This file contains a CVS FAQ. Until 1995 it was maintained by David
Grubbs. It was out of date and not being maintained, but it had a
certain following and in 1997 Pascal Molli decided to start
maintaining it with the FAQ-O-Matic package which allows any
contributor with a web browser to help maintain it. The following
text is (mostly automatically) extracted from the FAQ-O-Matic. The
odds are good that the file that you are currently reading is out of
date with respect to the online FAQ-O-Matic, which is part of Pascal
Molli's CVS web site at http://www.loria.fr/~molli/cvs-index.html
(currently under "Documentation"). The online version is also
somewhat better in terms of things like tables of contents (at least
until someone can write some code to extract data from a FAQ-O-Matic
and insert things like tables of contents).
The answers which are dated "6/13/1997" below are really from the 1995
FAQ, for the most part. Many of them are out of date. If you have
some time, you are encouraged to double-check them against other
sources like the Cederqvist manual and update the FAQ. If you don't
have such time, take them with a grain of salt or a few.
Category: /, all questions
Category: /
" [INLINE] "
1. About FAQ-O-Matic
This is FAQ-O-Matic, a quick-and-dirty Perl hack (aren't they all?) by
Jon Howell.
It seems like most FAQ maintainers make a valiant initial effort, then get
a life and don't have time to keep their FAQs up to date. Also, I often
find out a solution to a problem, and feel like I could write a single
FAQ answer on it in a few minutes, but where to post it?
Thus the FAQ-O-Matic was born. FAQ-O-Matic is a couple sleazy Perl scripts
that allow people to submit FAQ answers to this database, so it can stay
current, with just a tiny bit of work on any one person's part.
Yes, a bad guy could come along and wipe out all the FAQ entries. Please don't.
But to give the good guys some measure of comfort, each submission is stored
in an RCS file, so if someone does tamper, we can recover the database.
Guidelines for submissions:
1. Please _try to be fairly unbiased in matters of opinion._ Mailing lists are
the place to start flame wars (just kidding :v), but definitely not here.
2. Please _use HTML only conservatively_ in your entries. Links are appropriate
,
but put the URL in the plaintext also so it's useable on printed versions of
the FAQ. Inline images pointing off this site are inappropriate, as is much
fancy formatting. This is meant to be bandwidth-light and dumb-browser-friendly
.
3. If you feel there's a place for a _new category, or a reorganization of
existing questions_, don't hesitate to mail me (molli@loria.fr).
Category changes need to be done from my end.
4. Please _leave an email address_ at the bottom of your submission so that oth
ers
can drop you a note.
5. _If you only have a question_, not an answer, you should probably post
it to a mailing list, not here. If there are frequently asked questions to whic
h
the answer is not forthcoming on mailing lists (or perhaps there's no
useful answer yet other than "no one knows"), then it's appropriate to
post here, in hopes that someone will see it and know the answer.
6. Please refrain from crude or inconsiderate language. Please don't use
this as a forum for advertising. However, mention of worthy commercial
products is certainly appropriate (even if you sell said product). Just
don't overdo it. :v)
Last modified: _6/13/1997_
2. Adding a new category ?
just send me a mail at
molli@loria.fr
Last modified: _6/13/1997_
Category: /Advanced_Topics_/
" Advanced Topics "
Category: /Advanced_Topics_/Branching_and_Mergin/
" + Branching and Merging"
1. What is a branch?
Unfortunately, the word "branch" is an overloaded technical
term. It is used in too many different ways in three
categories. It might help to understand some of the issues by
going through the categories:
How Humans use the word "branch":
Most development starts with everyone working on the same
software, making changes and heading toward a single goal. This
is called something like "Main Line Development". Note that
though many people do main line development on CVS's "Main
Branch", that is a choice, not a requirement.
After a release or when one or more developers want to go off
and work on some project for a while, the Software Engineers
assigned to deal with large software issues generate a "Branch
in Development" to support the release or project. (Keep in
mind that a programmer is no more a Software Engineer than a
carpenter is a Civil Engineer.)
Essentially, the word "branch" implies a way to allow
simultaneous development on the same files by multiple people.
The above terms are human-oriented. They refer to actions that
people would like to take. They do *not* imply any particular
implementation or set of procedures. Branches in development
can be supported in many different ways.
How CVS uses the word "branch":
CVS uses the word "branch" in a number of ways. The two most
important are:
- The vendor branch holds releases from (normally) an outside
software vendor. It is implemented using a specific RCS branch
(i.e. 1.1.1).
- The "Main Branch", which normally holds your "Main Line
Development", but is defined as the collection of revisions you
get when you "checkout" something fresh, or when you use the
'-A' option to "update".
Important Note: The CVS "Main Branch" is *not* the same as the
RCS concept with the same name. If you are using Vendor
Branches, files you have never changed are on three branches at
the same time:
- The RCS 1.1.1 branch.
- The CVS Vendor branch.
- The CVS "Main Branch".
The concepts overlap, but they are not equivalent.
In referring to CVS, "branch" can be used in four other ways:
- A CVS working directory satisfies the definition of "branch"
for a single developer -- you are on a private "virtual branch"
that does not appear in any of the RCS files or the CVS control
files.
- The CVS "default branch" is the Repository source for the
collection of files in your working directory. It is *not* the
same as the RCS "default branch". Normally the CVS default
branch is the same as the CVS Main branch. If you use the "-r
<branch_tag>" option to the "checkout" command, you will record
a "sticky" tag that changes your default branch to the one you
checked out.
- A "magic" branch can be a branch that hasn't happened yet. It
is implemented by a special tag you can check out that is not
attached to a real RCS branch. When you commit a file to a
magic branch, the branch becomes real (i.e. a physical RCS
branch).
- And, of course, CVS uses "branch" to indicate a
human-oriented "branch in development".
How RCS uses the word "branch":
- The RCS "Main Branch" (Synonym: "The Trunk") contains a
series of two-part revision numbers separated by a single '.'
(e.g. 1.2). It is treated specially and is the initial default
branch. (The default default?)
- The RCS "Default" branch starts out attached to the RCS "Main
Branch". For RCS purposes, it can be changed to point to any
branch. Within CVS, you *must*not* alter the RCS default
branch. It is used to support the CVS idea of a "Main Branch"
and it must either point to the RCS Main Branch, or the Vendor
Branch (1.1.1) if you haven't made any changes to the file
since you executed "import".
Last modified: _6/13/1997_
2. Why (or when) would I want to create a branch?
Remember that you can think of your working directory as a "branch for
one". You can consider yourself to be on a branch all the time because
you can work without interfering with others until your project (big
or small) is done.
The four major situations when you should create a branch:
When you expect to take a long time or make a large set of changes
that the merging process will be difficult. Both "long" and "large"
are defined in your own environment.
When you want to be able to "commit" and "tag" your work repeatedly
without affecting others.
If you ever think you need Source Control for your own work, but don't
want your changes to affect others, create a private branch. (Put your
username in the branch tag, to make it obvious that it is private.)
When you need to share code among a group of developers, but not the
whole development organization working on the files.
Rather than trying to share a working directory, you can move onto a
branch and share your work with others by "committing" your work onto
the branch. Developers not working on the branch won't see your work
unless they switch to your branch or explicitly merge your branch into
theirs.
When you need to make minor changes to a released system.
Normally a "release" is labeled by a branch tag, allowing later work
on the released files. If the release is labeled by a non-branch tag,
it is easy to add a branch tag to a previously tagged module with the
"rtag" command. If the release is not tagged, you made a mistake.
Recovery requires identifying all revisions involved in the release
and adding a tag to them.
Last modified: _6/13/1997_
3. How do I create and checkout a branch?
Suggested technique:
Attach a non-branch tag to all the revisions you want to branch
from. (i.e. the branch point revisions)
When you decide you really need a branch, attach a branch tag to the
same revisions marked by the non-branch tag.
"Checkout" or "update" your working directory onto the branch.
Suggested procedure when using modules:
cvs rtag <branch_point_tag> module
cvs rtag -b -r <branch_point_tag> <branch_tag> <module>
cvs checkout -r <branch_tag> module
Suggested procedure when using your working directory, which
contains the revisions of your working files you want to branch from:
cvs tag <branch_point_tag>
cvs rtag -b -r <branch_point_tag> <branch_tag> <module>
cvs update -r <branch_tag>
In each procedure above, Step #1 applies a non-branch tag to all the
branch point revisions in the module/directory. Though this is not
strictly necessary, if you don't add a non-branch tag to the revisions
you branch from, you won't be able to refer to the branch point in the
future.
Between steps 1 & 2 you may commit changes. The result would be same
because "rtag -r <oldtag> <newtag>" applies <newtag> to the same
revision that <oldtag> is attached to. You can use this technique to
avoid attaching *any* branch tags until you need them.
Step B.2 has two corollaries:
If you plan to create the branch tag before committing anything in
your working directory, you can use "cvs tag -b <branch_tag>" instead
of the "rtag" command.
The <module> can be a relative path to a directory from which your
working directory was checked out.
If you have trouble figuring out what <module> to use (or pathname to
use in its place), you can aim it at whatever parent directories you
believe will cover all your work.
If you are sure the <branch_tag> is not being used anywhere else, you
can even aim it at the whole Repository ($CVSROOT), if you have to. It
might take some extra time, but assuming that your <tag> is a unique
string and you don't use the '-f' option to "rtag -r", "rtag" will
only add a <tag> to files in which it actually *finds* the earlier
<tag>.
In each procedure above, Step #3 may occur any time after step 2.
Unless you explicitly remove them with "tag -d", a <tag> is permanent.
The <branch_tag> is an unusual creature. It labels a branch in a way
that allows you to "checkout" the branch, to "commit" files to the end
of the branch and to refer to the end of the branch. It does not label
the base of the branch (the branch point).
There are two obvious ways to choose the <branch_point_tag> and
<branch_tag> names. But keep in mind that the <branch_tag> is typed by
any developer who wants to work on the branch -- you should make it
mean something to them.
Style #1 presumes that the simple version string refers to a set of
designed, documented or promised features, not to a specific set of
files. In this case, you tag the branch with the generic Version
string and assume that whenever you refer to "Version", you want the
"latest" set of files associated with that Version, including all
patches. (You can substitute whatever you like for "bp_", as long as
your <branch_point_tag> is some modification of the <branch_tag>.)
<branch_point_tag> Matching <branch_tag>
bp_V1_3 V1_3
bp_Release2-3-5 Release2-3-5
bp_Production4_5 Release4_5
Style #2 presumes that the simple version string refers to the
specific set of files used to construct the first release of
"version". In this case, you tag the branch-point revisions with the
generic Version string and assume that whenever you refer to this
Version, you want the original set of released revisions. To get the
latest patched revisions of the release, you refer to the branch tag
"latest_<branch_point_tag>". (You can substitute what ever you like
for "latest_", as long as your <branch_tag> is some modification of
the <branch_point_tag>.)
<branch_point_tag> Matching <branch_tag>
V1_3 latest_V1_3
Release2-3-5 latest_Release2-3-5
Release4_5 latest_Production4_5
In both styles you can find out what you had to change since the
original release of this Version by typing:
cvs diff -r <branch_point_tag> -r <branch_tag>
For Style 1, this is:
cvs diff -r bp_<branch_tag> -r <branch_tag>
For Style 2, this is:
cvs diff -r <branch_point_tag> -r latest_<branch_point_tag>
Notes on "being on a branch":
- "update -r <tag>" tells CVS to attach a "sticky tag" to working
directory (in ./CVS/Tag) and the checked-out files (on each line of
./CVS/Entries).
- A "sticky" <tag> (including a <branch_tag>) causes most CVS commands
to act as if "-r <tag>" were on the command line.
- A "sticky" <branch_tag> indicates that the working directory (and
working files) are "on the branch".
Last modified: _6/13/1997_
4. Once created, how do I manage a branch?
The most important thing you should know about managing a branch is
that the creation of a branch is not a lightweight act. When you
create a branch, you must also create a set of procedures to keep
track of it.
Specifically, you must:
- Remember that the branch exists. (This is non-trivial if you create
a lot of them.)
- Plan when to merge it back into the main line of development.
- Schedule the order that multiple branch merges are to be done.
- If you ever intend to merge branches into each other, instead of
limiting merges of branch work back into the "main line", you must
keep careful track of which parts of which branches have merged into
which other branches.
The simplest way to deal with branches is to limit their number,
"collapse" them back into the main line as quickly as is reasonable
and forget them. If a group wants to continue working, tell them to
create another branch off the fully merged main line.
Remember that CVS is just a tool. Over time, it will probably handle
branching better, requiring less careful attendance. But no matter how
good it becomes, the whole idea of "branching" is a complicated
management problem. Don't take it lightly.
Last modified: _6/13/1997_
5. Are there any extra issues in managing multiple branches?
If you plan to split from the "main line" and merge back after a time,
the only problem will be scheduling the order of branch merges. As
each branch is merged, the main line must be rebuilt and tested.
Merging multiple branches (i.e. "lines of development") before
building and testing creates more problems than you are ready for.
If you plan to collapse some branches into others, then move the
combined branches back into the main line, you have to be careful with
the revisions and tags you hand to your "update -j" command, but it
shouldn't be much trouble.
If you plan to allow every branch to incrementally take the work done
on other branches, you are creating an almost insurmountable
bookkeeping problem. Every developer will say "Hey, I can handle
taking just this little bit," but for the system as a whole it is
disaster. Try it once and see. If you are forced into this situation,
you will need to keep track of the beginning and end points of every
merge ever done. Good Luck.
Last modified: _6/13/1997_
6. How do I merge a whole branch back into the trunk?
If you don't have a working directory, you can checkout and merge in
one command:
cvs checkout -j <branch_tag> <module>
cd <module>
If you already have a working directory:
cd <working_directory>
cvs update <== Optional, to bring it up to date.
cvs update -j <branch_tag>
CVS will print lines beginning with
'U' for files that you hadn't changed, but the branch did.
'M' for files that you changed and the branch didn't
*and* for files that you both changed that were merged
without overlaps. (This overload is unfortunate.)
'C' for files that you both changed in a way that conflicts
with each other.
You need to go edit all the 'C' files and clean up the conflicts. Then
you must commit them.
Last modified: _6/13/1997_
7. How do I merge changes from the trunk into my branch or between
branches?
The idea is similar to the above, but since CVS doesn't treat the main
branch like other branches, you'll have to be more careful. There are
5 different ways to look at the problem.
The way to merge *all* changes made on the trunk into a working
branch is to move to the branch you want via "checkout -r" or "update
-r":
cvs update -r <branch_tag> {optional files}
Then merge the changes from the trunk into your working branch using
the pseudo-tag named "HEAD":
cvs up -j HEAD {optional files}
You will get everything from the branch point of the branch named
<branch_tag> up to the HEAD of the main branch. This is still kind of
strange. If the file is on a branch, HEAD should be the latest thing
on the branch, not the HEAD of MAIN. But that's not the way CVS
(currently) works.
If you run "cvs up -j HEAD" again after adding more revisions to the
trunk, you may get overlaps for the text you have already merged. It
depends on your version of your RCS "merge" command (actually the "co
-j" option, which depends on the version of "diff3" you configured RCS
to use).
You can merge the difference between any two <tags> using two "-j"
options on "update" or "checkout".
Identify the two tags on the branch you want to merge from.
cvs update -j <tag1> -j <tag2> {optional files}
This step assumes you were careful about tagging milestones. You can
use this technique for any two <tags> on the same branch, even the
trunk. It is also possible to use tags on different branches, but
you'll have to ponder the meaning of the difference between those two
tags.
In place of one of the <tags>, you can use a <branch_tag> to refer to
the latest revision on that branch. See 4C.11 and 4C.3 for info on
branch points.
Merges can also be performed by handing RCS revisions to the '-j'
options, but since revision numbers aren't the same in all files,
merging by number is normally limited to one file. Sets of files with
the exact same trees of branches and revision numbers would work too,
but that's a rare situation.
To "take" revisions from other branches instead of merging them, see
4C.19 for an idea.
A way to gain the effect of merging the main to the branch is to
merge the branch into the main using the normal
cvs update -A {optional files}
cvs update -j <branch_tag> {optional files}
cvs commit
cvs tag -F -b <same_branch_tag> {optional files}
See part B of 4D.5
Other oddities.
This also works, but is probably not officially supported:
cvs update -j N {optional files}
where N is a number. This will merge all the changes from the branch
point up to the highest revision on the main branch starting with N.
For example, if your highest trunk revision is 1.52, you can use this
to grab revisions from the trunk:
cvs update -j 1 {optional files}
Another example: Say you have a branch point at rev 1.2 for a branch
named "BR1" and trunk revisions 1.3, 1.4, 2.1, 2.2, 2.3, 3.1, 3.2.
Then:
cvs update -j 1 {optional files}
will merge the changes from 1.2 to 1.4
cvs update -j 2 {optional files}
will merge the changes from 1.2 to 2.3
cvs update -j 3 {optional files}
will merge the changes from 1.2 to 3.2, which in this example, is
equivalent to the use of "-j HEAD" in part A above.
The intuitive (at least to me):
cvs up -j MAIN (or TRUNK) {optional files}
doesn't work. If the trunk (i.e. "main branch") had an implicit branch
named "MAIN", you could use:
cvs up -j MAIN:10/26 -j MAIN:now {optional files}
and refer to date-stamped revisions on the trunk using the
<branch_tag>:<date> support that works on other branches.
You might also think you could place an explicit tag on branch 1 (or
higher) (e.g. MAINHACK:1) and use it in place of the implicit "MAIN",
but I haven't found the right combination.
[[If you find working techniques, I'll add them here.]]
Last modified: _6/13/1997_
8. How do I merge onto the Main Branch a file that exists only on a branch
other than the Main Branch? (i.e. it is in the Attic)
For how such a file can exist, see 3A.2 and 3A.3.
For how to avoid creating such a file, see 3A.5.
Though you might think that the "update -j" command could perform the
"merge" of a file from the side branch to the Main Branch, it isn't
(yet) smart enough. Unfortunately, there is no single CVS command to
do this -- it takes three steps:
To move something onto the Main Branch from the Attic, you have to
physically move the file from the Attic to the main Repository
directory associated with your working directory.
It is exactly like resurrecting a removed file. See 3L.4
I use something like this: (csh-like syntax)
set repos = `cat ./CVS/Repository` mv $repos/Attic/filename,v
$repos/filename,v
(If you use relative paths in your Repository files, that first line
becomes: set repos = $CVSROOT/`cat ./CVS/Repository`)
Now that the file is physically in the right place within the
Repository, "update -A" will make it appear in your working directory
on the Main Branch. Do that now.
You now have a choice. The act of physically moving the file has
fused together the <branch_tag> branch and the Main Branch for this
file. You can continue that way, making changes along the RCS Main
Branch which CVS will (for this type of file only) treat as both the
Main Branch and the <branch_tag> branch.
The other choice, which I would suggest, is to re-tag the file with
<branch_tag>, restoring a normal-looking magic branch tag to the file:
cvs tag -F -b <branch_tag> <file>
After you have done the above, you can run "update -A" or "update -r
<branch_tag>" to resume whatever you were doing before you started
this procedure.
Caveat: The final result is a file whose revision tree doesn't look
like it was ever on any branch but the Main Branch until the above
"tag -F -b" command was executed. CVS and RCS have no way of saving
the history of the actions you have just performed.
Last modified: _6/13/1997_
9. How do I know what branch I'm (working) on?
Type:
cvs status
and look at the "Sticky Tag" field for each file. If:
The *same* tag is on *every* file in your working tree, *and*
That tag matches the contents of the ./CVS/Tag file, *and*
That tag is a branch tag,
then you know what branch you are working on. You can get sticky Tag
information directly from the ./CVS/Entries file instead of "cvs
status".
If all the sticky Tags don't agree, then your directory is temporarily
inconsistent. This is a feature allowing you to make changes (or
perform merges) to individual files on multiple branches without
checking out the whole directory.
The sticky Tag on each file in the ./CVS/Entries file (as displayed by
the "status" command) indicates what branch the working file is on.
New files are added to the Tag stored in ./CVS/Tag.
To force your entire working directory onto the same branch, type:
cvs update -r <branch_tag>
Last modified: _6/13/1997_
10. Do I really have to know the name of the branch I'm working on?
If a developer can't be relied on to know what branch of development
to work on, then either the developer's manager isn't planning
branches properly or the developer has serious problems.
I have found that one of the hardest concepts to get across to
developers (and some managers) is that "a branch in development" (as
opposed to the use of RCS branches to support some other scheme) is a
heavyweight act. Every time you create a real branch in development,
you must spawn a set of managerial procedures and a schedule by which
you plan to merge each branch into each other branch. Unless you plan
to keep it simple and collapse (by merging and forgetting) branches
quickly, they are not to be created lightly.
In other words, if you don't regularly attend group meetings in which
the branch to be worked on is a major topic of discussion, then the
group is not managing branches properly.
We created a couple major branches a few months ago and even the
customer service people refer to the "XYZ branch" as a shorthand for
"continuing development on the XYZ project".
Last modified: _6/13/1997_
11. How do I refer to the revision where I branched so I can see what
changed since the Branch Point on another branch?
Given the current <branch_tag> format, there is no direct way to refer
to the branch point, which is more useful in many ways than referring
to the branch, which always refers to the latest revision on the
branch.
When CVS adds a branch tag, it attaches an RCS symbol to a
non-existent revision number containing the revision number of the
branch point as a prefix. (See Section 3O, on the "tag" command.) RCS
can't use the CVS magic branch tag and many of the CVS commands can't
refer to it.
To be certain of your ability to refer to a branch point, you must
create a "branch point" tag at the same time as the Branch tag. See
4C.3.
Last modified: _6/13/1997_
12. Why didn't the command "cvs admin -bBRANCH1 *" create a branch?
Because your command creates an RCS branch, not a CVS branch. See the
above discussion on branches. RCS branches are used to support CVS
branches, but they are not the same. You can't act as if you have
direct control over the RCS files.
The "admin" command was placed there as a convenience to allow you to
execute raw "rcs" commands on the Repository, taking advantage of
CVS's ability to find the files in the Repository.
But you have to remember that you are using RCS commands on a CVS
Repository, which is not generally safe unless you know exactly what
CVS depends on.
For one thing, CVS insists on control of the default branch. It is set
either to the Main branch or the Vendor branch depending on whether
you have changed the Vendor's code. If you change the default branch,
you are monkeying with the internals and you will get unexpected
results.
To set your "default CVS branch" to BRANCH1, you must use "checkout"
or "update" with the "-r BRANCH1" option. Then you have changed CVS's
idea of your "default branch", which has little to do with RCS's
default branch.
Last modified: _6/13/1997_
13. Is it possible to set the "default CVS branch" for everyone?
No. It doesn't work that way.
When using CVS, all administrative information (such as what branch
you checked out) is stored in CVS sub-directories, local to the user.
There is no global state, other than the description and logging files
in the $CVSROOT/CVSROOT directory.
You tell "checkout" or "update" what branch you want to check out via
the "-r <tag>" option. The default is CVS's "Main Branch".
I don't see a problem in *designing* a new way to indicate what branch
you get by default, instead of the main one, but that's not how it
currently works.
Last modified: _6/13/1997_
14. How do I perform a large merge?
Large merges require a bit more planning to be able to track what has
happened in the inevitable cases where something goes wrong. No tool
can force a "merge" to make perfect sense.
Though you can handle the details in many different ways, the two ends
of the spectrum of merge techniques are: gonzo and paranoid.
The gonzo method assumes that you know everything about your sources
so that recovery from failures is "just a matter of typing." You
created the branch this way:
cvs checkout <module>
cd <module>
cvs tag -b <branch_tag>
cvs update -r <branch_tag>
>>> Edit away.
cvs commit <<== Onto branch
Now you want to merge your branch back into the Main branch, you are
certain you can make it work, or at least detect all the failures, so
you dive in and hack away: (For simplicity, we will assume you are
collapsing (i.e. merging and forgetting) a side-branch into the Main
branch from your single working directory.)
cvs update -A
cvs update -j <branch_tag>
>>> Edit the 'C' files and remove the overlaps.
>>> Edit some more to make it all compile and work.
cvs commit
Looks simple. For more details on the output from the "update -j"
command, see 3P.2 and 4C.6.
Note: You could also checkout a whole new working directory and
perform the merge at the same time by replacing the two
update commands with these two commands:
cvs checkout -j <branch_tag> <module>
cd <module>
The paranoid way is more difficult, but it can catch all sorts of
problems. You created the branch this way:
cvs checkout <module>
cd <module>
cvs tag <branch_point_tag>
cvs tag -b <branch_tag>
cvs update -r <branch_tag>
>>> Edit away.
cvs commit <<== Onto branch
The extra tag command places a non-branch tag on the Branch Point, an
act that makes it easier to do "diffs" later. Now we decide to perform
the merge:
cvs tag <latest_on_branch_tag>
cvs update -A
*1* cvs diff -r <branch_point_tag> -r <latest_on_branch_tag>
>>> *1* shows all the changes on the branch.
*2* cvs diff -r <branch_point_tag> -r HEAD
>>> *2* shows the changes on the trunk since branching.
cvs tag <premerge_tag>
cvs update -j <branch_tag>
>>> Edit the 'C' files and remove the overlaps.
*3* cvs diff
>>> Verify that *3* matches *1*, except for line numbers.
cvs commit
cvs tag <just_merge_changes_tag>
>>> Edit some more to make it all compile and work.
cvs commit
cvs tag <after_merge_cleanup_tag>
The reason *3* and *1* match so closely is that they are the
differences between two pairs of starting points and ending points
after the same data was inserted. If they are significantly different,
you will want to figure out why.
NOTE: You will have to tell everyone to stay the hell out of the
Repository while you do this. If they commit something while you are
in the middle of a merge, your job will be much more difficult. If
they "update" at the wrong time, their work will be randomized until
you finish. It's better to call a halt.
See 3H.13 for some more information about dealing with merges after
import. The last part of the procedure is applicable to any large
merge.
Last modified: _6/13/1997_
15. Is a Vendor merge any different from a branch merge?
No. In most ways, a Vendor branch is exactly the same as any other
branch. In a Vendor merge, the data is append to the branch by the
"import" command, rather than by hand-editing, but the merge process
is the same.
See the "import" command in section 3H.
Last modified: _6/13/1997_
16. How do I go back to a previous version of the code on a branch?
You can avoid digging into RCS revision numbers (executing "update
-r (rev)" on each file) by trying one of these:
Use non-branch tags as you normally would. Non-branch tags
attach to specific revisions, so a "tag (tag)" command would
mark the revisions you have in your working directory, which
are on your branch. If you need to retrieve them, use "update
-r (non-branch-tag)"
Doing this overrides the sticky (branch-tag) attached to your
working directory with a non-branch tag, which means you won't
be able to commit until you again move forward to the end of
the branch with "update -r (branch-tag)".
Use the "update -r (branch-tag):(date)" trick.
This is almost like using the '-D' option, but it looks for
revisions extant on (date) only along the given branch.
As in #1, you can't commit to this kind of working area,
because it has a sticky date referring to revisions in the
middle of a branch.
[comment from the audience: You are dreaming..
this does not work.. try it, you get
No such tag: "MYTAG:May 1"
or similar. I wish it did because I need it. julian@whistle.com]
You can branch a branch.
If you add a branch tag to file in a working directory that was
checked out on a branch, you will branch the branch. This
works just fine, though you'll have to play some games to merge
everything back together again. You'll also create 6-part
revision numbers. (They'll be 8-part revision numbers if you
branch a branch that started out with some unmodified files on
the Vendor branch. Think about it. How does revision
1.2.4.2.4.2.2.1 grab you?)
(fixed formatting, kingdon@cyclic.com)
Last modified: _9/8/1997_
17. Once I've found the files I want, how do I start changing them? I keep
getting warnings about sticky tags.
What you probably did was type "cvs update -r <tag>" where <tag> is a
non-branch tag. "update" created a sticky tag for a specific revision,
not a branch. To start working right there, you have to create a
branch to work on.
You have two choices.
You can do it in place and keep working:
cvs tag -b <branch_tag> <<== To tag the current files.
cvs update -r <branch_tab> <<== To move onto the branch.
You can do it "externally" and create a new working directory:
cvs rtag -b -r <tag> <branch_tag> <module>
cvs checkout -r <branch_tag> <module>
<module> can be a relative path within the Repository.
<tag> in the above is the non-branch tag you placed earlier
that caused the error in your question. Be warned that
if <tag> is not set on all the files (or all the right
revisions) you won't get exactly what you wanted.
Last modified: _6/13/1997_
18. Why do I get the latest files on the branch when I tried to "update -r
<tag>"?
If "update -r <tag>" always retrieves the latest files on a branch,
then <tag> is really a <branch_tag>. A branch tag is supposed to be
used to grab a branch to work on. Since you can't modify a file in the
middle of a branch, checking out a <branch_tag> will give you the
latest revision on the branch.
If you want to "checkout" a specific collection of revisions, you must
use a "non-branch" tag. See the first part of 4C.16.
Last modified: _6/13/1997_
19. How can I avoid a merge? I just want to move the latest revision on my
working branch directly onto the trunk.
There is no direct way to do this using CVS, though the technique is
not difficult using shell commands. Here's one way:
Move your working directory to the Main Branch.
cvs update -A
Use "update -p" to grab the latest revision on the branch and write
it over your working files. Make sure you don't have an modified files
-- you will lose them. The following is in "csh" syntax. Change the
wildcard to grab the files you want
foreach i (Makefile *.cc *.hh)
cvs update -p -r <branch_tag> $i > $i
end
Commit all the working files onto the Main Branch.
cvs commit -m 'Moved branch <branch_tag> onto MAIN'
You should experiment with the above before blasting everything.
Last modified: _6/13/1997_
20. How to I avoid merge collisions in the RCS $\Log$ data?
In short, you can't. The RCS $\Log$ keyword is handled differently
from all other RCS keywords.
On the info-cvs mailing list, there is a periodic discussion that goes
something like this:
Question: How do I deal with $\Log$? Answer1: You can't do much with
it. Here's how it works. . . Answer2: I've found a limited way to use
it. . . Answer3: Get rid of it. $\Log$ is an abomination.
I tend to lean toward answer #3. There are only two sets of people who
would ever have access to logs stored within sources files, developers
and source customers.
For developers:
Log entries within sources files are notoriously incomplete, rushed,
poorly phrased and in many cases incorrect, making them useless for
debugging or file maintenance. I remember a maxim from "Software
Tools" (I believe): "Read the code, not the comments." No managerial
order or plan for programmer discipline will affect this in the real
world.
Log entries are usually in an unreadable mixture of styles. Many log
entries are just plain meaningless. Some are foolish. Some are even
insulting. Examples:
"Corrected spelling of misspelling." "Bug fix." "Reversed stupid
change in previous revisions." "If Joe could do his job, this would
already have worked."
Log entries are not managed well by the tools. Any merge can cause
conflicts in the $\Log$ data. Branch merges produce incomplete logs.
They can be edited into chaos and they are not regenerated. They waste
space duplicating information available to the developer with a single
command.
Even if correct when originally entered, as changes are made to the
file, log entries become false over time. Humans are not good at
reading down through a list and remembering only the last change
affecting something. Over time *most* of the log is wrong.
Even if still correct, the log data is almost useless to developers
without the code diffs. If you can get code diffs, you can display the
log.
For source customers the problem is even worse. The last thing you
want to show customers is a hodge-podge of tiny comments about large
changes followed by a series of emergency fixes before delivery. If
you distribute sources, then you should provide documentation, or
changelogs reviewed by people who won't let comments like "Fixed for
stupid customer." out the door.
Conclusion: Though some people would prefer to see in this FAQ
techniques for making the $\Log$ entries the best they can be, I
believe them to be a lost cause. My suggestion is to hunt down, root
out and destroy all occurrences of $\Log$ and the unusable data
attached to it wherever you may find it.
Last modified: _6/13/1997_
21. Why should I trust automatic merges?
Some developers have the feeling that three-way merging doesn't work.
They fear and distrust the way the "update" command automatically
merges committed changes from the Repository into the working file.
Experience has shown that most merges are utterly painless and most of
the rest are easily resolved. The few conflicts that cause headaches
are nearly all due to poor communication between developers, a problem
no source control system can obviate.
Some developers were troubled in the past by flaky Unix software. I
can't say that everything is perfect, but the tools CVS depends on
(RCS and diff, mainly) are fairly solid nowadays. They work.
Since it does seem to work for most of us, the algorithm is unlikely
to change soon. Why not test it on a couple trouble spots and if it
works for you, use it for a while? Then you can make an informed
decision.
Last modified: _6/13/1997_
22. How does CVS decide if it can safely perform a merge?
CVS can merge any text file, possibly discovering a conflict and
leaving overlaps for you to edit. Editing the conflict markers out of
the file is a moment's work, but resolving the conflict could take an
arbitrary amount of time. CVS works to determine if it *should* merge,
not if it *can*.
See 2B.6 for how the merge proceeds.
Last modified: _6/13/1997_
23. After resolving merge conflicts in a file, what if I want to keep my
previous version, and not take any of the branch changes?
If you want to retain your previous version, a version on the MAIN
branch greater than 1.1 (one you committed there), just throw the
merged file away and "cvs update" the file.
You don't need to commit something to remember it. The tags you place
before and after the merge should give all the handles you need to
find various versions. You don't have to create a new version of the
file.
If you want to retain the previous Vendor revision, you can grab a
copy of it using "cvs update -p" and commit it or use the technique
described in 3B.3 to revert back to the Vendor branch.
Last modified: _6/13/1997_
Category: /Advanced_Topics_/Engineering/
" + Engineering"
1. Where can I find out about Software Engineering?
A couple different people suggested this book:
Software Configuration Management: Coordination for Team Productivity;
Wayne A. Babich; Addison Wesley; 1986; ISBN 0-201-10161-0
A number of others suggested Appendix B of the book "Decline and Fall
of the American Programmer" by Ed Yourdon, called "The Programmer's
Bookshelf". It list 87 books you are expected to have read. Since they
publish many of the books, Prentice-Hall distributes this list as
"Prentice Hall Professional Technical reference PTR-125-AA3.
One interesting item from the Yourdon book: The total number of
professional computer books sold is less than the number of
programmers currently in the United States. It wasn't clear from the
book whether this meant "per year" or not, but it is still
frightening.
Last modified: _6/13/1997_
2. How do I flexibly arrange the modules file to describe my sources?
An equivalent question might be, "How do I structure my sources?" This
can be a difficult question especially in the areas that are more
political than technical.
Generally you want to think about which pieces of your system need to
be checked out together, built as one system or tagged as a consistent
whole. You should certainly create module names that correspond to
complete, buildable collections that you would tag and release as one
"product". It is also convenient to create module names for small
sections of the Repository containing files that will all be worked on
at the same time by the same person or group.
Once you have defined the structure of your work, you can usually see
how to lay it out in a Repository. After that the modules file is
easy. You set up module names and aliases to match what you need to
check out by name. If you like relative directories, it is possible,
but not recommended, to work completely without a modules file. See
1D.11 and 2C.7 for some info about the modules file.
Here are a few types of modules. You should experiment to see what
kind of structure each of these produces. They all have different
uses.
Connected projects in one group with two separate helper
directories. The helper directories can contain build tools, header
files, libraries, or whatever you like.
These are all aliases that checkout relative pathnames. The equivalent
results could be produced by placing the selected relative pathnames
on the "cvs checkout" command line.
pr1 -a P1 HELPERS
pr2 -a P2 HELPERS
pr3 -a P3 HELPERS
pr12 -a P1 P2 HELPERS
pr13 -a P1 P3 HELPERS
pr23 -a P2 P3 HELPERS
P1 -a group1/proj1
P2 -a group1/proj2
P3 -a group1/proj3
HELPERS -a group1/helper1 group1/helper2 MAKEFILE
MAKEFILE -a group1/Makefile
Actual Repository directory structure: (from $CVSROOT down)
group1/ Makefile The top level Makefile. helper1/ helper2/ Helper
files and dirs proj1/ Files and dirs proj2/ Files and dirs proj3/
Files and dirs
"checkout group1" produces a duplicate of the above. "checkout projX"
produces all but "projY" and "projZ". "checkout projXY" produces all
but "projZ".
Here is the exact same set of module names describing the same
Repository layout using module names (and aliases containing module
names) instead of merely aliases for relative pathnames.
There is one difference in the result. The name of the top level
directory in the checked out working tree will match the "module" name
(e.g. pr1) instead of always being "group1" as it was in the first
example above.
pr1 group1 proj1 &HELPERS
pr2 group1 proj2 &HELPERS
pr3 group1 proj3 &HELPERS
pr12 group1 proj1 proj2 &HELPERS
pr13 group1 proj1 proj3 &HELPERS
pr23 group1 proj2 proj3 &HELPERS
HELPERS -a helper1 helper2 group1-Makefile
helper1 group1/helper1
helper2 group1/helper2
group1-Makefile -d . group1 Makefile
The above line (with the -d in it) says that when the module named
"group1-Makefile" is checked out, the file named Makefile file will be
found in a directory named $CVSROOT/group1 and will be checked out
into a directory named '.', which obviously already exists.
The & references say to interpret those pathnames relative to the
directory where the whole module is stored. For the "pr1" module, that
directory is "group1", so the &HELPERS reference winds up placing
Makefile in '.' relative to "group1".
A short one containing the basic "module" actions:
m1 head/path file1 dir2 file3 dir4 file5
When checked out, a directory named "m1" appears in your current
directory. Elements named file1, dir2, file3, dir4, and file5 appear
in it. They were originally taken as relative paths from
$CVSROOT/head/path.
Here's another way to construct a working directory out of pieces of
the Repository:
projX projX Makefile &projX_inc &projX_src &projX_doc
# The first line selects a single file within projX, plus
# the contents of three other modules. Those three other
# modules rename their directories.
projX_inc -d include projX/inc projX_src -d source projX/src projX_doc
-d documentation projX/doc
A Unix tree. This is similar to what CVS was developed for and the
way I have used it for years.
# Top level
unix unix
u_bin unix/bin
u_etc unix/etc
u_man unix/man
usr-bin unix/usr.bin
# Subdirs of top level dirs. (tiny subset)
ls unix/bin/ls
fsck unix/etc/fsck
man8 unix/man/man8
# Programs without subdirs. (tiny subset)
cat unix/bin Makefile cat.c
uniq unix/usr.bin Makefile uniq.c
# /usr/local/src
localsrc localsrc
gnu localsrc/gnu
public localsrc/public
X11 localsrc/X11
# GNU and PD tools
cvs localsrc/gnu/cvs
emacs localsrc/gnu/emacs
rcs localsrc/gnu/rcs
btoa localsrc/public/btoa
tcsh localsrc/public/tcsh
# X11 related items.
tvtwm localsrc/X11/contrib/tvtwm
"unix" was checked out and built from the top down, using a set of
Makefiles that knew about the whole structure. "localsrc" was kept
checked out in /usr/local/src.
At any time I could run "checkout ls" or "checkout cat" and get a
simple directory with only that tool in it, plus a subset Makefile
that knew how to build that tool against the installed (or alternate,
via environment variables) headers and libraries.
I found it very handy to be able to run "ls" and see the three tools I
was porting that week.
Last modified: _6/13/1997_
3. Can I have multiple source repositories, one for each project?
Yes, you can have as many Repositories as you like. But each
Repository must be managed separately, creating additional work.
Question 4A.1 provides a short description of setting up a single
Repository. A few additional considerations:
It is a good idea to start by creating a single Repository and split
it up (or create additional Repositories) only if you believe it is
really necessary. I would only create a new Repository if the data is
completely disconnected from the rest of the main Repository.
If there is a lot of overlap among the developers working on the
collections of files you want to place in different Repositories, or
if there is any connection between those collections, I would go out
of my way to create a single Repository. It is much easier to manage.
Disk space should not be a factor since you can build up a
Repository using symbolic links and/or remote mounts.
Each Repository is completely distinct. You can't check out modules
from different Repositories at the same time. A better way of looking
at it is that if you *can* check out two modules or directories with a
single "checkout" command (without contortions or explicit absolute
pathnames), then they are in the same Repository.
To "checkout" modules from multiple Repositories, you must use the
"cvs -d" option on all CVS commands or alter your $CVSROOT variable
when you change focus to another Repository. If you work with multiple
Repositories, it is a good idea to configure CVS to use absolute
pathnames in the ./CVS/Repository file, since most commands (other
than "checkout") will use that file rather than $CVSROOT.
If you configure CVS to use relative pathnames in your
./CVS/Repository files, you must always be careful to set your
$CVSROOT properly or you will get unexpected results.
If you have two modules or directories by the same name at the same
relative path inside two different Repositories, you are asking for
disaster. You could unexpectedly update a directory with completely
unrelated files. This is not a fanciful example -- a Repository is
occasionally duplicated for release purposes in which case *all* the
paths in the two Repositories are the same.
Last modified: _6/13/1997_
4. Who should administer the Repository and manage the modules file?
This is a "management style" question. In large or traditional groups,
the CVS procedures are warped to conform to local conventions. In
small groups, in groups with strong personalities or on new projects
the choice of source control procedures can help create some of the
working environment. Here is a taxonomy of environments I have worked
in or helped set up:
Situation 1.
A small number of competent developers working on a medium size
project. We all got along and we all respected each other (at least
technically). Anyone edited anything.
Modules and Repository admin was mostly left to me. I never found a
problem in minor changes made by anyone else.
Situation 2.
A large number of experienced developers sprinkled with wackos. Many
of the developers didn't want to deal with any kind of source control.
They wanted a full-service source control system that caused them zero
thought.
I learned "big stick" diplomacy here. There was a small number of
"designated" (by me) people who were allowed to do *anything* other
than "update" and "commit". Even "checkouts" were controlled. This is
where I found "history" and "release" the most useful.
Situation 3.
A small number of developers who wanted me to "help", but who didn't
want to deal with anything other than their favorite algorithms.
I didn't have the time to baby-sit this group, so I designated one of
them to be my official contact and made him do it all. He felt sullied
by the requirement to pay attention to anything other than his pet
coding projects, but enjoyed the "status" of being the only one who
could touch the control files without my kicking the chair out from
under him.
Situation 4.
A huge number of developers of covering the whole spectrum of
competence and experience split into 20 groups, none of which
cooperated with the others, working on 57 different projects, most of
which didn't inter-operate.
Managing it in any coherent way was not my responsibility (and beyond
my tolerance for chaos). Too many people. So I privately designated a
person in each group to be the contact and kept watch on the
Repository activity. When something went wrong, I notified the contact
for the group and told him what was happening and *he* kept his troops
in line. They were tougher with their own group that I would have
been.
Eventually only a few people were willing to touch the control files,
since they were flamed from all directions if they screwed up.
Situation 5.
In a medium group of really *serious*, and seriously overworked,
people, someone else was designated the "master". I convinced the
master I knew what I was doing and went on my way.
No one else in the world was allowed to touch anything.
Situation 6.
In a large amorphous group of beginners, experts and clowns, over whom
no one had official control, I was forced to employ a group of
relative beginners (who became experts rather quickly) to police the
world. The ultimate in locking the barn after the horse was stolen, we
kept Chaos from destroying us only by use of superior firepower.
My choice, if allowed, is to let anyone touch anything. I keep backups
of important items and let people know individually whether I want
them to touch things or not. If someone on my "no touch" list touches
and succeeds, they are allowed more slack. If they screw up after
being warned, their screwup becomes public. After a few months, I
usually have no trouble keeping the world running smoothly, at least
from my (and CVS's) perspective.
Last modified: _6/13/1997_
5. Isn't disk space a big factor? CVS copies files out of the Repository,
duplicating everything.
Everyone knows that disk space is getting cheaper. How do we reconcile
this with the equally well-known problem that *all* disk is *always*
filled up?
In my opinion, the main reason disk space will never be an unlimited
resource is that it is the major variable in organizational time/space
tradeoffs. It isn't a problem of waste or an aspect of Murphy's law,
as some claim it is, but rather a direct consequence of good
management. Disk space is, and will always be, a limited resource.
First, the cost of *deploying* that disk is not dropping as fast as
the cost of the storage medium. The cost of machines to hold the disks
and the networks to connect them are dropping more slowly than disk
media. And the cost of the human time necessary to manage the
machines, networks, disks, and the developers using them, is not
dropping at all. The cost of human time continues to rise.
If management decides that expensive human time can be saved by using
all that new disk space to keep the last three releases online, then
that's what it will be used for. If each release takes up a Gigabyte
and you support 30 platforms, a simple time-saving suggestion has just
grabbed 100 Gigabytes of disk space. And we've ignored the potential
disk storage needed to support "better Customer Service", another
management refrain.
Even at 30 cents per Megabyte (next year's price), you've just used up
$30,000 of disk space. And that doesn't count the computers, tape
drives and humans necessary to maintain and deploy all of it. Spending
money to save time has its own overhead, too.
Binaries are getting bigger. Graphics and data collection devices can
eat up any amount of disk. There are more tools available, more
libraries, more raw data than you can ever store. My home computer has
a Gigabyte of disk on it. It could easily handle 30.
The "economy" of disk storage media will never remove the need to
manage disk space.
So, here's an un-reviewed suggestion originally from Graydon Dodson
<grdodson@lexmark.com>, which I've altered and edited heavily.
- Keep a directory where the whole tree is checked out. (It might be
built and tested once in a while to make sure it is worth linking to,
but that doesn't affect the source control aspect of this procedure).
Let's call it /master/build.
- Write a tool that creates a tree of directories (like the X11
"lndir" command) filled with links to the checked out files in the
/master/build tree.
This tool should also provide real copies of, not symlinks to, all the
files within the CVS administrative directories.
- You could also provide a way for the tool to take a list of whole
directories that you will never change, for which it would create a
single symlink to the directory and not a subtree of symlinks to
files. Or you could rm -r pieces of the resulting working directory
yourself and replace it with links.
- If you want to edit a file, you have to grab a real copy and keep it
until your revision shows up in the /master/build tree. I'd create a
script to do this: cvsgrab <file>
#!/bin/csh -f
set f = $1
if (! -l $f) then
echo "file $f is not a symlink"
exit 1
endif
rm $f
set rev = `grep "^/$f/" CVS/Entries | awk -F/ '{print $3}'`
cvs update -p -r $rev $f > $f
You can't do a plain "cvs update" since that would grab newer
revisions from the Repository, not the revision you wanted to start
with. After the file is no longer a symlink, you can work normally.
You'll have to run "update" before "commit" anyway if there are newer
revisions.
- Presumably there would also be a tool to traverse the link tree and
revert it to links if there are no modified files and/or if all the
real files match the revision of the /master/build tree.
- To avoid confusing CVS when the /master/build revisions are updated
but your CVS/Entries files is not, CVS would have to change to handle
symlinks. It currently causes problems with this scenario:
./<file> is a symlink.
./CVS/Entries says you are revision 1.2.
The corresponding CVS/Entries file in /master/build says the latest
revision is 1.3.
cvs update <file> shows a 'C' conflict flag.
Last modified: _6/13/1997_
Category: /Advanced_Topics_/Installing_CVS/
" + Installing CVS"
1. What do I have to do before I install CVS?
You must decide where to set up a Repository.
Though you can construct a Repository tree structure using links and
mount points, there must be a single copy of each real file across
your entire organization. You may not "rdist" files and expect to edit
both copies.
CVS does not support a truly distributed Repository. You can have
multiple Repositories, but each one must be mounted (not copied or
"rdist"ed) from a single place onto all machines where it will be
used.
Initially, a Repository takes about same amount of disk space as the
sources you want to put into it, plus a bit of overhead for the RCS
files.
See Section 4B. For multiple Repositories, see 4G.3
You need a directory in everyone's $PATH variable where you can
install all the executables. /usr/local/bin is a common place.
You need some helper tools besides CVS such as "RCS" and a good set
of "diff" and "diff3" programs. See 1B.4 for suggestions.
Read the README, INSTALL and ChangeLog files to see what you are
getting into.
Make sure you have versions of all the programs mentioned in the
"cvs/src/options.h" and "cvs/src/rcs.h" files.
Though you can probably muddle along without it, you should appoint
one or more "Repository Administrators" who will be responsible for
maintaining the Repository structure, administrative files and the
"modules" interface.
Someone at your site should probably be on the info-cvs mailing list.
See 1B.5.
Last modified: _6/13/1997_
2. How do I configure the CVS programs?
You should certainly start by reading the README file, the INSTALL
files and possibly the ChangeLogs in each directory, the Makefile.in
files and the "cvsinit.sh" program.
Edit the "options.h" file in the "src" directory.
You might need to specify a few site-specific pieces of information
including the names of a number of functions.
Hint1: You probably want to set the DIFF macro to use your version of
the GNU diff program with the '-a' option. Ours is set to "gdiff -a".
Hint2: You want to use RCS 5.6.0.1 or greater and set the "HAVE_RCS5"
macro.
Execute the ./configure command.
Type "make".
After running "make" you might try running the "sanity.sh" script:
./src/sanity.sh `pwd`/src/cvs
It writes into /tmp/cvs-sanity by default.
Finish reading the INSTALL file and test out the system.
Last modified: _6/13/1997_
3. What do I have to install?
Install the "cvs" executable and "mkmodules" from the CVS sources.
The man page is useful too. If you plan to report bugs, you should
also install "cvsbug".
Make sure you have versions of all the programs mentioned in the
options.h file, most of which are included in a standard Unix system.
Unless you plan to reimplement RCS [:-)], you must install RCS.
It is a very good idea to examine the RCS installation instructions
and make sure you are using the GNU versions of "diff" and "diff3" or
merges (an important part of CVS) will not work as well as you'd like.
Set your $CVSROOT environment variable and create the Repository
(which you planned out in 4A.1) with the "cvsinit" command at the top
of the CVS sources.
You'll need to edit the Repository control files created by
"cvsinit".
Install any helper programs mentioned in the modules file.
Last modified: _6/13/1997_
4. How do I work around the merge problems in GNU diff version 2.1 or
later?
See 1B.4 If you use recent versions of RCS and "diff", you won't run
into the above. If you do, see 5B.8
Last modified: _6/13/1997_
Category: /Advanced_Topics_/Internal_errors/
" + Internal errors"
1. Explain: "ci error: unexpected EOF in diff output"
RCS versions earlier than 5.5 print the above error when a file does
not end in a newline character. It can be caused by:
- Editing with Emacs and not using "require-final-newline".
- Committing a binary file.
- Filesystem failures (NFS!) that put nulls in your file.
The solution is to upgrade to RCS 5.5 or later. (Of course, this won't
fix filesystem failures. It will merely allow RCS (and therefore CVS)
to handle the file without error.)
Last modified: _6/13/1997_
2. Explain: "RCS file /Repository/module/file.c,v is in use"
This is an RCS error that occurs when its internal lock file has been
left around by an RCS command interrupted by some sort of system
crash, disk failure or SIGKILL signal.
Go into the Repository and look for files with names similar to
"file.c,v", usually starting with ',', '_' or '#'. Make sure they are
really crash remnants and do not belong to transactions in progress --
a recent last-modified timestamp is a good indicator of a live
transaction. Delete them if they are old.
Last modified: _6/13/1997_
3. Explain: "co error, line 2: Missing access list"
This is an error message from RCS Version 3 when it tries to read a
file created by a later version of RCS.
HP decided to "standardize" on an ancient version of RCS some time
ago. You can't use it for CVS. See 4H.6.
Since the error comes from having a later version of RCS than HP
supports, you probably did install the later version but must have
recently changed your $PATH or installed the HP package that has RCS
in it.
You should either reconfigure CVS to use absolute pathnames to the
proper versions of the RCS programs that CVS uses, or change your PATH
to look there first. If you haven't installed the latest version of
RCS, you should upgrade. See 1B.4
Last modified: _6/13/1997_
4. Explain: "error: RCS file name `xyz .c' contains white space"
RCS 5.6 doesn't allow white space in filenames. Apparently this
restriction will be removed in RCS 5.7, but CVS may still require that
filenames have no white space in them.
Last modified: _6/13/1997_
5. Explain: cvs checkout: warning: <X> is not (any longer) pertinent
This message occurs in three instances:
When there is an entry in the ./CVS/Entries for file <X> and there
is no RCS file in the Repository to back it up.
If the working file exists, and hasn't changed (determined from the
timestamp) it is removed.
When you try to check out a piece of the Repository with:
cvs checkout some/place/in/repository/tree
and at least the first element of the path (i.e. "some" in the above)
exists, but some part of the rest of it does not.
The checkout command checks the modules file first for the whole path,
then for a prefix of the path as a module name. If it doesn't find
*any* portion of your path in the modules file, it says:
cvs checkout: cannot find module `<module/path>' - ignored
If it finds some set of prefix directories, it prints the message you
see.
In practice this is usually a spelling error.
If the Repository files you are trying to check out or update are
not readable by you, the same problems can occur. Check the
permissions on the files involved.
Last modified: _6/13/1997_
6. Why did a Repository file change from <file>,v to ,<file>,?
This is an RCS problem, since the ,<file>, syntax for file names is
used by RCS and not CVS.
RCS constructs a new <file>,v in a temporary file named ,<file>,
(which doubles as a lock file) then renames it to <file>,v when it is
done. The only way this is reliable is if your system's version of
rename(2) is an atomic, as required by POSIX.
If your system has a non-atomic (and therefore non-POSIX) rename(2)
system call, RCS runs uses an internal version of this algorithm to
approximate the atomic rename:
rm <file>,v; ln ,<file>, <file>,v; rm ,<file>,
If the system crashes, or you lose your NFS connection between the
first "rm", but before the "ln", you can be left only with the
,<file>, file. If the crash or network failure occurs between the "ln"
and the final "rm", you could be left with a pair of linked names.
Recovery:
- If only the ,<file>, exists, rename it to <file>,v.
- If both ,<file>, and <file>,v exist and are linked, remove the
,<file>, file.
- If both ,<file>, and <file>,v exist and are separate files, look at
the dates, "diff" them and make your best guess. This sounds like the
remnants of two separate events.
Last modified: _6/13/1997_
Category: /Advanced_Topics_/Other_Systems/
" + Other Systems"
1. I use a NeXT. Is there anything I need to know?
NeXTSTEP 3.0's Interface Builder uses "nib" directories, rather than
the files used in previous revisions. It removes files it doesn't
recognize, making it impossible to place such a directory under CVS --
the CVS admin directory will be removed.
Some time ago, <Bob_Vadnais@pdh.com> posted a palette named CVSPalette
that claimed to resolve this problem. It was intended to preserve the
CVS administrative directories within nib documents (directories) that
Interface Builder usually removes.
CVSPalette is no longer in its announced place:
ftp.cs.orst.edu:/pub/next/submissions
though I did find two other interesting files on ftp.cs.orst.edu:
/software/NeXT/sources/tools/cvs-next-2_1_1.tar.Z
which is a port of CVS 1.3 (along with RCS and diff) and:
/software/NeXT/sources/programming/cvs.postamble-2.4.gz
which appears to be a set of wrappers for CVS commands that claim to
allow you to use CVS effectively (and without need for the "command
line") on a NeXT machine.
[[Anyone know the truth about CVS and NeXT?]]
Last modified: _6/13/1997_
2. I use OS/2 and/or DOS and/or Windows. Is there anything I need to know?
When using a local repository, be sure to specify the local access
method or CVS will interpret the drive letter as a remote host name
due to the : following it:
WRONG: CVSROOT=C:\SRC\CVSROOT
RIGHT: CVSROOT=:local:C:\SRC\CVSROOT
(larry.jones@sdrc.com)
You can share RCS files between Unix and DOS while avoiding the MS-DOS
file name limits by setting your RCSINIT environment variable to
'-x/,v'. New RCS files will be created without the standard ",v"
suffix, though files ending in ",v" will still be found if there is no
matching file in the same directory without the ",v".
Erik van Linstee offers an OS/2 and a DOS port of CVS 1.3 in:
ftp.informatik.tu-muenchen.de:/pub/comp/os/os2/gnu/devtools or
ftp.rrzn.uni-hannover.de:/pub/os2-local
The files are named:
cvs13p?[bs].zip
Where the ? stands for the patch level (currently 8) and the b is for
the binaries, the s for the sources.
There are three binaries. An OS/2 only one (32-bit), a DOS only one
(16-bit) and an EMX one that runs on both (32-bit).
There are many differences between the Unix and the DOS versions of
CVS. Read the material that comes with the DOS version before using
it.
[[Updates?]].
Last modified: _9/22/1997_
3. I use SCO Unix. Is there anything I need to know?
On SCO/UNIX 3.2 V2.0 POSIX signals don't work. Unfortunately the
configure program detects POSIXness and configures in the use of POSIX
signals. Workaround : Edit out the check for POSIXness in the
configure script. [[You could also remove all occurrences of
"-DPOSIX=1" from the Makefiles after configure is run. -dgg-]]
SCO/UNIX doesn't understand #!/<some shell> syntax. This breaks the
use of log.pl as it gets invoked by /bin/sh instead of
!#/usr/local/bin/perl. WorkAround : edit log.pl and change it into a
shell script which invokes perl with log.perl (renamed from log.pl) as
input.
Contributed by Joe Drumgoole
Last modified: _6/13/1997_
4. I use AIX. Is there anything I need to know?
The only report on AIX claims to have no trouble using it in concert
with SunOS and IRIX platforms.
Last modified: _6/13/1997_
5. I use IRIX. Is there anything I need to know?
If you see "uid" numbers where you would expect user names, try adding
-lsun to the link line. Without it CVS is unable to retrieve "passwd"
data through NIS.
Last modified: _6/13/1997_
6. I use an HP system. Is there anything I need to know?
HP distributes RCS version 3 (a circa 1983 release!) with HP-UX. CVS
does not work with RCS version 3; it requires RCS version 4 or later.
Your best bet is to find the latest version of RCS and install it
somewhere.
HP-UX 8.07 has a serious bug with the mmap system call and NFS files;
the bug can crash the operating system. Make sure that you configure
RCS to avoid mmap by setting has_mmap to 0 in RCS's conf.h. This bug
is fixed in HP-UX 9.
Contributed by Paul Eggert
If using the setgid() trick described in 4D.13, you will have to
create an entry in the /etc/privgroup file to give the group assigned
to the cvs executable setgid permission (see setprivgrp(1m)).
Additionally, if you are restricting "read" access to the Repository
by limiting access to the executable (this requires yet another
group), then you will require that /etc/logingroup exists and is
configured correctly (usually it's just alink to /etc/group).
Contributed by Dale Woolridge
Last modified: _6/13/1997_
7. I use AFS. Is there anything I need to know?
There is a problem with the way CVS performs its locking when the
files are within AFS. When your current PTS id != your uid, the locks
are not deleted. The stat() system call returns the PTS id of the
owner. If that id != your uid, CVS assumes you did not lock it, and
leaves the lock files alone. The next time you try to use it, it
complains that someone has the repository locked.
Contributed by Michael Ganzberger
[[This was against CVS 1.3. Is it still in CVS 1.4?]]
Last modified: _6/13/1997_
8. I use A/UX. Is there anything I need to know?
[[??]]
Last modified: _6/13/1997_
Category: /Advanced_Topics_/Related_Software/
" + Related Software"
1. How do I use CVS under Emacs? Is there an Emacs cvs-mode?
The pcl-cvs package distributed with CVS is an emacs package that
helps with the update/commit process. When you are ready to update,
you use the 'cvs-update' command within emacs. This executes "update"
and fills a cvs-mode buffer with a line for each file that changed.
The most helpful features are: descriptive words for what happened
(i.e. Merged or Conflict rather than 'U' or 'C'), single keys bound to
diffs and commits, and the ability to mark arbitrary groups of files,
possibly from different directories, for commit as a whole.
All the developers in my group that use emacs find pcl-cvs a much
friendlier and more helpful way to update/commit than raw cvs. One vi
user even converted to emacs just to use pcl-cvs.
Contributed by Jeffrey M Loomis
Last modified: _6/13/1997_
2. What is GIC (Graphical Interface to CVS)?
GIC provides a graphical user interface to the Concurrent Version
System (CVS), a powerful revision control system. GIC is
implemented in the Tcl/Tk programming language and is intended to
augment the sometimes cumbersome CVS command line interface.
Note that according to the official GIC page at
http://www.cpsc.ucalgary.ca/redirect/grouplab/projects/gic/
GIC is no longer being maintained and tkCVS is recommended
instead. For more on tkCVS, see http://www.cyclic.com/tkcvs/
kingdon@cyclic.com
Last modified: _9/6/1997_
3. What is CAVEMAN?
CAVEMAN is a front end to CVS written in PERL providing a collection
of features desired by the site where it was developed.
- The ability to spread a "project" over multiple Repositories.
- Optional automatic tagging after each commit.
- Additional locking of files.
- Extra before and after program hooks.
- A layer of event logging.
- All sorts of error messages.
- Many changes to the semantics of commands.
It is available via anonymous ftp on ftp.llnl.gov [128.115.54.18] in
gnu/caveman_vX.Y.Z.tar.gz (The numbers X, Y, & Z vary.)
contact Kathleen Dyer kdyer@llnl.gov
(510)423-6803
(510)423-5112 FAX
[[Does someone want to elaborate?]]
Last modified: _6/13/1997_
Category: /Advanced_Topics_/Setting_up_and_Manag/
" + Setting up and Managing the Repository"
1. What do I do first? How do I create a Repository?
First, install all the programs. (See Section 4A.)
Then create a Repository by executing "cvsinit", which works only from
within the head of the CVS source directory. (It needs files from the
distribution to work.)
If you want a very primitive Repository and don't want to save a
history log, refer to modules, or use any of the "info" files for
logging, pre-commit checks, or editing templates, you can dispense
with "cvsinit" entirely. I would advise executing it.
The cvsinit program will create a short modules file containing the
module named "CVSROOT". Change to your work directory and type:
cvs checkout CVSROOT
Then read the files that are checked out.
You will certainly want to add modules of your own. Edit the "modules"
file and add lines to describe the items you want to "checkout" by
module name. Here's a short list that could be used for storing a
small number of GNU and PD sources:
local local
gnu local/gnu
emacs local/gnu/emacs
cvs local/gnu/cvs
public local/public
pdprog1 local/public/pdprog1
pdprog2 local/public/pdprog2
test test
junk test/junk
When you are done editing, "commit" the modules file. If you
configured CVS to use "dbm", you might have to edit and commit the
modules file twice to change the pathname of the mkmodules program in
the modules file.
Try using the "import" command to insert the "junk" module and play
around until you are comfortable.
Last modified: _11/7/1997_
2. What are those files in $CVSROOT/CVSROOT?
There are eight Repository control (or "database") files of interest
in the CVSROOT directory:
modules contains the "modules" database. See 1D.11, 2C.7, 4B.6 and
4B.7 for more details.
commitinfo contains two columns: 1. a regular expression to match
against pathnames within the Repository and
a <command> to execute for matching pathnames.
When you execute "commit", CVS passes the Repository pathname for each
directory (and the files to commit within that directory) to
<command>. If <command> exits with a non-zero status, the commit is
blocked.
A <command> associated with a pathname of "DEFAULT" is executed if
nothing else matches. Every <command> associated with a pathname of
"ALL" is executed separately.
rcsinfo contains the same first column as commitinfo, but the second
column is a template file for specifying the log entry you are
required to enter for each commit.
"DEFAULT" and "ALL" work the same as in the commitinfo file.
editinfo contains the same two columns as commitinfo, but the
<command> in the second column is intended to do some consistency
checking on the commit log.
"DEFAULT" works as in commitinfo.
loginfo contains the same two columns as commitinfo, but the
<command> is expected to read a log message from its standard input.
The <command> can do anything it wants with the log information, but
normally it is appended to a log file or sent to mailing lists.
"DEFAULT" & "ALL" work the same as in commitinfo.
cvsignore contains "ignore" patterns that are added to the built-in
ignore list. See 2D.10.
checkoutlist contains a list of other files kept under RCS in
$CVSROOT/CVSROOT that should be checked out by mkmodules to provide a
readable copy.
history contains a stream of text records, one for each event that
the "history" command is interested in. Though the contents of the
history file can be read, it is intended to be read and displayed by
the "history" command. This file is the only one in the above list
that is not under RCS.
Last modified: _6/13/1997_
3. Is there any other state stored in the Repository besides in the
$CVSROOT/CVSROOT directory?
Only in the RCS files. The Repository holds exactly two things: the
tree of RCS files (each usually ending in ",v") and the CVSROOT
directory described above.
Last modified: _6/13/1997_
4. How do I put sources into the Repository?
There are three main ways to put files in the Repository:
Use the "import" command described in Section 3H.
This method is the fastest way to put trees of new code into the
Repository and the *only* way to handle source releases from a 3rd
party software vendor.
Use "add" followed by "commit".
This is how to add new files and directories to the Repository, a few
at a time. Directories don't need to be committed.
You can move RCS files directly into the Repository.
You should create a directory hierarchy to hold them, but you can just
move arbitrary ",v" files into the Repository. The only "state" in the
Repository other than within ",v" files is in the required CVSROOT
directory at the top of the Repository.
Last modified: _6/13/1997_
5. What file permissions should I use on (and in) the Repository?
If you run a completely open environment (which usually means that you
don't have, or don't want to waste, the time to deal with it):
- Set all directory permissions to 777.
- Have everyone set their umasks to 0.
(BTW, I don't suggest this. I am merely reporting it.)
If you are a normal Unix shop and want to use groups effectively:
- Set all the directory permissions in the Repository to 775.
If you are using a system that handles both System V and BSD
filesystems, you might have to set the permissions to 2775.)
If you are using one of the many recent versions of Unix that don't
allow you to use the full octal mode, then you'll have to type: chmod
u=rwx,g=rwx,o=rx,g+s <dir>
- Change all the groups on the directories to match the groups you
want to write to various directories.
- Make sure every user is in the appropriate groups.
- Have everyone set their umask to 002, including root.
If you don't want non-group members to even read the files, do the
above, but change:
- Repository directory permissions to 770. (or 2770)
- umasks to 007.
If you work in an environment where people can't be trusted to set
their "umask" to something reasonable, you might want to set the umask
for them:
mv /usr/local/bin/cvs /usr/local/bin/cvs.real
cat > /usr/local/bin/cvs
#!/bin/sh
umask 2 # Or whatever your site standard is.
exec /usr/local/bin/cvs.real ${1+"$@"}
^D
Last modified: _6/13/1997_
6. How do I structure my Repository?
The Repository holds your software. It can be all interrelated or it
can be a bunch of separately managed directories.
How you break a whole system down into its component parts, while
defining interfaces between them, is one aspect of "Software
Engineering", a discipline that requires the study of dozens of
strange and wonderful areas of the computer and management worlds.
CVS provides a way to keep track of changes to individual files, a way
to "tag" collections of files, and a way to "name" collections of
files and directories. That's all. Everything else is in the way you
apply it.
In other words, you should structure your Repository to match your
needs, usually tied in with the other tools you use to build, install
and distribute your work. Common needs include the ability to:
- mount (or automount) directories from many places in your
organization.
- check out just what you need and no more.
- check out multiple sections in a fixed relation to each other.
- check out large sections to match the assumptions built into your
build system. (Makefiles?)
In my opinion, you should start small and keep everything in one tree,
placing each major sub-system into a separate directory. Later, when
you know what you are doing, you can make it more sophisticated.
Last modified: _6/13/1997_
7. Why would anyone use "modules"? They are too restrictive. I want to be
able to select just the files I want to edit.
Any form of structure is restrictive. If you believe that total chaos
is a viable working paradigm, or if you believe you can keep track of
the interrelations between all portions of your Repository in your
head, then you can do what you please.
If you believe that systems of files require management and structure,
then the "modules" idea is very useful. It is a way to impose a naming
scheme on a tree of files, a naming scheme that can be simpler than a
large list of relative pathnames.
The "modules" file represents a published interface to the Repository
set up by your Repository Administrator. If s/he did a creditable job,
the modules offered will be internally consistent and will smoothly
interact with the rest of your environment.
Last modified: _6/13/1997_
8. How do I rename a file or directory? What are the consequences?
In CVS there is no single "rename" command.
See 2C.4 for the suggested way to rename a file or directory.
The rest of this section covers some of the consequences of renaming.
A "renaming database" has been proposed that would keep track of name
changes so that "update -r <tag>" would continue to work across the
renaming. But as it stands, you have to pick one of the following
options:
Use the technique described in 2C.4. (For each file, duplicate the
file in the Repository, "remove" the old version so it winds up in the
Attic and strip all Tags off the new version.)
- "update -r <tag>" produces the correct files.
- The duplicated revision history can be slightly misleading.
- A plain (i.e. without the "-r <tag>") "checkout" or "update -d" will
create directories "renamed" this way, but you can delete it and a
plain "update" won't bring it back.
Move the files and directories in the Repository to the new names.
- You save the revision history under a different file name.
- You save a little space.
- "update -r <tag>" produces the wrong files or directories.
This is not a good general solution, but if you plan never to look
back (someone may be gaining on you!), it is sometimes a useful
notion.
If you are clever with Makefiles, you might be able to rework them to
handle either the new or old names, depending on which ones exist at
the time. Then you can move an old <tag> onto the new, more
sophisticated, revision of the Makefile. (Yes, this changes the
"released" file if <tag> indicates a release. But it is an option.)
- Important Note: If you rename a directory, you must rename the
corresponding directory in every checked-out working directory. At the
same time, you must edit the pathname stored in the ./CVS/Repository
file within each of the moved directories.
The easiest way to move a lot of directories around is to tell
everyone to remove their working directories and check them out again
from scratch.
- The file exists in the working directory and in the ./CVS/Entries
file, but not in the Repository. For the old file, "update" prints:
cvs update: xyz.c is no longer in the repository
and deletes the file. If the file was modified, "update" prints:
cvs update: conflict: xyz.c is modified but no longer in the
repository C xyz.c
and leaves the file alone. In the new directory, you see:
U xyz.c
as you would if someone else executed "add" and "commit".
For each file, copy the working file to a new name in the working
directory and use the "cvs remove" to get rid of the old old file and
"cvs add" to add the new one. Since there is no way for CVS to remove
a directory, this only works for files.
- This is what most people think of first. Without a "rename" command,
the remove/add technique seems obvious.
- You lose the connection of your new working file to its past
revision history.
Last modified: _6/13/1997_
9. What are "Attic" directories?
When you use the "remove" command on a file, CVS doesn't delete the
file, it only registers your desire to delete it.
When you "commit" a removed file, CVS moves the Repository's matching
RCS file into a sub-directory named "Attic" within the Repository.
Attic files are examined when the '-r' or '-D' option is used on
"checkout" or "update". If the specified revision, tag or date matches
one on a file in the Attic, that file is checked out with the others.
You can think of the Attic as a sort of dead branch, which is only
looked at when you refer to a <tag> or <date>.
Last modified: _6/13/1997_
10. Is it OK to remove anything from the Repository?
In general, removing anything from the Repository is a bad idea. The
information in a deleted object is lost forever. There are many ways
to skip over files, directories and revisions without deleting them.
Here are some of the consequences of removing the following things
stored in the Repository:
CVSROOT files (Repository control files)
The Repository will work without any of them, but you should
understand what you are losing by deleting them. See 4B.2.
Revisions
The only way to remove revisions is to use the "admin -o" command (or
the equivalent RCS command "rcs -o").
They are lost forever. Any tags formerly attached to deleted revisions
are now pointing into the Phantom Zone. You'll need to contact Jor-el
to get them back.
Files
You should not remove a file unless you truly never want to see it
again. If you want to be able to check out an old revision of this
file, use "cvs remove" instead.
Tags
Tags take up little space and you can't recover from deleting them. If
you depend on tags for releases you will lose vital information.
Directories
There is no Attic for directories, so the only way to remove them is
to use "rm -r". They are gone forever.
If you delete (or move) a directory, all checked-out versions of that
directory will cause CVS to halt. You'll have to visit each
checked-out directory and remove the matching working directory by
hand.
Attic files
The "remove" command sends files to the Attic. To really delete them,
you have to go into the Attic and use "rm".
If a file in the Attic has a Tag on it that you might ever want to
check out again, you probably don't want to delete it.
Lock files (named: "#cvs.[wr]fl.<pid>")
These are lock files. If you are getting "lock" errors and the dates
on the lock files indicate that they are old, you can delete them.
Deleting lock files still in use by a CVS process might produce
unusual errors.
Last modified: _6/13/1997_
11. Can I convert to CVS from RCS without losing my revision history?
Yes, you can simply move (or copy) your RCS files into a directory
within the Repository, check out that directory and start working.
Last modified: _6/13/1997_
12. Can I move RCS files with branches in them into the Repository?
Yes, but they may not work if you created branches in a way that
conflicts with CVS's assumptions:
You can't use .0. branches. (They are reserved for "Magic" branch
tags.)
If you use branch 1.1.1, you can't use the Vendor branch.
You can use other RCS branches under CVS. There is no need to create
"magic" branch tags because the physical branch already exists.
Last modified: _6/13/1997_
13. Can I use raw RCS commands on the Repository?
You can use raw rcs commands directly on the Repository if you take a
little care. The Repository itself contains no "CVS state" (as opposed
to RCS revision histories) outside the CVSROOT directory.
But using raw RCS commands to change branches, tags or other things
that CVS depends on may render the files unusable.
See 4D.7 on RCS/CVS sharing of the Repository and Section 3B on the
"admin" command.
Last modified: _6/13/1997_
14. How do I convert from SCCS to RCS?
You'll have to execute something like "sccs2rcs" (in the CVS contrib
directory) on every file. Then you can move the resulting RCS files
into the Repository as described above.
Last modified: _6/13/1997_
15. How do I limit access to the Repository?
There are all sorts of ways to restrict access to Repository files,
none of which are hooked directly into CVS.
Techniques for limiting access include:
Training, management and good backups.
The best form of Repository control is a combination of:
- A reliable backup scheme (verify it!)
- Enough training to ensure your developers are competent and
knowledgeable about all areas of your sources.
- Effective management of the boundaries and grey areas.
In many cases, technical solutions to "security" problems are
inadequate. You should first try to avoid them.
Personal Opinion: In an environment where "unknowns" are allowed to
touch important sources the "owner" of the CVS Repository must be a
large, loud, vigorous lout with a well-balanced truncheon and the
right to use it. Don't underestimate the effectiveness of letting
everyone know they will be strapped into the stocks on the Town Common
and pelted with vegetables if they break something they don't
understand without first asking the experts.
Set Unix groups and permissions. See 4B.5. You can set different
owners, groups and permissions for each sub-directory within the
Repository if that helps.
Catch invocations of "commit" by defining pre-commit programs in the
"commitinfo" file. This is fairly powerful, since it can block commits
based on anything you can program. Take a look at the programs in the
"contrib" directory of the CVS source tree.
Use multiple Repositories, each with its own protection scheme. If
you use NFS (or AFS) you can even use "export" restrictions to various
groups of machines to keep (for example) the Engineering Repository
off the Customer Service machines.
Try the "setgid" trick described in 4D.13.
Try to use the RCS access control lists, though I don't think CVS
will handle them cleanly.
Edit the source code to CVS to add your own access control.
Last modified: _6/13/1997_
16. What are the Repository Administrator's responsibilities?
Generally, the Administrator should set "policy", create the
Repository and monitor its size and control files.
Some specific responsibilities include:
Examining the Repository once in a while to clean up:
Trash files left by misguided developers who mistake the Repository
for a working directory.
Non-RCS files. Other than the files CVS needs in the
$CVSROOT/CVSROOT directory, every file in the Repository should be an
RCS file.
Lock files (both CVS '#*' and RCS ',*' files) left around after
crashes.
Wrong permissions, groups and ownerships.
Locked files. (RCS locks, that is.)
Attic files that should never have been under CVS at all. Don't
blindly delete files from Attic directories -- they were mostly put
there (via the "cvs remove") for a reason. Files that should be
deleted are binary files (e.g. '*.o', 'core', executables) that were
mistakenly inserted by "import -I !".
Maintaining the modules file.
Storing site-specific ignore patterns in the
$CVSROOT/CVSROOT/cvsignore file.
Storing the names of non-standard CVSROOT files (See 4B.2) in the
$CVSROOT/CVSROOT/checkoutlist
Maintaining the other Repository control files: commitinfo, loginfo,
rcsinfo and editinfo.
Pruning the history file every once in a while. (Try the
"cln_hist.pl" script in the "contrib" directory.)
Staying aware of developments on the info-cvs mailing list and what
is available in the FTP and WWW archives.
Running "ps ax" once in a while and kill off any "update" programs
not running as "root". It is too easy to leave the "cvs" off the front
of the "cvs update" command.
Executing monitor programs to check the internal consistency of the
Repository files. Ideas:
Files that have a default RCS branch that is not 1.1.1 (From an
abuse of "admin -b".)
Files that have only Revisions 1.1 and 1.1.1.1, with a default
branch of "MAIN". (From an abuse of "admin -o".)
Existing branch tags and various branch consistency checks.
Last modified: _6/13/1997_
17. How do I move the whole Repository?
Copy or move the tree. (On Unix systems, a set of piped "tar" commands
works great. If the Repository does not contain any symlinks, which it
normally doesn't, you can also use "cp -r".)
If you can avoid changing $CVSROOT (i.e. the "logical" pathname of the
Repository) by replacing the old location with a symbolic link to the
new location, you don't have to do anything else.
(You could also mount the new location on top of the old location if
you are using NFS or some other filesystem that allows it.)
If you must change $CVSROOT, you must also tell everyone to change the
CVSROOT environment variable in all running shells and in any personal
configuration files ('.' files on Unix) where it is set.
The Repository itself contains no references to its own name, except
possibly in some of the files in the CVSROOT directory. If your
modules (or loginfo, commitinfo, etc.) file mentions helper programs
directly in the Repository, you'll have to change the pathnames to
point to the new Repository location.
The main changes you'll have to make are to all the CVS administrative
files (./CVS/Repository and ./CVS/Root) in every working directory
ever checked out from the previous location of the Repository you just
moved.
You have three choices:
If all ./CVS/Repository files in all working directories contain
relative pathnames, you don't have to do anything else.
Have everyone "release" or delete their working directories (after
committing, or just saving, their work) and check them all out again
from the new Repository after the move.
Use "find . ( -name Repository -o -name Root )" and a PERL or shell
script to run through all the ./CVS/Repository and ./CVS/Root files
and edit the values in the files.
Last modified: _6/13/1997_
18. How do I change permissions on a file in the Repository by using a CVS
command? (i.e. without using "chmod 777 $CVSROOT/dir/file")
When you first "import" or "add"/"commit" a file, the read and execute
bits on the Repository file are inherited from the original source
file, while the write bits on the Repository file are are turned off.
This is a standard RCS action.
After that, there is no way to alter the permissions on a file in the
Repository using CVS (or RCS) commands. You have to change the
permissions on both your working file and on the Repository file from
which it was retrieved.
Whenever you "checkout" the file or retrieve a new revision via
"update" (or after a "commit"), your working file is set to match the
permissions of the Repository file, minus any "umask" bits you have
set.
Last modified: _6/13/1997_
Category: /Advanced_Topics_/Tricks_of_the_Trade/
" + Tricks of the Trade"
1. How can you even check in binary files, let alone allow CVS to do its
auto-merge trick on them?
First of all, if you want to use binary files, you should get RCS 5.7
and CVS 1.9 or later (earlier versions had some support, but there have been
bug fixes). Secondly, follow the instructions for installing RCS very
carefully (it is easy to get it installed so it works for everything
except binary files).
Then, specify 'cvs add -kb' instead of just 'cvs add' to add a binary
file. If you want to set an existing file to binary, run 'cvs admin
-kb' (and then check in a new copy of the file). Note that old
versions of CVS used -ko instead of -kb for binary files, so if you
see a reference to -ko in the context of binary files, you should
think -kb instead.
Of course when 'cvs update' finds that a merge is needed, it can't
do this for binary files the same way as for text files. With the
latest versions (e.g. CVS 1.9.14), it should be able to give you both
versions and let you merge manually. Another approach is to
run 'cvs admin -l' to lock files, as described in
"How can I lock files while I'm working on them the way RCS does?"
elsewhere in this FAQ. See also
"Is there any way to import binary files?" and
"How do I "add" a binary file?" elsewhere in this FAQ.
kingdon@cyclic.com
Last modified: _9/6/1997_
2. Can I edit the RCS (",v") files in the Repository?
Yes, but be very careful. The RCS files are not free-form files, they
have a structure that is easily broken by hand-editing. The only time
I would suggest doing this is to recover from emergency failures that
are difficult to deal with using CVS commands, including the "admin"
command, which can talk directly to RCS.
Though no one actively encourages the editing of RCS files, many
people have succumbed to the urge to do so when pressed for time. The
reasons given, usually with evident contrition, include:
- Editing mistakes in, or adding text to, log entries. (If you have
RCS 5.6 or later, you should use `cvs admin -m'.)
- Renaming or moving symbolic names. (You should `cvs admin -N'
instead.)
- Unlocking a file by changing the "locker" from someone else to
yourself. (It's safer to use `cvs admin -u -l'.)
- Making global changes to past history. Example: Eradicating former
employees names from old documents and Author entries. (And someone
thought the "history" command was evidence of Big Brother! I never
realized how much help a wide-open revision control system could have
provided to The Ministry of Truth.)
Last modified: _6/13/1997_
3. Can I edit the ./CVS/{Entries,Repository,Tag} files?
Yes, but with CVS 1.3 and later, there is almost no reason to edit any
of the CVS administrative files.
If you move pieces of your Repository around it can be faster to edit
all the ./CVS/Repository files rather than checking out a large tree.
But that is nearly the only reason to do so.
Last modified: _6/13/1997_
4. Someone executed "admin -o" and removed revisions to which tags/symbols
were attached. How do I fix them?
It depends on what you mean by "fix". I can think of three ways to fix
your predicament:
Remove the tags.
Assuming you really wanted to get rid of the revision and its
associated tags, you can remove them with the "admin" command. The
"tag -d" command will only remove tags attached to existing revisions.
You can remove a tag, even if it is attached to a non-existent
revision, by typing:
cvs admin -N<tag> <file>
Retrieve the outdated revision.
You should first look in your backup system for recent versions of the
file. If you can't use them, you can carefully extract each revision
that followed the earliest outdated revision using RCS (or "cvs
admin") commands and reconstruct the file with all the right
revisions, branches and tags. This is a lot of work.
You *can't* insert a revision into the current RCS file.
Move the Tags to another revision in each file.
If you want to move the tags to another valid revision, you have two
choices, both of which require that you find all the revision numbers
of the files you want to "tag" and execute the following command
sequences on each <file>.
Use "update" to grab the revision you want, then execute a normal
"tag" command to Tag that revision:
cvs update -r <rev> <file>
cvs tag <tag> <file>
Use "admin" to set the tag to a specific revision:
cvs admin -N<tag>:<rev> <file>
Last modified: _6/13/1997_
5. How do I move or rename a magic branch tag?
(To rename a non-branch <tag> see 3O.9.)
Before reading this, read 3M.3 and 3M.4 and understand exactly how tag
and rtag use '-r' and why it won't do the right job here.
First, I have to explain exactly what a magic branch tag is.
A magic <branch_tag> is an artificial tag attached to a non-existent
revision on a non-existent branch number zero. It looks like this:
TAG1:<X>.0.Y
<X> is the "branch point revision", a normal revision with an
odd number of '.'s in it. (e.g. 1.5, 1.3.1.6, etc)
Y is an even number (e.g. 2, 4, 6, etc.) All CVS branches,
other than the Vendor branch, are even numbered.
TAG1 is considered by CVS to be attached to revision <X>. The first
"update -r TAG1 <file>" after applying TAG1 will produce a copy of
revision <X> with a sticky tag of TAG1. The first "commit" to that
file will cause CVS to construct an RCS branch named <X>.Y and check
in revision <X>.Y.1 on the new branch.
Note: TAG1 is *not* considered to be attached to <X> by RCS, which
explains why you can't refer directly to the branch point revision for
some CVS commands.
Moving a magic <branch_tag> is the act of reapplying the same tag to
different revisions in the file:
TAG1:<X>.0.Y
to
TAG1:<X>.0.Z or TAG1:<A>.0.B
You can move a magic branch tag to the revisions of your choice by
using "update" to find the revisions you want to tag and reapplying
the tag to all the files with the '-F' option to force it to move the
existing <branch_tag>.
cvs update -r <tag/rev> (or '-A' for the Main Branch)
cvs tag -F -b <branch_tag>
If the earlier location of TAG1 refers to a physical branch within any
RCS file, moving it will make the existing branch in the file seem to
disappear from CVS's view. This is not a good idea unless you really
want to forget the existence of those RCS branches.
If the "update" above retrieves the original branch point revision
(<X>), the "tag" command above will create the tag:
TAG1:<X>.0.Z
Where Z is 2 greater than the highest magic branch already on revision
<X>. The TAG1 branch will still have the same branch point (i.e.
revision <X>), but the first commit to the new TAG1 branch will create
a different RCS branch number (<X>.Z instead of <X>.Y).
Renaming a magic <branch_tag> is the act of changing
TAG1:<X>.0.Y
to
TAG2:<X>.0.Y
There is no harm in changing a tag name as long as you forget that
TAG1 ever existed and you clean up any working directories with sticky
TAG1 tags on them by using "update -A", "update -r <other_tag>" or by
removing the working directories.
On the other hand, actually changing the tag is not easy.
See 3M.3 for why the seemingly obvious solution won't work:
cvs tag -b -r <old_branch_tag> <new_branch_tag>
The only direct way to rename a magic tag is to use the "admin"
command on each file: (You might want to use '-n'. Read "man rcs" and
look at the '-n' and '-N' options.)
cvs admin -N<new_branch_tag>:<old_branch_tag> .
cvs tag -d <old_branch_tag>
But you have to be careful because "admin" is different from other CVS
commands:
"admin" can be used recursively, but only by specifying directory
names in its argument list (e.g. '.'),
Where "rtag -r <old_branch_tag>" would interpret <old_branch_tag> as
a magic CVS branch tag, "admin" is a direct interface to RCS which
sees a magic branch tag as a simple (though non-existent) RCS revision
number.
This is good for us in this particular case, but different from normal
CVS.
"admin" also skips the Attic and produces different kinds of errors
than CVS usually does. (Because they are coming directly from RCS.)
The other way to rename a magic <branch_tag> is to edit the Repository
files with a script of some kind. I've done it in the past, but I'll
leave it as an exercise for the reader.
Last modified: _6/13/1997_
6. Can I use RCS locally to record my changes without making them globally
visible by committing them?
You can, but it will probably confuse CVS to have ",v" files in your
working directory. And you will lose all your log entries when you
finally commit it.
Your best bet is to create your own CVS branch and work there. You can
commit as many revisions as you want, then merge it back into the main
line (or parent branch) when you are finished.
Last modified: _6/13/1997_
7. How can I allow access to the Repository by both CVS and RCS?
The first step is to try not to. If some people are using CVS, there
is no reason for everyone not to. It is not hard to learn the basics
and CVS makes certain operations *easier* than a series of RCS
commands. Personal preference in what software tools can be applied to
a shared Repository has to take second place to system integration
needs. If you disagree, try writing some Lisp code for inclusion in
your Unix kernel and see what kind of reception you get.
If you really must allow routine RCS access to the CVS Repository, you
can link an RCS sub-directory into a piece of the Repository:
ln -s /Repository/some/directory/I/want RCS
and RCS will work just fine.
Those who are using RCS will have to keep the following in mind:
If a file was originally added to the Repository by "import" and has
not been changed using CVS, the *RCS* default branch will remain
attached to the Vendor branch, causing revisions checked-in by "ci" to
wind up on the Vendor branch, instead of the main branch. Only CVS
moves the RCS default branch on first commit.
The way around this is to checkin (using "ci") all the files first and
move them into the Repository. That way they won't have Vendor
branches. Then RCS will work OK.
It is possible to use "rcs" and "ci" to make the files unusable by
CVS. The same is true of the CVS "admin" command.
Normal RCS practice locks a file on checkout with "co -l". In such
an environment, RCS users should plan to keep survival gear and food
for at least 30 days near their desks. When faced with bizarre and
unexpected permission errors, howling mobs of slavering CVS users will
run the RCS users out of town with pitchforks and machetes.
See 3C.8 for a way to avoid machetes aroused by lock collisions.
Though files checked in by RCS users will correctly cause
"up-to-date" failures during CVS "commits" and they will be
auto-merged into CVS working directories during "update", the opposite
won't happen.
RCS users will get no warning and will not be required to merge older
work into their code. They can easily checkin an old file on top of a
new revision added by CVS, discarding work committed earlier by CVS
users.
See the howling mob scenario described above.
RCS is great. I have used it for years. But I wouldn't mix it this
way. In a two-camp society, you are asking for real trouble, both in
technical hassles to clean up and in political hassles to soothe.
Branch merges will also be a major problem.
Last modified: _6/13/1997_
8. I "updated" a file my friend, "bubba", committed yesterday. Why doesn't
the file now have a modified date of yesterday?
CVS restores dates from the RCS files only on first "checkout". After
that, it is more important to maintain a timestamp relative to the
other files in the working directory.
Example: You committed a source file at 5PM. Bubba updated his copy of
the file, grabbing your changes, then changed and committed a new
revision of the file at 6PM. At 7PM, you compile your file. Then you
execute "update". If CVS sets the date to the one in the RCS file, the
file would be given a timestamp of 6PM and your Makefile wouldn't
rebuild anything that depended on it. Bad news.
Note that the same logic applies to retrieving a revision out of the
Repository to replace a deleted file. If CVS changes your file in an
existing working directory, whether it was because a new revision was
committed by someone else or because you deleted your working file,
the timestamp on the retrieved working file *must* be set to the
current time.
When you first retrieve a file, there is no reason to expect any
particular timestamp on the file within your working area. But later,
when dependency checking is performed during a build, it is more
important for the timestamps on the local files to be consistent with
each other than than it is for working files to match the timestamps
on the files in the Repository. See 4D.17 for some more about
timestamps.
Last modified: _6/13/1997_
9. While in the middle of a large "commit", how do I run other commands,
like "diff" or "stat" without seeing lock errors?
Type:
cvs -n <command>
The '-n' option to the main cvs command turns off lock checking, a
reasonable act for read-only commands given the promise offered by
'-n' not to alter anything. The "diff", "log" and "stat" commands
provide the same information (for files that are not being committed)
when used with and without the '-n' option.
Warning: Ignoring locks can produce inconsistent information across a
collection of files if you are looking at the revisions affected by an
active commit. Be careful when creating "patches" from the output of
"cvs -n diff". If you are looking only at your working files, tagged
revisions, and BASE revisions (revisions whose numbers are read from
your ./CVS/Entries files), you should get consistent results. Of
course, if you catch a single file in the middle of RCS activity, you
might get some strange errors.
Note that the suggested command is "cvs -n <command>". The visually
similar command "cvs <command> -n" has no relation to the suggested
usage and has an entirely different meaning for each command.
"cvs -n update" also works in the middle of a commit, providing
slightly different information from a plain "cvs update". But, of
course, it also avoids modifying anything.
You could also use the RCS functions, "rlog" and "rcsdiff" to display
some of the information by referring directly to the Repository files.
You need RCS version 5 or later for the commands described above to
work reliably.
Last modified: _6/13/1997_
10. Where did the ./CVS/Entries.Static file come from? What is it for?
Each CVS working directory contains a ./CVS/Entries file listing the
files managed by CVS in that working directory. Normally, if the
"update" command finds a file in the Repository that is not in the
./CVS/Entries file, "update" copies the appropriate revision of the
"new" file out of the Repository and adds the filename to the Entries
file. This happens for files:
Added to the Repository from another working directory.
Dragged out of the Attic when switching branches with "update -A" or
"update -r".
Whose names were deleted from the ./CVS/Entries file.
If the ./CVS/Entries.Static file exists, CVS will only bring out
revisions of files that are contained in either ./CVS/Entries or
./CVS/Entries.Static. If a Repository file is found in *neither* file,
it is ignored.
The ./CVS/Entries.Static file is created when you check out an
individual file or a module that creates working directories that
don't contain all files in the corresponding Repository directory. In
those cases, without an ./CVS/Entries.Static file, a simple "update"
would bring more files out of the Repository than the original
"checkout" wanted.
The ./CVS/Entries.Static file can be removed by hand. It is
automatically removed if you run "update -d" to create new directories
(even if no new directories are created). (Internally, since
"checkout" turns on the '-d' flag and calls the "update" routine, a
"checkout" of a module or directory that writes into an existing
directory will also remove the ./CVS/Entries.Static file.)
Last modified: _6/13/1997_
11. Why did I get the wrong Repository in the loginfo message?
You probably:
Use multiple Repositories.
Configured CVS to use absolute pathnames in the ./CVS/Repository
file.
Configured CVS not to use the ./CVS/Root file.
Typed the "commit" command in one Repository with your $CVSROOT
pointing at another.
"commit" and all other CVS commands will heed an absolute pathname in
the ./CVS/Repository file (or in the "-d CVSrootdir" override), but
the log function doesn't take arguments -- it just looks at $CVSROOT.
If you avoid even one of the four steps above, you won't see this
problem. If you configure ./CVS/Root, you won't be allowed to execute
the program causing the error.
Last modified: _6/13/1997_
12. How do I run CVS setuid so I can only allow access through the CVS
program itself?
Setuid to root is not a great idea. Any program that modifies files
and is used by a widely distributed group of users is not a good
candidate for a setuid program. (The worst suggestion I've ever heard
was to make *Emacs* setuid to root.)
Root access on Unix is too powerful. Also, it might not work in some
(secure?) environments.
Running it setuid to some user other than root might work, if you add
this line to main.c near the beginning:
setuid(geteuid());
Otherwise it uses *your* access rights, rather than the effective
uid's.
Also, you have to invent a fake user whose name will show up in
various places. But many sites, especially those who might want a
setuid CVS for "security", want personal accountability -- no generic
accounts. I don't know whether accountability outweighs file security.
And finally, unless you take action to limit the "admin" command, you
are leaving yourself unprotected anyway.
Last modified: _6/13/1997_
13. How about using groups and setgid() then?
Here is a way to run CVS setgid in some environments:
Stick this near the front of the main() in main.c:
setgid(getegid());
This will allow "access" to work on systems where it only works on the
real gid.
Create a group named "cvsg". (This example uses "cvsg". You can name
it as you wish.)
Put *no* users in the "cvsg" group. You can put Repository
administrators in this group if you want to.
Set the cvs executable to setgid (not setuid):
cd /usr/local/bin; chown root.cvsg cvs; chmod 2755 cvs
Make sure every file in the Repository is in group "cvsg":
chown -R root.cvsg $CVSROOT
Change all directory permissions to 770. This allows all access to
the files by the "cvsg" group (which has no members!) and no access at
all to anyone else.
find $CVSROOT -type d -exec chmod 2770 {} \;
On some systems you might have to type:
find $CVSROOT -type d -exec chmod u=rwx,g=rwx,o=,g+s {} \;
This should allow only the cvs program (or other "setgid to group
cvsg") programs to write into the area, but no one else. Yes the user
winds up owning the file, but s/he can't find it again later since
s/he can't traverse the tree. (If you enable the world execute bit
(mode 2771) on directories, users can traverse the tree and the user
who last wrote the file can still write to it.)
If you want to allow read access, check out an entire tree somewhere.
You have to do this anyway to build it.
Note: If you are using a stupid file system that can't inherit file
groups from the parent directory (even with the "setgid" (Octal 2000)
bit set), you might have to modify CVS (or RCS) to reset the group
every time you create a new file. I have not tested this.
The setgid() method shares with the setuid() method the problem of
keeping "admin" from breaking things.
Last modified: _6/13/1997_
14. How do I use the "commitinfo" file?
Go read 4B.2 first.
The "commitinfo" file allows you to execute "sanity check" functions
before allowing a commit. If any function called from within the
commitinfo file exits with a non-zero status, the commit is denied.
To fill out a "commitinfo" file, ask yourself (and those sharing your
Repository) these questions:
- Is there anything you want to check or change before someone is
allowed to commit a file? If not, forget commitinfo.
If you want to serialize binary files, you might consider something
like the rcslock.pl program in the contrib directory of the CVS
sources.
- Do you want to execute the same exact thing before committing to
every file in the Repository? (This is useful if you want to program
the restrictions yourself.) If so, set up a single line in the
commitinfo:
DEFAULT /absolute/path/to/program
CVS executes the program once for each directory that "commit"
traverses, passing as arguments the directory and the files to be
committed within that directory.
Write your program accordingly. Some examples exist in the contrib
directory.
- Do you want a different kind of sanity check performed for different
directories? If so, you'll have to decide what to do for all
directories and enter lines like this:
regexp1 /absolute/path/to/program-for-regexp1
regexp2 /absolute/path/to/program-for-regexp2
DEFAULT /absolute/path/to/program-for-all-else
- Is there anything you want to happen before *all* commits, in
addition to other pattern matches? If so, include a line like this:
ALL /absolute/path/to/program
It is executed independently of all the above. And it's repeatable --
you can have as many ALL lines as you like.
Last modified: _6/13/1997_
15. How do I use the "loginfo" files?
See 4B.2 and the "commitinfo" question above.
The "loginfo" file has the same format as the "commitinfo" file, but
its function is different. Where the "commitinfo" information is used
before a commit, the "loginfo" file is used after a commit.
All the commands in the "loginfo" file should read data from standard
input, then either append it to a file or send a message to a mailing
list. If you want to make it simple, you can put shell (the shell used
by "popen(3)") command lines directly in the "loginfo" (or
"commitinfo") file. These seem to work:
^special /usr/ucb/Mail -s %s special-mailing-list ^other /usr/ucb/Mail
-s %s other-mailing-list DEFAULT (echo '===='; echo %s; cat) >
/path/name/to/log/file
Last modified: _6/13/1997_
16. How can I keep people with restrictive umask values from blocking
access to the Repository?
If a user creates a new file with restricted permissions (e.g. 0600),
and commits it, the Repository will have a file in it that is
unreadable by everyone. The 0600 example would be unreadable by
*anyone* but root and the user who created it.
There are 3 solutions to this:
Let it happen. This is a valid way to protect things. If everyone is
working alone, a umask of 077 is OK. If everyone is working only in
small groups, a umask of 007 is OK.
Train your users not to create such things if you expect to share
them.
See 4B.5 for a small script that will reset the umask.
I personally don't like the idea of a program automatically
*loosening* security. It would be better for you all to talk about the
issue and decide how to work together.
Last modified: _6/13/1997_
17. Why do timestamps sometimes get set to the date of the revision,
sometimes not? The inconsistency causes unnecessary recompiles.
The "checkout" command normally sets the timestamp of a working file
to match the timestamp stored on the revision in the Repository's RCS
file.
The "commit" command retains the timestamp of the file, if the act of
checking it in didn't change it (by expanding keywords).
The "update" command sets the time to the revision time the first time
it sees the file. After that, it sets the time of the file to the
current time. See 4D.8 for a reason why.
Here's a two-line PERL program to set timestamps on files based on
other timestamps. I've found this program useful. When you are certain
you don't want a source file to be recompiled, you can set its
timestamp to the stamp on the object file.
#!/usr/local/bin/perl
#
# Set timestamp of args 2nd-Last to that of the first arg.
#
($dev,$ino,$mode,$nlink,$uid,$gid,$rdev,$size,$atime,$mtime,$ctime)
= stat(shift);
utime($atime,$mtime,@ARGV);
Last modified: _6/13/1997_
Category: /Commands_/
" Commands "
Category: /Commands_/add_ad_new/
" + "add", "ad", "new""
1. What is "add" for?
To add a new directory to the Repository or to register the desire to
add a new file to the Repository.
The directory is created immediately, while the desire to add the file
is recorded in the local ./CVS administrative directory. To really add
the file to the Repository, you must then "commit" it.
Last modified: _6/13/1997_
2. How do I add a new file to the branch I'm working on?
The user actions for adding a file to any branch, including the Main
Branch, are exactly the same.
You are in a directory checked out (or updated) with the '-A' option
(to place you on the Main Branch) or the "-r <branch_tag>" option (to
place you on a branch tagged with <branch_tag>). To add <file> to the
branch you are on, you type:
cvs add <file>
cvs commit <file>
If no ./CVS/Tag file exists (the '-A' option deletes it), the file
will be added to the Main Branch. If a ./CVS/Tag file exists (the "-r
<branch_tag>" option creates it), the file will be added to the branch
named (i.e. tagged with) <branch_tag>.
Unless you took steps to first add the file to the Main Branch, your
new file ends up in the Attic.
Last modified: _6/13/1997_
3. Why did my new file end up in the Attic?
The file is thrown into the Attic to keep it from being visible when
you check out the Main Branch, since it was never committed to the
Main Branch.
Last modified: _6/13/1997_
4. Now that it's in the Attic, how do I connect it to the Main branch?
That can be considered a kind of "merge". See 4C.8
Last modified: _6/13/1997_
5. How do I avoid the hassle of reconnecting an Attic-only file to the Main
Branch?
You create it on the Main Branch first, then branch it.
If you haven't yet added the file or if you decided to delete the new
Attic file and start over, then do the following: (If you added the
file (or worse, the 157 files) to the Attic and don't want to start
over, try the procedure in 4C.8.)
Temporarily remove the sticky branch information. Either:
Move the whole directory back to the Main Branch. [This might not be
a good idea if you have modified files, since it will require a merge
in each direction.]
cvs update -A
*or*
Move the ./CVS/Tag file out of the way.
mv ./CVS/Tag HOLD_Tag
Add and branch the file "normally":
cvs add <file>
cvs commit <file>
cvs tag -b <branch_tag> <file>
[<branch_tag> is the same Branch Tag as you used on all the other
files. Look at ./CVS/Entries or the output from "cvs stat" for sticky
tags.]
Clean up the temporary step.
If you moved the ./CVS/Tag file, put it back. Then move the new file
onto the branch where you are working.
mv HOLD_Tag ./CVS/Tag
cvs update -r <branch_tag> <file>
If you ran "update -A" rather than moving the ./CVS/Tag file, move
the whole directory (including the new file) back onto the branch
where you were working:
cvs update -r <branch_tag>
Last modified: _6/13/1997_
6. How do I cancel an "add"?
If you want to remove the file entirely and cancel the "add" at the
same time, type:
cvs remove -f <file>
If you want to cancel the "add", but leave the file as it was before
you typed "cvs add", then you have to fake it:
mv <file> <file>.hold
cvs remove <file>
mv <file>.hold <file>
Last modified: _6/13/1997_
7. What are the ./CVS/file,p and ./CVS/file,t files for?
The ./CVS/file,p and ./CVS/file,t files are created by the "add"
command to hold command line options and message text between the time
of the "add" command and the expected "commit".
The ./CVS/file,p file is always null, since its function was absorbed
by the "options" field in the ./CVS/Entries file. If you put something
in this file it will be used as arguments to the RCS "ci" command that
commit uses to check the file in, but CVS itself doesn't put anything
there.
The ./CVS/file,t file is null unless you specify an initial message in
an "add -m 'message'" command. The text is handed to "rcs -i
-t./CVS/file,t" to create the initial RCS file container.
Both files must exist to commit a newly added file. If the
./CVS/file,p file doesn't exist, CVS prints an error and aborts the
commit. If the ./CVS/file,t file doesn't exist, RCS prints an error
and CVS gets confused, but does no harm.
To recover from missing ,p and ,t files, just create two zero-length
files and rerun the "commit".
Last modified: _6/13/1997_
8. How do I "add" a binary file?
If you configured CVS to use the GNU version of "diff" and "diff3",
you only need to turn off RCS keyword expansion.
First you turn off RCS keyword expansion for the initial checkin by
using "add -ko". It works like "update -ko" in creating a "sticky"
option only for the copy of the file in the current working directory.
cvs add -ko <file>
Commit the file normally. The sticky -ko option will be used.
cvs commit <file>
Then mark the RCS file in the Repository so that keyword expansion is
turned off for all checked out versions of the file.
cvs admin -ko <file>
Since "admin -ko" records the keyword substitution value in the
Repository's RCS file, you no longer need the sticky option. You can
turn it off with the "update -A" command, but if you were on a branch,
you'll have to follow it "update -r <branch_tag>" to put yourself back
on the branch.
Managing that binary file is another problem. See 4D.1.
Last modified: _6/13/1997_
Category: /Commands_/admin_adm_rcs/
" + "admin", "adm", "rcs""
1. What is "admin" for?
To provide direct access to the underlying "rcs" command (which is not
documented in this FAQ) bypassing all safeguards and CVS assumptions.
Last modified: _6/13/1997_
2. Wow! Isn't that dangerous?
Yes.
Though you can't hurt the internal structure of an RCS file using its
own "rcs" command, you *can* change the underlying RCS files using
"admin" in ways that CVS can't handle.
If you feel the need to use "admin", create some test files with the
RCS "ci" command and experiment on them with "rcs" before blasting any
CVS files.
Last modified: _6/13/1997_
3. What would I normally use "admin" for?
Normally, you wouldn't use admin at all. In unusual circumstances,
experts can use it to set up or restore the internal RCS state that
CVS requires.
You can use "admin -o" (for "outdate") to remove revisions you don't
care about. This has its own problems, such as leaving dangling Tags
and confusing the "update" command.
There is some feeling among manipulators of binary files that "admin
-l" should be used to serialize access. See 3C.8.
An interesting use for "admin" came up while maintaining CVS itself. I
import versions of CVS onto the Vendor branch of my copy of CVS, make
changes to some files and ship the diffs (created by "cvs diff -c -r
TO_BRIAN") off to Brian Berliner. After creating the diff, I retag
("cvs tag -F TO_BRIAN") the working directory, which is then ready to
produce the next patch.
I'll use "add.c" as an example (only because the name is short).
When the next release came out, I discovered that the released "add.c"
(version 1.1.1.3 on the Vendor branch) was exactly the same as my
modified file (version 1.3). I didn't care about the changelog on
versions 1.2 and 1.3 (or the evidence of having done the work), so I
decided to revert the file to the state where it looked like I had not
touched the file -- where I was just using the latest on the vendor
branch after a sequence of imports.
To do that, I removed all the revisions on the main branch, except for
the original 1.1 from which the Vendor branch sprouts:
cvs admin -o1.2: add.c
Then I set the RCS "default branch" back to the Vendor branch, the way
import would have created it:
cvs admin -b1.1.1 add.c
And I moved the "TO_BRIAN" Tag to the latest revision on the Vendor
branch, since that is the base from which further patches would be
created (if I made any):
cvs admin -NTO_BRIAN:1.1.1.3 add.c
Instead of 1.1.1.3, I could have used one of the "Release Tags" last
applied by "import" (3rd through Nth arguments).
Suggestion: Practice on non-essential files.
Last modified: _6/13/1997_
4. What should I avoid when using "admin"?
If you know exactly what you are doing, hack away. But under normal
circumstances:
Never use "admin" to alter branches (using the '-b' option), which CVS
takes very seriously. If you change the default branch, CVS will not
work as expected. If you create new branches without using the "tag
-b" command, you may not be able to treat them as CVS branches.
See 3C.8 for a short discussion of how to use "admin -l" for
serializing access to binary files.
The "admin -o <file>" allows you to delete revisions, usually a bad
idea. You should commit a correction rather than back out a revision.
Outdating a revision is prone to all sorts of problems:
Discarding data is always a bad idea. Unless something in the
revision you just committed is a threat to your job or your life,
(like naming a function "<boss's name>_is_a_dweeb", or including the
combination to the local Mafioso's safe in a C comment), just leave it
there. No one cares about simple mistakes -- just commit a corrected
revision.
The time travel paradoxes you can cause by changing history are not
worth the trouble. Even if CVS can't interfere with your parents'
introduction, it *can* log commits in at least two ways (history and
loginfo). The reports now lie -- the revision referred to in the logs
no longer exists.
If you used "import" to place <file> into CVS, outdating all the
revisions on the Main branch back to and including revision 1.2 (or
worse, 1.1), will produce an invalid CVS file.
If the <file>,v file only contains revision 1.1 (and the connected
branch revision 1.1.1.1), then the default branch must be set to the
Vendor branch as it was when you first imported the file. Outdating
back through 1.2 doesn't restore the branch setting. Despite the above
admonition against it, "admin -b" is the only way to recover:
cvs admin -b1.1.1 <file>
Although you can't outdate a physical (RCS) branch point without
removing the whole branch, you *can* outdate a revision referred to by
a magic branch tag. If you do so, you will invalidate the branch.
If you "outdate" a tagged revision, you will invalidate all uses of
the <tag>, not just the one on <file>. A tag is supposed to be
attached to a consistent set of files, usually a set built as a unit.
By discarding one of the files in the set, you have destroyed the
utility of the <tag>. And it leaves a dangling tag, which points to
nothing.
And even worse, if you commit a revision already tagged, you will
alter what the <tag> pointed to without using the "tag" command. For
example, if revision 1.3 has <tag> attached to it and you "outdate"
the 1.3 revision, <tag> will point to a nonexistent revision. Although
this is annoying, it is nowhere near as much trouble as the problem
that will occur when you commit to this file again, recreating
revision 1.3. The old tag will point to the new revision, a file that
was not in existence when the <tag> was applied. And the discrepancy
is nearly undetectable.
If you don't understand the above, you should not use the admin
command at all.
Last modified: _6/13/1997_
5. How do I restrict the "admin" command? The -i flag in the modules file
can restrict commits. What's the equivalent for "admin"?
At this writing, to disable the "admin" command, you will have to
change the program source code, recompile and reinstall.
Last modified: _6/13/1997_
6. I backed out a revision with "admin -o" and committed a replacement. Why
doesn't "update" retrieve the new revision?
CVS is confused because the revision in the ./CVS/Entries file matches
the latest revision in the Repository *and* the timestamp in the
./CVS/Entries file matches your working file. CVS believes that your
file is "up-to-date" and doesn't need to be updated.
You can cause CVS to notice the change by "touch"ing the file.
Unfortunately what CVS will tell you is that you have a "Modified"
file. If you then "commit" the file, you will bypass the normal CVS
check for "up-to-date" and will probably commit the revision that was
originally removed by "admin -o".
Changing a file without changing the revision number confuses CVS no
matter whether you did it by replacing the revision (using "admin -o"
and "commit" or raw RCS commands) or by applying an editor directly to
a Repository (",v") file. Don't do it unless you are absolutely
certain no one has the latest revision of the file checked out.
The best solution to this is to institute a program of deterrent
flogging of abusers of "admin -o".
The "admin" command has other problems." See 3B.4 above.
Last modified: _6/13/1997_
Category: /Commands_/checkout_co_get/
" + "checkout", "co", "get""
1. What is "checkout" for?
To acquire a copy of a module (or set of files) to work on.
All work on files controlled by CVS starts with a "checkout".
Last modified: _6/13/1997_
2. What is the "module" that "checkout" takes on the command line?
It is a name for a directory or a collection of files in the
Repository. It provides a compact name space and the ability to
execute before and after helper functions based on definitions in the
modules file.
See 1D.11.
Last modified: _6/13/1997_
3. Isn't a CVS "checkout" just a bunch of RCS checkouts?
Like much of CVS, a similar RCS concept is used to support a CVS
function. But a CVS checkout is *not* the same as an RCS checkout.
Differences include:
CVS does not lock the files. Others may access them at the same
time.
CVS works best when you provide a name for a collection of files (a
module or a directory) rather than an explicit list of files to work
on.
CVS remembers what revisions you checked out and what branch you are
on, simplifying later commands.
Last modified: _6/13/1997_
4. What's the difference between "update" and "checkout"?
The "checkout" and "update" commands are nearly equivalent in how they
treat individual files. They differ in the following ways:
The "checkout" command always creates a directory, moves into it,
then becomes equivalent to "update -d".
The "update" command does not create directories unless you add the
'-d' option.
"Update" is intended to be executed within a working directory
created by "checkout". It doesn't take a module or directory argument,
but figures out what Repository files to look at by reading the files
in the ./CVS administrative directory.
The two commands generate completely different types of records in
the "history" file.
Last modified: _6/13/1997_
5. Why can't I check out a file from within my working directory?
Though you *can* check out a file, you normally check out a module or
directory. And you normally do it only once at the beginning of a
project.
After the initial "checkout", you can use the "update" command to
retrieve any file you want within the checked-out directory. There is
no need for further "checkout" commands.
If you want to retrieve another module or directory to work on, you
must provide two pathnames: where to find it in the Repository and
where to put it on disk. The "modules" file and your current directory
supply two pieces of naming information. While inside a checked-out
working directory, the CVS administrative information provides most of
the rest.
You should be careful not to confuse CVS with RCS and use "checkout"
in the RCS sense. An RCS "checkout" (which is performed by the RCS
"co" command) is closer to a "cvs update" than to a "cvs checkout".
Last modified: _6/13/1997_
6. How do I avoid dealing with those long relative pathnames?
This question has also been phrased:
How do I avoid all those layers of directories on checkout? or Why do
I have to go to the top of my working directory and checkout some long
pathname to get a file or two?
This type of question occurs only among groups of people who decide
not to use "modules". The answer is to use "modules".
When you hand the "checkout" command a relative pathname rather than a
module name, all directories in the path are created, maintaining the
same directory hierarchy as in the Repository. The same kind of
environment results if you specify a "module" that is really an alias
expanding into a list of relative pathnames rather than a list of
module names.
If you use "module" names, "checkout" creates a single directory by
the name of the module in your current directory. This "module"
directory becomes your working directory.
The "module" concept combines the ability to "name" a collection of
files with the ability to structure the Repository so that consistent
sets of files are checked out together. It is the responsibility of
the Repository Administrators to set up a modules file that describes
the software within the Repository.
Last modified: _6/13/1997_
7. Can I move a checked-out directory? Does CVS remember where it was
checked out?
Yes and Yes.
The ./CVS/Repository file in each working directory contains a
pathname pointing to the matching directory within the Repository. The
pathname is either absolute or relative to $CVSROOT, depending on how
you configured CVS.
When you move a checked-out directory, the CVS administrative files
will move along with it. As long as you don't move the Repository
itself, or alter your $CVSROOT variable, the moved directory will
continue to be usable.
CVS remembers where you checked out the directory in the "history"
file, which can be edited, or even ignored if you don't use the
"working directory" information displayed by the "history" command.
Last modified: _6/13/1997_
8. How can I lock files while I'm working on them the way RCS does?
Until the day arrives of the all-powerful merge tool, there are still
files that must be accessed serially. For those instances, here's a
potential solution:
Install a pre-commit program in the "commitinfo" file to check for
RCS locks. The program "rcslock.pl" performs this function. It can be
found in the contrib directory of the CVS source distribution.
When you want to make a change to a file you know can't be merged,
first use "cvs admin -l" to lock the file. If you can't acquire the
lock, use the standard "locked out" protocol: go talk to the person
holding the lock.
Make sure the pre-commit program prints a message and exits with a
non-zero status if someone besides the user running "commit" has the
file locked. This non-zero exist status will cause the "commit" to
fail cleanly.
Make sure the pre-commit program exits with a zero status if the
file is either unlocked or locked by the user running "commit". The
"cvs commit" command that kicked off the pre-commit program will take
a zero exist status as an OK and checkin the file, which has the
side-effect of unlocking it.
===> The following is opinion and context. Don't read it if you are
looking for a quick fix.
The topic of locking CVS files resurfaces on the network every so
often, producing the same results each time:
The Big Endians:
CVS was designed to avoid locks, using a copy-modify-merge model.
Locking is not necessary and you should take the time to learn the CVS
model which many people find workable. So why not get with the program
and learn how to think the CVS way?
The Little Endians:
The users determine how a tool is to be used, not the designers. We,
the users, have always used locking, our bosses demand locking,
locking is good, locking is God. I don't want to hear any more
lectures on the CVS model. Make locking work.
Any organization making active changes to a source base will
eventually face the need to do parallel development. Parallel
development implies merges. (If you plan to keep separate copies of
everything and never merge, good luck. Tell me who you work for so I
can buy stock in your disk suppliers this year and sell your stock
short next year.)
Merges will never go away. CVS chose to make "merges" stand front and
center as an important, common occurrence in development. It is one
way of looking at things.
For free-format text, the merge paradigm gives you a considerable
amount of freedom. It does take a bit of management, but any project
should be ready to deal with it.
On the other hand, there are many files that can't be merged using
text merge techniques. Straight text merge programs like "diff3" are
guaranteed to fail on executables (with relative branch statements),
files with self-referential counts stored in the file (such as TAGS
files), or files with relative motion statements in them (such as
Frame MIF files, many postscript files). They aren't all binary files.
For these types of files, and many others, there are only two
solutions:
Complex merge tools that are intimately aware of the contents of the
files to be merged. (ClearCase, and probably others, allow you to
define your own "files types" with associated "merge tools".)
Serialization of access to the file. The only technical solution to
the problem of serialization is "locking".
Since you can call a program that offers:
"Which one do you want? A/B?"
a "merge tool", more and more merge tools will appear which can be
hooked into a merge-intensive program like CVS. Think of a bitmap
"merge" tool that displays the bitmaps on the screen and offers a
"paint" interface to allow you to cut and paste, overlay, invert or
fuse the two images such that the result is a "merged" file.
My conclusion is that the need for locking is temporary, awaiting
better technology. For large development groups, locking is not an
alternative to merging for text files.
Last modified: _6/13/1997_
9. What is "checkout -s"? How is it different from "checkout -c"?
The '-c' and '-s' options to "checkout" both cause the modules file to
appear on standard output, but formatted differently.
"checkout -c" lists the modules file alphabetized by the module name.
It also prints all data (including options like '-a' and "-o <prog>")
specified in the modules file.
"checkout -s" lists the modules file sorted by "status" field, then by
module name. The status field was intended to allow you to mark
modules with strings of your choice to get a quick sorted report based
on the data you chose to put in the status fields. I have used it for
priority ("Showstopper", etc as tied into a bug database), for porting
status ("Ported", "Compiled", etc. when porting a large collection of
modules), for "assignee" (the person responsible for maintenance), and
for "test suite" (which automatic test procedure to run for a
particular module).
Last modified: _6/13/1997_
Category: /Commands_/commit_ci_com/
" + "commit", "ci", "com""
1. What is "commit" for?
To store new revisions in the Repository, making them visible to other
users.
Last modified: _6/13/1997_
2. If I edit ten files, do I have to type "commit" ten times?
No. The "commit" command will take multiple filenames, directory names
and relative pathnames on the command line and commit them all with
the same log message. If a file is unchanged, even if it is explicitly
listed on the command line, CVS will skip it.
Like all CVS commands, "commit" will work on the whole directory by
default. Just type "cvs commit" to tell CVS to commit all modified
files (i.e. the files that "update" would display preceded by 'M') in
the current directory and in all sub-directories.
Last modified: _6/13/1997_
3. Explain: cvs commit: Up-to-date check failed for `<file>'
You may not "commit" a file if your BASE revision (i.e. the revision
you last checked out, committed or retrieved via "update") doesn't
match the HEAD revision (i.e the latest revision on your branch,
usually the Main Branch).
In other words, someone committed a revision since you last executed
"checkout", "update" or "commit". You must now execute "update" to
merge the other person's changes into your working file before
"commit" will work. You are thus protected (somewhat) from a common
form of race condition in source control systems, where a checkin of a
minor alteration of a second copy of the same base file obliterates
the changes made in the first.
Normally, the "update" command's auto-merge should be followed by
another round of building and testing before the "commit".
Last modified: _6/13/1997_
4. What happens if two people try to "commit" conflicting changes?
Conflicts can occur only when two developers check out the same
revision of the same file and make changes. The first developer to
commit the file has no chance of seeing the conflict. Only the second
developer runs into it, usually when faced with the "Up-to-date" error
explained in the previous question.
There are two types of conflicts:
When two developers make changes to the same section of code, the
auto-merge caused by "update" will print a 'C' on your terminal and
leave "overlap" markers in the file.
You are expected to examine and clean them up before committing the
file. (That may be obvious to *some* of you, but . . .)
A more difficult problem arises when two developers change different
sections of code, but make calls to, or somehow depend on, the old
version of each other's code.
The auto-merge does the "right" thing, if you view the file as a
series of text lines. But as a program, the two developers have
created a problem for themselves.
This is no different from making cross-referential changes in
*separate* files. CVS can't help you. In a perfect world, you would
each refer to the specification and resolve it independently. In the
real world you have to talk/argue, read code, test and debug until the
combined changes work again.
Welcome to the world of parallel development.
Last modified: _6/13/1997_
5. I committed something and I don't like it. How do I remove it?
Though you *can* use the "admin -o" (synonym: "rcs -o") command to
delete revisions, unless the file you committed is so embarrassing
that the need to eradicate it overrides the need to be careful, you
should just grab an old version of the file ("update -p -r
<previous-rev>" might help here) and commit it on top of the offending
revision.
See Section 3B on "admin".
Last modified: _6/13/1997_
6. Explain: cvs commit: sticky tag `V3' for file `X' is not a branch
The message implies two things:
You created your working directory by using "checkout -r V3", or you
recently executed "update -r V3".
The tag named V3 is not a branch tag.
CVS records (i.e. makes "sticky") any "-r <tag/rev>" argument handed
to the "checkout" or "update" commands. The <tag/rev> is recorded as
the CVS working branch, which is the branch to which "commit" will add
a new revision.
Branch tags are created when you use the -b switch on the "tag" or
"rtag" commands. Branch tags are magic tags that don't create a
physical branch, but merely mark the revision to branch from when the
branch is needed. The first commit to a magic branch creates a
physical branch in the RCS files.
You can commit onto the end of the Main Trunk, if you have no sticky
tag at all, or onto the end of a branch, if you have a sticky branch
tag. But you can't commit a file that has a sticky tag not pointing to
a branch. CVS assumes a sticky Tag or Revision that does not refer to
a branch is attached to the middle of a series of revisions. You can't
squeeze a new revision between two others. Sticky dates also block
commits since they never refer to a branch.
Scenario1:
If you don't want a branch and were just looking at an old revision,
then you can move back to the Main Branch by typing:
cvs update -A {files or dirs, default is '.'}
or you can move to the branch named <branch_tag> by:
cvs update -r <branch_tag> {files or dirs, default is '.'}
Scenario2:
If you really wanted to be on a branch and made an earlier mistake by
tagging your branch point with a non-branch tag, you can recover by
adding a new branch tag to the old non-branch tag:
cvs rtag -b -r <oldtag> <newtag> <module>
(It was not a big mistake. Branch-point tags can be useful. But the
<newtag> must have a different name.)
If you don't know the <module> name or don't use "modules", you can
also use "tag" this way:
cvs update -r <oldtag>
cvs tag -b <newtag> .
Then, to put your working directory onto the branch, you type:
cvs update -r <newtag>
You can't delete <oldtag> before adding <newtag>, and I would not
advise deleting the <oldtag> at all, because it is useful in referring
to the branch point. If you must, you can delete the non-branch tag
by:
cvs rtag -d <oldtag> <module>
or
cvs tag -d <oldtag> .
Scenario3:
If you made the same mistake as in Scenario2 (of placing a non-branch
tag where you wanted a branch tag), but really want <oldtag> to be the
name of your branch, you can execute a slightly different series of
commands to rename it and move your working directory onto the branch.
Warning: This is not a way to rename a branch tag. It is a way to turn
a non-branch tag into a branch tag with the same name.
cvs rtag -r <oldtag> <branch_point_tag> <module>
cvs rtag -d <oldtag> <module>
cvs rtag -b -r <branch_point_tag> <oldtag> <module>
Then, if you really must, delete the <branch_point_tag>:
cvs rtag -d <branch_point_tag> <module>
Note: The unwieldy mixture of "tag" and "rtag" is mostly because you
can't specify a revision (-r <tag>) to the "tag" command.
See 4C.3 for more info on creating a branch.
Last modified: _6/13/1997_
7. Why does "commit -r <tag/rev>" put newly added files in the Attic?
If you specify "-r <rev>" (where <rev> is a dotted numeric number like
2.4), it correctly sets the initial revision to <rev>, but it also
attaches the numeric <rev> as a sticky tag and throws the file into
the Attic. This is a bug. The obvious solution is to move the file out
of the Attic into the associated Repository directory and "update -A"
the file. There are no Tags to clean up.
If you specify "-r <tag>" to commit a newly added file, the <tag> is
treated like a <branch_tag>, which becomes a symbolic RCS label
pointing to the string '1', which can be considered to be the "Main
branch number" when the main branch is still at revision 1.N. The file
is also thrown into the Attic. See 4C.8 for a way to recover from
this.
In fact, a plain "commit" without the "-r" will throw a newly added
file into the Attic if you added it to a directory checked out on a
branch. See 3A.[2-5].
See Section 4C, on Branching, for many more details.
Last modified: _6/13/1997_
8. Why would a "commit" of a newly added file not produce rev 1.1?
When committing a newly added file CVS looks for the highest main
branch major number in all files in the ./CVS/Entries file. Normally
it is '1', but if you have a file of revision 3.27 in your directory,
CVS will find the '3' and create revision 3.1 for the first rev of
<file>. Normally, the first revision is 1.1.
Last modified: _6/13/1997_
Category: /Commands_/diff_di_dif/
" + "diff", "di", "dif""
1. What is "diff" for?
To display the difference between a working file and its BASE
revision (the revision last checked out, updated or committed):
cvs diff <file>
To display the difference between a working file and a committed
revision of the same file:
cvs diff -r <tag/rev> <file>
To display the difference between two committed revisions of the
same file:
cvs diff -r <tag1/rev1> -r <tag2/rev2> <file>
You can specify any number of <file> arguments. Without any <file>
arguments, it compares the whole directory.
In the examples above, "-D <date>" may be substituted wherever "-r
<tag/rev>" appears. The revision a <date> refers to is the revision
that existed on that date.
Last modified: _6/13/1997_
2. Why did "diff" display nothing when I know there are later committed
revisions in the Repository?
By default, "diff" displays the difference between your working file
and the BASE revision. If you haven't made any changes to the file
since your last "checkout", "update" or "commit" there is no
difference to display.
To display the difference between your working file and the latest
revision committed to your current branch, type:
cvs diff -r HEAD <file>
Last modified: _6/13/1997_
3. How do I display what changed in the Repository since I last executed
"checkout", "update" or "commit"?
A special tag (interpreted by CVS -- it does not appear in the Tag
list) named "BASE" always refers to the revision you last checked out,
updated or committed. Another special tag named "HEAD" always refers
to the latest revision on your working branch.
To compare BASE and HEAD, you type:
cvs diff -r BASE -r HEAD <file>
Last modified: _6/13/1997_
4. How do I display the difference between my working file and what I
checked in last Thursday?
cvs diff -D "last Thursday" <file>
where "last Thursday" is a date string. To be more precise, the
argument to the '-D' option is a timestamp. Many formats are accepted.
See the man page under "-D date_spec" for details.
Last modified: _6/13/1997_
5. Why can't I pass long options, like --unified, to "diff"?
CVS only handles single character '-X' arguments, not the FSF long
options. CVS also passes through only arguments it knows about,
because a few arguments are captured and interpreted by CVS.
If you didn't configure RCS and CVS to use the GNU version of diff,
long options wouldn't work even if future versions of CVS acquire the
ability to pass them through.
Most of the long options have equivalent single-character options,
which do work. The "--unified" option is equivalent to '-u' in
revisions of GNU diff since 1.15.
Last modified: _6/13/1997_
Category: /Commands_/export_exp_ex/
" + "export", "exp", "ex""
1. What is "export" for?
"export" checks out a copy of a module in a form intended for export
outside the CVS environment. The "export" command produces the same
directory and file structure as the "checkout" command, but it doesn't
create "CVS" sub-directories and it removes all the RCS keywords from
the files.
Last modified: _6/13/1997_
2. Why does it remove the RCS keywords so I can't use the "ident" command
on the source files?
It removes the RCS keywords, so that if the recipient of the exported
sources checks them into another set of RCS files (with or without
CVS), and then makes modifications through RCS or CVS commands, the
revision numbers that they had when you exported them will be
preserved. (That ident no longer works is just an unfortunate side
effect.)
The theory is that you are exporting the sources to someone else who
will make independent changes, and at some point you or they will want
to know what revisions from your Repository they started with
(probably to merge changes, or to try to decide whether to merge
changes).
A better way to handle this situation would be to give them their own
branch of your Repository. They would need to remember to checkin the
exported sources with RCS IDs intact (ci -k) so that their changes
would get revision numbers from the branch, rather than starting at
1.1 again. Perhaps a future version of CVS will provide a way to
export sources this way.
Contributed by Dan Franklin
Last modified: _6/13/1997_
3. Can I override the '-kv' flag CVS passes to RCS?
Not as of CVS version 1.4.
Last modified: _6/13/1997_
4. Why doesn't "export" have a '-k' flag like "import" does?
Export is intended for a specific purpose -- to remove all trace of
revision control on the way *out* of CVS.
Last modified: _6/13/1997_
5. Why does "export -D" check out every file in the Attic?
See 5B.3 for an explanation of the same problem with "update".
Last modified: _6/13/1997_
Category: /Commands_/history_hi_his/
" + "history", "hi", "his""
1. What is "history" for?
To provide information difficult or impossible to extract out of the
RCS files, such as a "tag" history or a summary of module activities.
Last modified: _6/13/1997_
2. Of what use is it?
I have found it useful in a number of ways, including:
Providing a list of files changed since
- A tagged release.
- Yesterday, last Thursday, or a specific date.
- Someone changed a specific file.
Providing a list of special events:
- Files added or removed since one of the above events.
- Merge failures since one of the above events. (Where did the
conflicts occur?)
- Has anyone (and who) grabbed the revision of this file I committed
last week, or are they still working blind?
Telling me how often a file/directory/module has been changed.
Dumping a summary of work done on a particular module, including who
last worked on it and what changed.
Displaying the checked-out modules and where they are being worked
on.
To tell me what users "joe" and "malcolm" have done this week.
Last modified: _6/13/1997_
3. What is this, Big Brother?
War is Peace.
Freedom is Slavery.
Ignorance is Strength.
Normally manager types and those with the power to play Big Brother
don't care about this information. The Software Engineer responsible
for integration usually wants to know who is working on what and what
changed. Use your imagination.
Last modified: _6/13/1997_
4. I deleted my working directory and "history" still says I have it
checked out. How do I fix it?
You can use "release -f" to forcibly add a "release" record to the
history file for a working directory associated with a "module". If
your version of "release" doesn't have the '-f' option, or you checked
out the directory using a relative path, you have to edit the
$CVSROOT/CVSROOT/history file.
You can remove the last 'O' line in the history file referring to the
module in question or add an 'F' record.
Last modified: _6/13/1997_
5. So I *can* edit the History file?
Yes, but if you are using history at all, you should take a little
care not to lose information. I normally use Emacs on the file, since
it can detect that a file has changed out from under it. You could
also copy and zero out the history file, edit the copy and append any
new records to the edited copy before replacing it.
Last modified: _6/13/1997_
6. Why does the history file grow so quickly?
It stores 'U' records, which come in handy sometimes when you are
tracking whether people have updated each other's code before testing.
There should (and probably will sometime) be a way to choose what
kinds of events go into the history file.
The contributed "cln_hist.pl" script will remove all the 'U' records,
plus matching pairs of 'O' and 'F' records during your normal clean up
of the history file.
Last modified: _6/13/1997_
7. What is the difference between "cvs history -r <tag/rev>" and "cvs
history -t <tag>"?
The '-t' option looks for a Tag record stored by "rtag" in the history
file and limits the search to dates after the last <tag> of the given
name was added.
The '-r' option was intended to search all files looking for the <tag>
in the RCS files. It takes forever and needs to be rewritten.
Last modified: _6/13/1997_
8. Why does "cvs history -c -t <tag>" fail to print anything?
You have been using "tag" instead of "rtag". The "tag" command
currently doesn't store a history record. This is another remnant of
CVS's earlier firm belief in "modules". But it also has a basis in how
"rtag" and "tag" were originally used.
"rtag" was intended for large-scale tagging of large chunks of the
Repository, an event work recording. "tag" was intended for adding and
updating tags on a few files or directories, though it could also be
used to tag the entire checked-out working tree when there is no
module defined to match the tree or when the working tree is the only
place where the right collection of revisions to tag can be found.
Last modified: _6/13/1997_
9. "cvs history -a -o" only printed one line for each checked-out module.
Shouldn't it print all the directories where the modules are checked out?
Not as designed.
Command Question it is supposed to answer.
---------------- ------------------------------------------
cvs history -o What modules do I have checked out?
cvs history -a -o <same for all users>
cvs history -o -w What working directories have I created
and what modules are in them?
cvs history -a -o -w <same for every user>
The -o option chooses the "checked out modules" report, which is the
default history report.
Last modified: _6/13/1997_
10. I can't figure out "history", can you give me concrete examples?
Default output selects records only for the user who executes the
"history" command. To see records for other users, add one or more "-u
user" options or the '-a' option to select *all* users.
To list (for the selected users): Type "cvs history" and:
* Checked out modules: -o (the default)
* Files added since creation: -x A
* Modified files since creation: -c
* Modified files since last Friday: -c -D 'last Friday'
* Modified files since TAG was added: -c -t <tag>
* Modified files since TAG on files: -c -r <tag>
* Last modifier of file/Repository X? -c -l -[fp] X
* Modified files since string "str": -c -b str
* Tag history: (Actually "rtag".) -T
* History of file/Repository/module X: -[fpn] X
* Module report on "module": -m module
Last modified: _6/13/1997_
11. Can we merge history files when we merge Repositories?
Assuming that the two Repositories have different sets of pathnames,
it should be possible to merge two history files by sorting them
together by the timestamp fields.
You should be able to run:
sort +0.1 ${dir1}/history ${dir2}/history > history
If you "diff" a standard history file before and after such a sort,
you might see other differences caused by garbage (split lines, nulls,
etc) in the file. If your Repository is mounted through NFS onto
multiple machines you will also see a few differences caused by
different clocks on different machines. (Especially if you don't use
NTP to keep the clocks in sync.)
Last modified: _6/13/1997_
Category: /Commands_/import_im_imp/
" + "import", "im", "imp""
1. What is "import" for?
The "import" command is a fast way to insert a whole tree of files
into CVS.
The first "import" to a particular file within the Repository creates
an RCS file with a single revision on the "Vendor branch." Subsequent
"import"s of the same file within the Repository append a new revision
onto the Vendor branch. It does not, as some seem to believe, create a
new branch for each "import". All "imports" are appended to the single
Vendor branch.
If the file hasn't changed, no new revision is created -- the new
"Release-Tag" is added to the previous revision.
After the import is finished, files you have not changed locally are
considered to have changed in the "Main line of development". Files
you *have* changed locally must have the new Vendor code merged into
them before they are visible on the "Main line".
See 4C.6 and 4C.15
Last modified: _6/13/1997_
2. How am I supposed to use "import"?
Create a source directory containing only the files you want to
import. Make sure you clean up any cruft left over from previous
builds or editing. You want to make sure that the directory contains
only what you want to call "source" from which everything else is
built.
If this is not the first import from this "Vendor", you should also
compare the output of "find . ! -name CVS -print | sort" executed both
at the head of a checked out working directory and at the head of the
sources to be imported. If you find any deleted or renamed files, you
have to deal with them by hand. (See 4B.8 on renaming.)
"cd" into your source directory and type:
cvs import -m "Message" <repos> <Vendor-Tag> <Release-Tag>
where <repos> is the relative directory pathname within the Repository
that corresponds to the sources you are importing.
You might also consider using the "-I !" option to avoid ignoring
anything. It is easier to remove bogus files from the Repository than
to create a sparse tree of the ignored files and rerun "import".
For example, if the FSF, CVS, Make and I are still active in the year
2015, I'll import version 89.53 of GNU make this way:
cvs import -m "GNUmake V89.53" gnu/make GNU GNUMAKE_89_53
See 3H.13 for more details.
Last modified: _6/13/1997_
3. Why does import put files on a branch? Why can't I work on the main
trunk instead of a Vendor branch?
This was a Design choice. The Vendor branch is the way "import" deals
with a Vendor release. It is a solution to the Engineering problem of
how to merge multiple external releases of Vendor-supplied sources
into your ongoing work. The Vendor releases are kept on a separate,
special, "Vendor" branch and your work is kept on the RCS trunk. New
Vendor releases are imported onto the Vendor branch and then merged
into your work, if there is any, on the trunk.
This way, you can use CVS to find out not only about your work, but
you can also find out what the Vendor changed by diffing between two
of the Release Tags you handed to "import".
CVS was designed to work this way. If you use CVS in some other way,
you should think carefully about what you are doing.
Note that the CVS "Main Branch" and the RCS Main Trunk are not the
same. Placing files on the Vendor Branch doesn't keep you from
creating a development branch to work on.
See Section 4C, on Branching.
If you are not working with 3rd party (i.e. Vendor) sources, you can
skip the "import" and avoid the Vendor branch entirely. It works just
as well to move pre-existing RCS files into Repository directories.
You can create a whole Repository tree by copying a directory
hierarchy of normal source files directly into the Repository and
applying CVS to it. Here's an idea you should *test* before using:
cd <your source tree>
set source = `pwd`
set module = xyzzy <<== Your choice of directory name
mkdir $CVSROOT/$module
cd $CVSROOT/$module
(cd $source; tar cf - .) | tar xvpBf -
find . -type f -exec ci -t-Original. {} \;
The RCS "ci" command, without -u or -l options, will turn your source
file into an RCS (",v") and delete the original source.
Last modified: _6/13/1997_
4. Is there any way to import binary files?
If you configured CVS to use the GNU version of "diff" and "diff3",
then you can import any kind of file.
Binary files with RCS keywords in them are a problem, since you don't
want them to expand.
If the tree you are about to "import" is entirely filled with binary
files, you can use the '-ko' option on "import". Otherwise, I would
run the import normally, then fix the binary files as described below
in 3H.5.
See 4D.1 on Binary files.
Last modified: _6/13/1997_
5. Why does "import" corrupt some binary files?
The RCS "co" command, when it is invoked by a CVS "checkout" or
"update" (or after a "commit") command, searches for and expands a
list of keywords within the file. They are documented in the RCS "co"
man page. Strings such as "$\Id$" (or "$\Id:"), or "$\Revision$" (or
"$\Revision:") are altered to the include the indicated information.
[[Note: The keywords should appear in the text without the '\'
character I have inserted to *avoid* expansion here. The only real RCS
keywords in this document are at the top of the file, where I store
the Revision and Date.]]
If RCS keyword strings show up in a binary file, they will be altered
unless you set the '-ko' option on the RCS files to tell RCS to keep
the original keyword values and not to expand new ones. After
"import", you can set the '-ko' option this way:
cvs admin -ko <file>
rm <file>
cvs update <file>
After an import that didn't use '-ko' (because the whole tree wasn't
of binary files) you should fix up the binary files as described above
before checking out any new copies of the files and before updating
any working directories you checked out earlier.
See 4D.1 on Binary files.
Last modified: _6/13/1997_
6. How do I retain the original $\Revision$ strings in the sources?
If you want to leave old RCS keywords as they are, you can use the
'-ko' tricks described above.
Last modified: _6/13/1997_
7. I imported some files for the Yarg compiler that compiles files with a
suffix of ".yarg" and whose comment prefix is "YARG> ". When I check them
out, they will no longer compile because they have this junk in them. Why?
YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>
YARG> $\Log:
# Revision 1.3 1998/03/03 00:16:16 bubba
# What is 2+2 anyway?
#
# Revision 1.2 1998/03/03 00:15:15 bubba
# Added scorekeeping.
YARG>
YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>YARG>
Well bubba, "Yarg" hasn't hit the big time yet. Neither RCS nor CVS
know about your suffix or your comment prefix. So you have two
choices:
Check out the Yarg-less module, and tell all the files about your
comment prefix. Visit each directory and type:
cvs admin -c"YARG> " *.yarg
If *all* files in the whole directory tree are Yarg files, you can use
this instead:
cvs admin -c"YARG> " .
Then save any changes you made, remove all the "*.yarg" files and grab
new copies from the Repository:
rm *.yarg (or: find . -name '*.yarg' -exec rm {} ';') (or: find .
-name '*.yarg' -print | xargs rm) (or: find . -name '*.yarg' -print0 |
xargs -0 rm if you have spaces in filenames and the GNU find/xargs.)
cvs update
It might be faster to remove the whole directory and check it out
again.
Change the import.c file in the CVS sources and add the .yarg
suffix, along with the "YARG> " comment prefix to the "comtable"
array.
If you ever plan to add new files with $\Log in them, you should also
go into the RCS sources and make the same change in the table
contained in the "rcsfnms.c" file.
Then delete the imported files from the Repository and re-"import" the
sources.
Last modified: _6/13/1997_
8. How do I make "import" save the timestamps on the original files?
Use "import -d" to save the current timestamps on the files as the RCS
revision times.
See 4D.8 for another aspect of file timestamps.
Last modified: _6/13/1997_
9. Why can't I "import" 3 releases on different branches?
I'll bet you typed something like this:
cd /src/blasto.v2
cvs import -b 1.1.2 VENDOR2 Version2
cd /src/blasto.v3
cvs import -b 1.1.3 VENDOR3 Version3
cd /src/blasto.v4
cvs import -b 1.1.4 VENDOR4 Version4
This is wrong, or at least it won't help you much. You have created
three separate Vendor branches, which is probably not what you wanted.
Earlier versions of CVS, as described in Brian Berliner's Usenix
paper, tried to support multiple Vendor branches on the theory that
you might receive source for the *same* program from multiple vendors.
It turns out that this is very rare, whereas the need to branch in
*your* development, for releases and for project branches, is much
greater.
So the model now is to use a single vendor branch to contain a series
of releases from the same vendor. Your work moves along on the Main
Trunk, or on a CVS branch to support a real "branch in development".
To set this up, you should type this instead of the above:
cd /src/blasto.v2
cvs import VENDOR Version2
cd /src/blasto.v3
cvs import VENDOR Version3
cd /src/blasto.v4
cvs import VENDOR Version4
Last modified: _6/13/1997_
10. What do I do if the Vendor adds or deletes files between releases?
Added files show up with no extra effort. To handle "removed" files,
you should always compare the tree structure of the new release
against the one you have in your Repository. If the Vendor has removed
files since the previous release, go into a working directory
containing your current version of the sources and "cvs remove"
(followed by "cvs commit" to make it really take effect) each file
that is no longer in the latest release.
Using this scheme will allow you to "checkout" any version of the
vendor's code, with the correct revisions and files, by using
"checkout -r Version[234]".
Renames are harder to find, since you have to compare file contents to
determine that one has occurred. If you notice one, see 4B.8 on
renaming files.
Last modified: _6/13/1997_
11. What about if the Vendor changes the names of files or directories, or
rearranges the whole structure between releases?
Currently CVS can't handle this cleanly. It requires "renaming" a
bunch of files or directories.
See 4B.8 on "renaming" for more details.
What I generally do is to close the Repository for a while and make
changes in both the Repository and in a copy of the vendor release
until the structure matches, then execute the import.
If you ever have to check out and build an old version, you may have
to use the new, or completely different Makefiles.
Last modified: _6/13/1997_
12. I thought "import" was for Vendor releases, why would I use it for code
of my own? Do I have to use import?
For code you produce yourself, "import" is a convenience for fast
insertion of whole trees. It is not necessary. You can just as easily
create ",v" files using the RCS "ci" command and move them directly
into the Repository.
Other than the CVSROOT directory, the Repository consists entirely of
directories of ",v" files. The Repository contains no other state
information.
See Section 4B, on Setting up and Managing the Repository.
Last modified: _6/13/1997_
13. How do I import a large Vendor release?
When the sum of the changes made by the Vendor and the changes made by
local developers is small, "import" is not a big problem. But when you
are managing a large Repository, any care taken up front will save you
time later.
First read the following, then, before executing "import", see the
questions in Section 4C dealing with branch merges and Vendor branch
merges.
If this is not the first import of this code, before starting, rtag
the whole directory you will be changing.
The first step is to make sure the structure of the new files
matches the structure of the current Repository.
Run "find . -print | sort" on both trees and "diff" the output.
Alter the "source" tree until the "diff" (of the list of filenames,
not of the whole trees) shows that the directory structures are
equivalent.
The "comm" command, if you have it, can help figure out what has been
added or deleted between releases.
If they deleted any files, you can handle them cleanly with "cvs
remove". The command "comm -23 files.old files.new" will show you a
list of files that need to be removed.
You should examine the list first to see if any have been renamed
rather than simply deleted.
If they renamed any files, see 4B.8 on renaming files.
Remember to *SAVE* the output from the import command.
When you have dealt with removed and renamed files, then you can
execute the import:
cd <new source>
cvs import -I ! -m "Message" <repos> <VendorTag> <ReleaseTag>
Where
"-I !" is an optional argument that keeps "import" from ignoring
files. The comparison of the "find" commands above will probably avoid
the need for this, but it is easier to remove files from the
Repository than to run a subset "import" to catch just the ignored
files. [You might have to quote or backwhack the '!'.]
Message is the log message to be stored in the RCS files.
<repos> is a relative path to a directory within the
Repository. The directory <new source> must be at
the same relative level within the new sources as
the <repos> you give is within the Repository. (I
realize this is not obvious. Experiment first.)
<VendorTag> is a Tag used to identify the Vendor who sent you
the files you are importing. All "imports" into
the same <repos> *must* use the same VendorTag.
You can find it later by using the "log" command.
<ReleaseTag> is a Tag used to identify the particular release of the
software you are importing. It must be unique and should be mnemonic
-- at least include the revision number in it. (Note: you can't use
'.' characters in a Tag. Substitute '_' or '-'.)
There will be six categories of files to deal with. (Actually there
are eight, but you have already dealt with "removed" and "renamed"
files.)
If this is the first "import" into a given <repos> directory, only the
first three of these ('I', 'L' and 'N') can occur.
Ignored file.
CVS prints: I filename
You'll need to examine it to see if it *should* have been ignored. If
you use "-I !", nothing will be ignored.
Symbolic link.
CVS prints: L linkname
Links are "ignored", but you'll probably want to create a "checkout
helper" function to regenerate them.
New file.
CVS prints: N filename
CVS creates a new file in the Repository. You don't have to do
anything to the file, but you might have to change Makefiles to refer
to it if this is really a new file.
A file unchanged by the Vendor since its last release.
CVS prints: U filename
CVS will notice this and simply add the new ReleaseTag to the latest
rev on the Vendor branch.
No work will be needed by you, whether you have changed the file or
not. No one will notice anything.
A file changed by the Vendor, but not by you.
CVS prints: U filename
CVS should add the file onto the vendor branch and attach the Release
Tag to it.
When you next execute "update" in any working directory you'll get the
new revision.
A file changed by both the Vendor and by you.
CVS prints: C filename
These are the trouble files. For each of these files (or in groups --
I usually do one directory at a time), you must execute:
cvs update -j <PreviousReleaseTag> -j <ReleaseTag>
or
cvs update -j <VendorTag:yesterday> -j <VendorTag>
It will print either 'M' (if no overlaps) or 'C', if overlaps. If a
'C' shows up, you'll need to edit the file by hand.
Then, for every file, you'll need to execute "cvs commit".
See the part of Section 4C dealing with branch merges.
If you are truly performing a large import, you will most likely
need help. Managing those people is another problem area.
Since the merge of the Vendor branch is just like any other merge, you
should read section 4C for more info about performing and cleaning up
merges.
The larger the import, and the larger the group of people involved,
the more often you should use "tag" and "rtag" to record even trivial
milestones. See 4C.14, especially the "paranoid" section.
Before starting the import, you should install and test a "commitinfo"
procedure to record all commits in a file or via Email to a mail
archive. Along with the tags you placed on the Repository before the
import, this archive will help to track what was changed, if problems
occur
There are four stages to the recovery:
Parcel out the work -- Effective Emacs Engineering.
As input to the assignment process, you might want to examine the tree
and record the last person who changed the file. You can also
research, if you don't already know, who is expert in each area of the
software.
Examine the import log (you saved the output, right?), estimate how
much work is involved in each area and assign groups of files to
individual developers. Unless some directory is immense, it is easier
to manage if you assign whole directories to one person.
Keep a list. Suggest a completion date/time. Tell them to "commit" the
file when they are finished with the merge. If you tagged the
Repository before starting the import, you should have no trouble
figuring out what happened.
If you can, find out (or tell them) which working directory to use.
You should verify that the working directory they use is on the Main
Branch ("update -A") and without modified files.
If you trust your crew, have them notify you by Email. Have them send
you the output from "cvs update" in their working directory. You might
have to poll some people until you are certain they have finished, or
have given up. (This is not an invention. I've heard a false, "Yeah,
sure. I finished yesterday," more times that you'd believe.)
When all reports are in, go on to the Source Verification stage.
Source Verification -- CVS and other Tools.
If you didn't dictate which ones to use, find all working directories
and run "cvs -n update" in all of them. The history command and the
"commitinfo" log you set up might help to find checked out working
directories.
Sticky conflict flags will help, but they can't recover from
sloppiness or incompetence. You might want to check everything out
into a tree and grep for the parts of the merge conflict markers CVS
doesn't look for. CVS looks for the string '^>>>>>>> '. The merge
operation also puts '^<<<<<<< ' and '^======= ' markers in the file
that careless developers might leave there.
If you find problems simply by looking at the source files and working
directories, start the flogging now. Resolving the textual conflicts
is the easy part. Weed the turkeys out before reaching the next part
of the cleanup -- the resolution of logical conflicts.
Then apply a set of post-commit tags.
Logical Verification -- Diff and powerful eyeballs.
No source control system can solve the problem of resolving
distributed conflicts in program logic. If you change the argument
template for function A (defined in file A.c) and add new calls to
function A from within function B (defined in file B.c) using the old
argument format, you are outside the realm of CVS's competence.
Assign someone to understand what the Vendor changed by running "cvs
diff -c -r <PreviousReleaseTag> <ReleaseTag>", where the tags were
those handed to the last two invocations of "import".
Then have the same person compare that output (logically or you can
actually diff the diffs) to the output of the similar "cvs diff -c -r
<pre-import-tag> <post-commit-tag>". The two sets of differences
should be almost identical. They should both show only the work *you*
have performed.
Product Verification -- Build and Test.
Don't let your help off the hook until you verify that the merge
actually produced something that can compile and pass tests. Compiling
should really be part of the logical verification phase, but you
should test the output of the build system before declaring victory
and releasing the troops.
After it is all built, apply another set of tags to mark the end of
the "import process". You can delete the intermediate tags you added
during source and logic testing, but keep the "pre-import" and
"post-import" tags forever.
Of course, experience can tell you when to skip a step. But I'd start
out by considering each one as necessary unless you can prove
otherwise.
Last modified: _6/13/1997_
14. Explain: ERROR: cannot create link to <file>: Permission denied
This error appears when you try to execute a second (or later)
"import" into the same module from a directory to which you don't have
write access.
The "link error" is caused by a feature purposely added to speed up
the import.
Though the error message is somewhat strange, it indicates that
"import" is supposed to be executed only in writable directories.
Last modified: _6/13/1997_
15. Where does the -m <message> go when the file doesn't change?
The <message> handed to import is used as an RCS log message, but only
if the imported file changed since the last version on the Vendor
branch. If the imported file hasn't changed, then no new revision is
created. The <ReleaseTag> is still applied, but to the previous
revision. So the Tags are still correct, but the message is lost.
Maybe it should be appended to the previous log message. But currently
it isn't.
Last modified: _6/13/1997_
16. How do I "import" just the files ignored by a previous "import"?
A real answer follows, but first, an editorial:
I am now convinced that you should always use the "-I !" option.
Removing a few extraneous files from the Repository is a lot easier
than the recovery step described below.
Let's assume your original import procedure was: (We assume there is
enough disk space in /tmp.)
cd <head-of-vendor-tree>
cvs import -m 'xyz 1.3' gnu/xyz GNU GNUXYZ_1_3 | tee /tmp/IMP
To import just the files ignored by "import", I would do this:
Create a list of the ignored files to import:
cd <head-of-vendor-tree> awk '/^I / {print $2}' /tmp/IMP | sed
's|^gnu/xyz/||' > /tmp/IG [Edit the IG file to contain just the files
you want.]
Then create a sparse directory by handing your list to the GNU
version of "tar", installed in many places as "gtar":
mkdir /tmp/FIXUP gtar -T /tmp/IG -c -f - . | (cd /tmp/FIXUP; gtar xvBf
-)
Then rerun the import. Use the exact same command, but execute it in
the sparse directory tree you just created. And this time, tell it not
to ignore anything.
cd /tmp/FIXUP
cvs import -I ! -m 'xyz 1.3' gnu/xyz GNU GNUXYZ_1_3
Last modified: _6/13/1997_
17. Why did "import" ignore all the symlinks?
This is another design choice.
Like the Unix "tar" command, "import" could sprout an option to follow
symbolic links, but I don't think CVS will ever follow symbolic links
by default.
Two possible future enhancements have been seriously discussed:
Treat symbolic links as data in its parent directory (the way
ClearCase does) in some sort of per-directory control file.
Treat symbolic links as version-controlled elements themselves,
whose data is the value of readlink(2).
For now, they are simply ignored.
If you want to save and reconstruct symlinks, you might want to define
a "checkout" or "update" program in the modules file which could
consult a file kept under CVS in your working directory and make sure
the specified links are in place.
Last modified: _6/13/1997_
Category: /Commands_/log_lo_rlog/
" + "log", "lo", "rlog""
1. What is "log" for?
To provide an interface to the RCS "rlog" command, which displays
information about the underlying RCS files, including the revision
history and Tag (RCS calls it a "symbol") list.
Last modified: _6/13/1997_
2. How do I extract the log entries between two revisions?
If both <rev1> and <rev2> are on the same branch, you can get what you
are looking for with: (If they aren't on the same branch you'll either
get an error or a display of the whole change log.)
cvs log -r<rev1>:<rev2> <file>
If you want all the revisions on the branch from <rev1> to the end of
the branch <rev1> is on, you can use:
cvs log -r<rev1>: <file>
(If <rev1> is a numeric RCS symbol attached to a branch revision with
an even number of '.'s in it, you get the whole branch.)
If you want all the revisions on the branch from the beginning of the
branch <rev2> is on up to revision <rev2>, you can use:
cvs log -r:<rev2> <file>
Note: Depending on whether <rev1> and <rev2> are:
- numeric or symbolic
- in the file or not
- on the same branch or not
the RCS "rlog" (and therefore the "cvs log") command will
display some combination of:
- error messages
- (intuitively correct) partial log listings
- a display of the entire change log.
Last modified: _6/13/1997_
3. How do I extract the log entries on a whole branch?
cvs log -r<rev> <file>
where <rev> must be a branch revision (one with an even number of
dots) or a *non-branch* tag on a branch revision. Non-branch tags on a
branch revision are not normally attached by CVS, to add one you will
have to explicitly tag a physical branch number within each file.
Since these branch numbers are almost never the same in different
files, this command is not all that useful.
The intuitive command (at least from the CVS perspective):
cvs log -r<branch_tag> <file>
does not work.
Last modified: _6/13/1997_
4. How do I generate ChangeLogs from RCS logs?
A program called rcs2log is distributed as part of GNU Emacs 19. A
(possibly older) version of this program appears in the contrib
directory of the cvs source tree.
Last modified: _6/13/1997_
5. Why does "log" tell me a file was committed exactly 5 hours later
than I know it was?
I can tell by this question that you were working in a time zone that
is 5 hours behind GMT (e.g. the U.S. East Coast in winter).
RCS file dates are stored in GMT to allow users in different time
zones to agree on the meaning of a timestamp. At first glance this
doesn't seem necessary, but many companies use distributed file
systems, such as NFS or AFS, across multiple timezones.
Some standard form must be used. GMT, as the "grid origin", is an
obvious candidate. The only other reasonable choice is to put the
timezone information in all the time stamps, but that changes the RCS
file format incompatibly, a step which has been avoided in the last
few RCS releases.
Last modified: _6/13/1997_
Category: /Commands_/patch_pa_rdiff/
" + "patch", "pa", "rdiff""
1. What is "patch" for?
To produce a "diff" between tagged releases to be handed to the
"patch" command at other sites. This is the standard way that source
patches are distributed on the network.
Last modified: _6/13/1997_
2. Why does "patch" include files from the Attic when I use '-D'?
See the explanation of the same problem with "update -D" contained in
section 5B.
Last modified: _6/13/1997_
3. How do I make "patch" produce a patch for one or two files? It seems to
work only with modules.
Patch is intended for producing patches of whole modules between
releases to be distributed to remote sites. Instead of "patch", you
can use the "diff" command with the '-c' context option:
cvs diff -c -r <rev/tag> -r <rev/tag> <file1> . . .
The patch command will be able to merge such a "diff" into the remote
source files.
If you configured CVS to use a version of "diff" that supports the
'-u' option, you can produce a more compact "patch" in "unidiff"
format. The latest revisions of the patch command can parse and apply
patches in "unidiff" format.
Last modified: _6/13/1997_
Category: /Commands_/release_re_rel/
" + "release", "re", "rel""
1. What is "release" for?
To register that a module is no longer in use. It is intended to
reverse the effects of a "checkout" by adding a record to the history
file to balance the checkout record and by optionally allowing you to
delete the checked-out directory associated with the module name.
Last modified: _6/13/1997_
2. Why can't I reverse a "cvs checkout path/name/subdir" with a "cvs
release path/name/subdir" without an "unknown module name"?
A simplistic implementation. (I can say this -- I wrote it.)
The "release" function was written for CVS 1.2 under the assumption
that the "module name" is a first class, unavoidable interface to the
Repository, allowing no way to retrieve anything other than by module
name. Though it is easier to program that way, many users of CVS
believe the modules support to be too primitive to allow such a
limitation.
Since "release" was written, other parts of CVS broke that assumption.
It needs to be revised.
Last modified: _6/13/1997_
3. Why can't I "release" portions of a checked out directory? I should be
able to "release" any file or sub-directory within my working directory.
This isn't really a limitation in "release", per se. CVS doesn't try
to keep track of which files in which directories are "checked out"
and which are just lying there. You can delete directories and
"update" will not bring them back unless you add a special "-d"
option.
In other words, CVS doesn't keep track of how you adjust the partition
between files you consider part of your working set and files that
were checked out because they are part of the same module or
directory. And neither does "release".
In future CVS releases, "release" might become sophisticated enough to
handle both the reversal of a "checkout" and the deletion of random
portions of the working directory, but it isn't that way now.
Last modified: _6/13/1997_
4. I removed the tree that I was about to start working on. How do I tell
cvs that I want to release it if I don't have it anymore?
See 3G.4.
Last modified: _6/13/1997_
5. Why doesn't "release -d module" reverse a "checkout module"?
It does, if you are using "module" in a way that "release" expects: a
non-alias string in the left column of the "modules" database.
If "module" is really an alias, or if you are using a relative path in
the place of "module", or if you renamed the directory with the -d
option in the modules file or on the "checkout" command line, then the
current version of "release" won't work.
Future versions of "release" will probably fix most of these.
Last modified: _6/13/1997_
6. Why can't I release a module renamed with "cvs checkout -d"?
The current version of "release" doesn't know how to track the
renaming option ('-d') of the "checkout" command. It will probably be
fixed in the future.
Last modified: _6/13/1997_
Category: /Commands_/remove_rm_delete/
" + "remove", "rm", "delete""
1. What is "remove" for?
To remove a file from the working branch. It removes a file from the
main branch by placing it in an "Attic" directory.
Last modified: _6/13/1997_
2. Why doesn't "remove" work on directories when it appears to try?
Oversight. It should be able to delete an empty directory, but you
still don't have a way to remember when it was there and when it
disappeared to allow the "-D " option to work.
You'll have to remove the working directory and the matching directory
in the Repository.
Note that you want to do a _cvs remove dir_ in the working directory,
do a cvs commit, and then do a _rmdir dir_ in the Repository.
(msusrtsp.mark at eds dot com)
Last modified: _12/18/1997_
3. I don't like removing files. Is there another way to ignore them?
There's no reason to be hasty in using the "remove" command.
If there is a way to ignore files in your build procedures, I'd just
do that. Later, when you decide that the files are really ancient, you
can execute a "remove" command to clean up.
The CVS "ignore" concept can't ignore files already in CVS.
Last modified: _6/13/1997_
4. I just removed a file. How do I resurrect it?
If you executed "remove", but haven't typed "commit" (you can tell
this by the 'R' notation that "update" prints next to the file), you
can execute "add" to reverse the "remove".
If you followed the "remove" with a "commit", you'll have to move it
back out of the Attic by hand:
I use something like this: (csh-like syntax)
set repos = `cat ./CVS/Repository`
mv $repos/Attic/filename,v $repos/filename,v
(If you use relative paths in your Repository files, that first line
becomes: set repos = $CVSROOT/`cat ./CVS/Repository`)
While a file is in the Attic, you can't "add" another file by the same
name. To add such a file you either have to move it by hand as in the
above, or delete it from the Attic.
The main reason for the Attic is to retain files with tags in them. If
you execute: "update -r <oldtag>", files with <oldtag> attached to
some revision will be taken from the normal Repository area and from
the Attic. That's why you can't "add" a file with the same name.
"remove" only moves a file off the main branch, it doesn't obliterate
it.
Last modified: _6/13/1997_
5. Why doesn't "remove" delete the file? Instead, it prints an error
message and tells me to remove the file by hand.
Design choice. Unix software written within last decade, usually
requires an extra verification step, such as answering a question or
adding a flag on the command line. CVS currently requires that you
delete the file first unless you specify the '-f' (force) option,
which deletes the file before performing "cvs remove".
Last modified: _6/13/1997_
Category: /Commands_/rtag_rt_rfreeze/
" + "rtag", "rt", "rfreeze""
1. What is "rtag" for?
To add a symbolic label (a "tag") to the last committed revisions of a
module directly in the Repository.
Last modified: _6/13/1997_
2. Why use "rtag"? It assumes no one is changing the Repository.
Though the "tag" command is more useful in marking the revisions you
have in a particular working directory, "rtag" is much handier for
whole-Repository actions, which occur at major release boundaries.
Last modified: _6/13/1997_
3. What revision does "rtag -r <tag1> <tag2>" actually put the tag on?
In short, the '-r' option is another way to select the revision to
tag. The revision is selected the same way for all commands that
accept a "-r <tag/rev>" option.
Depending on whether <tag1> is a <branch_tag>, or a non-branch <tag>
and on whether you use the '-b' option to "rtag", you get four
different results:
rtag -r <tag1> <tag2>
Adds the non-branch tag <tag2> to the same revision that the
non-branch tag <tag1> is attached to.
Example:
<tag1> --> TT1
<tag2> --> TT2
<file> --> Symbols: TT1:1.4
After --> Symbols: TT1:1.4,TT2:1.4
rtag -r <branch_tag1> <tag2>
Adds the non-branch tag <tag2> to the HEAD of (the highest revision
number on) the branch labelled with tag <branch_tag1>.
Example:
<branch_tag1> --> BR1
<tag2> --> TT2
<file> --> Symbols: BR1:1.2.0.2 (1.2.2.5 is HEAD)
After --> Symbols: BR1:1.2.0.2,TT2:1.2.2.5
If the branch tagged by <branch_tag1> has not been created, then the
tag shows up on the branch point revision:
Example:
<branch_tag1> --> BR1
<tag2> --> TT2
<file> --> Symbols: BR1:1.2.0.2 (No 1.2.X exists.)
After --> Symbols: BR1:1.2.0.2,TT2:1.2
rtag -b -r <tag1> <branch_tag2>
Adds the magic branch tag <branch_tag2> to the revision that the
non-branch tag <tag1> is attached to, preparing it to be a branch
point.
Example:
<tag1> --> TT1
<branch_tag2> --> BR2
<file> --> Symbol: TT1:1.4
After --> Symbol: TT1:1.4, BR2:1.4.0.2
rtag -b -r <branch_tag1> <branch_tag2>
Adds the magic branch tag <branch_tag2> to the revision at the HEAD of
(the highest revision number on) the branch labelled with
<branch_tag1>, preparing it to be a branch point.
Example:
<branch_tag1> --> BR1
<branch_tag2> --> BR2
<file> --> Symbol: BR1:1.2.0.2 (1.2.2.5 is HEAD)
After --> Symbol: BR1:1.2.0.2,BR2:1.2.2.5.0.2
If the branch tagged by <branch_tag1> has not been created, then the
tag shows up as a second branch off the same branch point revision:
Example:
<branch_tag1> --> BR1
<tag2> --> TT2
<file> --> Symbols: BR1:1.2.0.2 (No 1.2.X exists.)
After --> Symbols: BR1:1.2.0.2,TT2:1.2.0.4
In all four cases above, if <tag2> already exists on the file, you get
an error unless you specify the '-F' option.
In all four cases, if <tag1> does not exist on the file, <tag2> is not
added unless you specify the '-f' option.
Last modified: _6/13/1997_
4. What happens if the tags are the same in "rtag -r <tag> <tag>"?
Again, there are four cases depending on whether <tag> is a branch
tag, or a non-branch tag and on whether you use the '-b' option to
"rtag":
rtag -r <tag> <tag>
Is a no-op. It does nothing even with '-F' specified.
If you add the '-f' option ("rtag -f -r <tag> <tag>"), then <tag> is
attached to the latest revision on the Main Branch if the file does
*not* already have <tag> on some revision.
If the <tag> is already on the file, using "rtag -f" is still a no-op.
rtag -r <branch_tag> <branch_tag>
Produces an error, since the <branch_tag> is already on some revision
of the file.
But, "rtag -F -r <branch_tag> <branch_tag>" turns the magic branch tag
into a non-branch tag.
Symbols: BR1:1.4.0.2 becomes Symbols: BR1:1.4
rtag -b -r <tag> <tag>
Produces an error, since the <tag> is already on the file.
But, "rtag -F -b -r <tag> <tag>" turns the non-branch tag into a magic
branch tag.
Symbols: BR1:1.4 becomes Symbols: BR1:1.4.0.2
rtag -b -r <branch_tag> <branch_tag>
Produces an error, since the <branch_tag> is already on the file.
But, "rtag -F -b -r <branch_tag> <branch_tag>" increments the branch
number. It essentially removes the branch and creates a new one by the
same name.
Symbols: BR1:1.2.0.4 becomes Symbols: BR1:1.2.0.6
Last modified: _6/13/1997_
5. Why doesn't "rtag -b -r <branch_tag1> <branch_tag2>" rename or duplicate
a magic branch tag?
None of the "tag" or "rtag" options rename anything. They only apply
(or, with the '-F' option, move) tags to specific revisions in the
file.
See 3M.[3-4] above for details of how it works.
To rename a non-branch tag, see 3O.9. To rename a magic branch tag,
see 4D.5
Last modified: _6/13/1997_
Category: /Commands_/status_st_stat/
" + "status", "st", "stat""
1. What is "status" for?
To display the status of files, including the revision and branch you
are working on and the existence of "sticky" information.
Last modified: _6/13/1997_
2. Why does "status" limit the File: at the top to 17 characters?
Designed that way to line up with other data. You can find the whole
filename in the line beginning with "RCS version:", which is not
limited in length.
Last modified: _6/13/1997_
3. Why does it print "Sticky" lines when the values are "(none)"?
Oversight. It should probably elide lines without information.
Last modified: _6/13/1997_
4. Shouldn't the status "Needs Checkout" be "Needs Update"?
Probably.
[[Did this show up in CVS 1.4?]]
Last modified: _6/13/1997_
Category: /Commands_/tag_ta_freeze/
" + "tag", "ta", "freeze""
1. What is "tag" for?
To add a symbolic label (a "tag") to the RCS files last checked out,
updated or committed in a working directory.
Last modified: _6/13/1997_
2. What is the difference between "tag" and "rtag"?
The end result of both commands is that a <tag>, or symbolic name, is
attached to a single revision in each of a collection of files.
The differences lie in:
The collection of files they work on.
"rtag" works on the collection of files referred to by a "module" name
as defined in the "modules" file, or a relative path within the
Repository.
"tag" works on files and directories specified on the command line
within the user's working directory. (Default is '.')
Both commands recursively follow directory hierarchies within the
named files and directories.
The revisions they choose to tag.
"rtag" places a tag on the latest committed revision of each file on
the branch specified by the '-r' option. By default it tags the Main
Branch.
"tag" places a tag on the BASE (i.e. last checked out, updated or
committed) revision of each file found in the working directory. (The
BASE revision of a file is the one stored in the ./CVS/Entries file.)
A different set of command line options.
For example, "rtag" takes a "-r <oldtag>" option to retag an existing
tag. The "tag" command does not.
How it is logged.
Currently "rtag" records the <tag> and the module in the "history"
file, while "tag" does not.
Last modified: _6/13/1997_
3. Why does "tag -b" not put a tag on the Branch Point revision? How do I
refer to the Branch Point?
This is probably an oversight, or a disbelief in the need for it. If
everything works perfectly, the "update -j" command will do the merge
you need and you don't need to check up on it by playing with the
branch point revision.
The '-b' option attaches a magic branch tag to allow CVS later to
figure out the branch point. The actual revision that <tag> is
attached to does not exist. References to the branch tag are
equivalent to references to the latest revision on the branch.
There is no way to refer to the branch point without adding a
non-branch tag. You might want to add non-branch tags as a habit and
add branch tags later, possibly immediate after adding the non-branch
tag. See 4C.3 on Creating a Branch.
Last modified: _6/13/1997_
4. So "{r}tag" labels a bunch of files. What do you use a Tag for?
You use it to "checkout" the labeled collection of files as a single
object, referring to it by name.
Anywhere a revision number can be used a Tag can be used. In fact tags
are more useful because they draw a line through a collection of
files, marking a development milestone.
The way to think about a Tag is as a curve drawn through a matrix of
filename vs. revision number. Consider this:
Say we have 5 files (in some arbitrary modules, some may be in 2 or
more modules by name, some may be in 2 or more modules because of the
Repository tree structure) with the following revisions:
file1 file2 file3 file4 file5
1.1 1.1 1.1 1.1 /--1.1* <-*- <tag>
1.2*- 1.2 1.2 -1.2*-
1.3 \- 1.3*- 1.3 / 1.3
1.4 \ 1.4 / 1.4
\-1.5*- 1.5
1.6
At some time in the past, the '*' versions were tagged. Think of the
<tag> as a handle attached to the curve drawn through the tagged
revisions. When you pull on the handle, you get all the tagged
revisions. Another way to look at it is that you draw a straight line
through the set of revisions you care about and shuffle the other
revisions accordingly. Like this:
file1 file2 file3 file4 file5
1.1
1.2
1.1 1.3 _
1.1 1.2 1.4 1.1 /
1.2*----1.3*----1.5*----1.2*----1.1 (--- <-- Look here
1.3 1.6 1.3 \_
1.4 1.4
1.5
I find that using these visual aids, it is much easier to understand
what a <tag> is and what it is useful for.
Last modified: _6/13/1997_
5. How do I get "tag" and "rtag" to send mail the way "commit" does?
The "commit" command is supported by two files ("commitinfo" and
"loginfo") not used by other commands. To do logging the same way for
"tag" and "rtag" would require another file like loginfo, which
currently doesn't exist.
The "rtag" command requires a "module" entry, which can specify a
"tag" program using the "-t programname" option on the module line.
There is no equivalent support for "tag".
Last modified: _6/13/1997_
6. Why can't "tag" handle the '-r' option that "rtag" takes?
Oversight. The answer is probably "Fixed in a Future Release."
Last modified: _6/13/1997_
7. After a "tag <tag>" in my working directory, why doesn't "checkout -r
<tag>" somewhere else produce copies of my current files?
The only reason this would fail, other than misspelling the <tag>
string, is that you didn't "commit" your work before "tagging" it.
Only committed revisions may be tagged. Modified files are not marked
for later tagging.
Last modified: _6/13/1997_
8. Why doesn't "tag" write a history record the way "rtag" does?
The "rtag" command was originally intended to place major "release"
tags onto modules. The "tag" functionality was developed to *move* the
more significant tag when slight changes to individual files sneaked
in after the release tag was stamped onto the Repository.
The significant event was the "rtag", which was recorded in the
"history" file for the "history -T" option to work.
It turns out that "tag" is generally more useful than "rtag", so the
model has changed. Future revisions of CVS will probably store both
kinds of tags in the history file.
Last modified: _6/13/1997_
9. How do I rename a <tag>?
For a procedure to rename a branch tag, See section 4D.5 The following
covers only non-branch tags.
First, pick a <newtag> that is not in use. You could reuse (i.e. move)
an existing tag to the new revisions using the '-F' option, but that
will confuse matters when both tags are not already on a file. (It
will probably confuse "rtag -f" too.)
Use "rtag" to place <newtag> only on revisions attached to <oldtag> in
the whole Repository, then delete the old one.
cvs rtag -r <oldtag> <newtag> world
cvs rtag -d <oldtag> world.
You can also checkout or update your working directory to the <oldtag>
and "tag" rather than "rtag" the result. But that will take longer and
it has the chance of producing conflicts.
cvs update -r <oldtag>
cvs tag <newtag>
cvs tag -d <oldtag>
cvs update -A (or cvs update -r <previous_tag>)
Last modified: _6/13/1997_
Category: /Commands_/update_up_upd/
" + "update", "up", "upd""
1. What is "update" for?
The "update" command is by far the most important command and is
probably also the most used command.
It has five purposes: (And many options.)
To display the status of your working files.
Though a plain "update" also displays the status, it does so after
possibly altering your working directory. To see the status of your
working files without changing anything, type:
cvs -n update {optional list of files}
To merge changes made by others to the branch you are working on
into your working files.
Each working directory is attached to a branch, usually the Main
branch. To merge changes made on your working branch since your last
checkout, update or commit, type:
cvs update {optional list of files}
To merge changes made on another branch into the branch you are
working on (your "working branch").
If you want to grab a whole branch, from the branch point, which is
assumed to be on the Main Branch, to the end of the branch, you type:
cvs update -j <branch_tag> {optional files}
If you want to grab the changes made between two tags or revisions,
you type:
cvs update -j <tag1> -j <tag2> {optional files}
(If you are working with a single file, the Tags could also be
revisions numbers. Unless you take great care to match revision
numbers across different files (a waste of time given the way Tags
work), using revision numbers in place of the Tags for multiple files
would be meaningless.)
To move your working directory to another branch.
A working directory is presumed to be attached to (or working on) a
particular branch, usually the Main branch. To alter what CVS believes
to be your working branch, you "move" to that branch.
To move to a tagged branch, type:
cvs update -r <branch_tag> {optional files}
To move to the Main Branch, type:
cvs update -A {optional files}
If you have modified files in your working directory, this is not a
clean move. CVS will attempt to merge the changes necessary to make it
look like you made the same changes to the new branch as you made in
the old one. But if you do this twice without resolving the merge
conflicts each time, you can lose work.
To retrieve old revisions of files.
This option is similar to 4 above but you are not restricted to using
a <branch_tag>. You may specify any revision or <tag> with '-r' and
get the specified revision or the tagged revision:
cvs update -r <tag/rev> {optional files}
Or you may specify any date with '-D':
cvs update -D <date> {optional files}
The '-p' option sends the revisions to standard output (normally your
terminal) rather than setting the "sticky" tag and changing the files.
Last modified: _6/13/1997_
2. What do 'U', 'M' and 'C' mean when I type "update"? Are they different
for "cvs -n update"?
"cvs update" merges changes made to the Repository, since your last
"checkout", "update" or "commit", into your working files. You can
think of it as changing your BASE revision.
"cvs update" prints lines beginning with:
'U' after replacing your unmodified file with a different
revision from the Repository.
'M' for two different reasons:
for files you have modified that have not changed in the Repository.
after a merge, if it detected no conflicts.
'C' after a merge, if it detected conflicts. See 2D.7 and 3P.6 for
more info on conflict resolution and "sticky conflicts."
"cvs -n update" shows what it *would* do, rather than doing it. Or,
another way of looking at it, "cvs -n update" displays the
relationship between your current BASE revisions (identified in your
./CVS/Entries file) and the HEAD revisions (the latest revisions in
the Repository).
"cvs -n update" prints lines beginning with:
'U' for files you have not modified that have changed in the
Repository.
'M' for files you have modified that have not changed in the
Repository.
'C' for files you have modified that have also been changed in the
Repository.
See 4C.6 for what the letters mean when merging in from another
branch. The output is almost the same for a normal update if you
consider the Repository as the branch and your working directory as
the "trunk".
Last modified: _6/13/1997_
3. What's the difference between "update" and "checkout"?
See 3C.4 above.
Last modified: _6/13/1997_
4. Why don't I get new files when I execute "update"?
There are six reasons for nothing to happen during an "update":
Nothing on your branch changed in the Repository.
If no one has committed anything to the branch you are working on
(normally the Main branch) since the last time you executed
"checkout", "update" or "commit", nothing will happen.
It's like shouting "xyzzy" or "plugh" in the wrong room.
You have a "sticky" non-branch <tag> or <date> attached to the
working files you are trying to "update".
At some time in the past you checked out or updated your directory
with the "-r <tag>" or "-D <date>" option. Until you do it again with
a different tag or date, or go back to the Main Branch with "update
-A", you will never again see any updates.
The ./CVS/Entries.Static file exists and you are expecting a new
file.
If your ./CVS administrative directory contains a file named
Entries.Static, no files will be checked out that aren't already in
the Entries or Entries.Static file.
You forgot to use the '-d' option and are looking for new
directories.
If you execute "update" without the '-d' option, it will not create
new directories that have been added to the Repository.
You typed "update" instead of "cvs update".
On most Unix systems, your disk caches are now furiously being flushed
by multiple update daemons, destroying performance and proving to
management that you need more CPU power. :-)
On HP systems you might be asked what package you want to install from
the "update server".
Someone removed (using "admin -o") your BASE revision (the revision
CVS thought you had in your working directory), then committed a
"replacement". CVS is now confused because the revision in the
Repository matches your BASE revision when the files themselves don't
match. See 3B.6.
Last modified: _6/13/1997_
5. Why does "update" say 'M' both for plain modified files and for
successful (i.e. conflict-free) merges? Aren't they different?
A design choice. Yes, they are different internally, but that
shouldn't matter. Your files are in the same condition after the
"update" as they were before -- a "diff" will display only your
modifications. And you are expected to continue onward with parts two
and three of the normal development cycle: "emacs" (a synonym for
"edit" in most of the civilized world) and "commit".
Last modified: _6/13/1997_
6. What's a "sticky conflict"? How does it know a conflict occurred?
When a "cvs update" (or an "update -j") creates a conflict, it prints
a 'C' and stores the timestamp of the file after the merge in a
special field in the ./CVS/Entries file.
This conflict indication implies that the merge command altered your
working file to contain conflict markers surrounding the overlapping
code segments. For example, say that
- Two developers acquire revision 1.2 of <file> via "checkout" or
"update".
- Developer A changes line 1 from "9999" to "5555", then commits the
file, creating revision 1.3.
- Developer B changes line 1 from "9999" to "7777", then tries to
commit the file, but is blocked because the file is not up to date.
Developer B then runs "update" and sees the conflict marker 'C'. The
beginning of the file would look like this:
<<<<<<< <file> The working <file> in question.
7777 Change made to the working <file>.
=======
5555 Change made in the first commit (1.3)
>>>>>>> 1.3 The revision created by the first commit.
The conflict is "sticky", which means that until the conflict is
cleared, the "update" command will continue to display the file's
status as 'C' and the "status" command will show the file's status as
"Unresolved Conflict".
Until the conflict is cleared, "commit" is blocked for this file.
The sticky conflict indicator can be cleared by:
Resolving the conflict by editing the file. Two things must happen
before the conflict is considered resolved:
The timestamp of the file must change. *and* The file must contain no
conflict markers. (The string searched for in the file is the regexp:
"^>>>>>>> ".)
After clearing the sticky conflict indicator, you may then commit the
file normally.
Removing the file and running "update". This throws away the local
changes and accepts the latest committed file on this branch. No
commit is needed.
Forcing the commit to happen by using "commit -f". This is probably
a mistake since there are few lines of real text that begin with
">>>>>>> ".
Last modified: _6/13/1997_
7. Is there a feature to tell me what I have changed, added and removed
without changing anything?
The command "cvs -n update" will do exactly that.
Last modified: _6/13/1997_
8. Why were all my files deleted when I executed "update"?
You probably executed "update -r <tag>" some time ago, then removed
<tag> from the Repository files. "update -r <tag>" will delete a file
that doesn't contain <tag>.
A way to fix this is to "cd" into your working directory and type:
cvs update -A
If you don't want the latest revisions on the Main (or Vendor) Branch,
then decide what Tag (normal or branch) you want and type:
cvs update -r <the_tag_you_want>
Another way to make a file disappear is to execute "update -D <date>"
where <date> is before the date stamped onto the first revision in the
RCS file.
Last modified: _6/13/1997_
Category: /Past__Future_/
" Past & Future "
Category: /Past__Future_/Bugs_and_Patches/
" + Bugs and Patches"
1. Why can't CVS handle deletion of directories?
An oversight, probably. [[Fixed in a future release?]]
Last modified: _6/13/1997_
2. Why can't CVS handle the moving of sources from one place in the
directory hierarchy to another?
A "renaming database" has been proposed to track the history of
pathname changes in the Repository. A general solution is a difficult
problem. See 4B.8.
Last modified: _6/13/1997_
3. When I typed "cvs update -D <date>", why did it check out all
sorts of ancient files from the Attic? Shouldn't it just create the
set of files and revisions that existed at that date?
This seems to be a bug, but is really the lack of any obvious place to
store the date when a file is "removed".
There are four ranges of dates that CVS has to deal with when trying
to determine what revision was available on <date>:
Dates before the earliest revision in the file.
Dates between any two revisions in the file.
Dates between the latest revision in the file and the date when the
file was moved to the Attic by "commit".
Dates after moving the file to the Attic.
Since the date when a file is moved to the Attic is not stored
anywhere, CVS can't tell the difference between #3 and #4. To avoid
not producing a file that should exist in case #3, it produces
extraneous files in case #4.
For the above reason, if you have removed files in the Attic, it is
better to use "-r <tag>, or even "-r HEAD" than to use a date spec.
If you must use "-D <date>", then you should either archive and delete
Attic files (losing some past history) or construct your Makefiles to
work with an explicit list of files and let the old source files stay
in the working directory. The contents of the revision-controlled
Makefile can then be considered to contain deletion "information".
Last modified: _6/13/1997_
4. When I typed "cvs update -D <date>" in my branch, why did it screw up
all my files?
Currently, the internal routine ("version_ts") that looks up info
about a file, overrides both the tag and date if *either* the tag or
date is specified on the command line. If only the date is specified,
it should not override a branch tag, but it does.
In CVS 1.3, the documented "-D <branch_tag>:<date>" syntax only works
with the Main Branch and the Vendor Branch.
[[Is this fixed in CVS 1.4? This is one item I didn't check.]]
Last modified: _6/13/1997_
5. When I executed "checkout" into an existing directory I got "No such
file or directory" errors. Why?
Though the man page says that "checkout" turns into an "update -d" in
directories that already exist, it is referring to directories that
already exist *and* were created by CVS.
When you try to run "checkout" on top of an existing directory
structure, some of which wasn't created by CVS, it will handle
directories and non-CVS files within directories already under CVS,
but it will display the above error on non-CVS files within non-CVS
directories.
Last modified: _6/13/1997_
6. Why does "update" send all output to the terminal after 26 files have
been updated?
CVS uses the "tmpnam()" function to generate temporary file names. The
ANSI standard for the "tmpnam()" function says:
"The tmpnam function generates a different string each time it is
called, up to TMP_MAX times. If it is called more than TMP_MAX times,
the behavior is implementation defined."
Later it says that the value of "TMP_MAX shall be at least 25."
On some platforms, the above specification is taken literally by
turning "at least 25" into "exactly 26" and by doing something foolish
(i.e. "implementation defined") after that. Some systems return the
same name repeatedly, which causes one form of trouble. Others return
NULL or garbage, which causes a different form of trouble.
The broken systems appear to be cycling a single character through the
alphabet. SunOS cycles 3 characters through the alphabet, so it won't
cause trouble until 26 cubed or 17576 calls to "tmpnam()".
Since CVS doesn't depend on the exact format of the tmp files, the
workaround is to provide a "tmpnam()" that doesn't have a limit on the
number of calls to it.
Last modified: _6/13/1997_
7. Why does the merge occasionally resurrect lines of code?
The diff3 program provided by GNU diff version 1.15 has a bug that
occasionally causes text to come back from the dead.
This is an old problem which you can avoid by upgrading to the latest
GNU "diffutils" package. If you were using GNU diff version 1.15 and
plan to upgrade to the latest GNU diff program, see the next question.
Last modified: _6/13/1997_
8. Why does the merge fail when my "rcsmerge" program is configured to use
GNU diff version 2.1 or later?
A change in the overlap format was introduced in GNU diff3 between
versions 2.0 and 2.1 that causes RCS versions before 5.6.0.1 to fail
during a merge.
To get consistent rcsmerge behavior, you have four choices:
Go back to using GNU diff 1.15 or 2.0 with RCS versions 5.5 or 5.6.
If you want to use GNU diff 2.1 or later, you'll have to pick one of
the other three choices in this list.
Grab RCS version 5.6.0.1 from an FSF archive and set the DIFF3_A
macro to '1' as it tells you to in the Makefile:
#define DIFF3_A 1
Patch the RCS 5.6 source. Change line 84 in "merger.c" from:
DIFF3, "-am", "-L", label[0], "-L", label[1], to DIFF3, "-amE", "-L",
label[0], "-L", "", "-L", label[1],
Wait both for RCS version 5.7 to be released and for a new version
of CVS that can deal with it.
Last modified: _6/13/1997_
Category: /Past__Future_/Contributors/
" + Contributors"
1. Who wrote CVS?
Brian Berliner <berliner@sun.com> converted a collection of scripts
written by Dick Grune <dick@cs.vu.nl> into a C program, then added all
sorts of features. He continues to maintain CVS.
Jeff Polk <polk@bsdi.com> wrote much of the code added between
revisions 1.2 and 1.3. Many others were involved at some level.
david d zuhn <zoo@armadillo.com> fixed a number of bugs, added some of
the new features, reworked the whole thing to be more portable, and
provided much of the energy to push CVS 1.4 out the door.
Jim Kingdon implemented CVS 1.5's remote repository access features,
fixed many bugs, and managed the release of version 1.5.
Take a look at the README and the ChangeLog files in the CVS sources
for more contributors.
Last modified: _6/13/1997_
2. You didn't write all of this FAQ, did you?
In the original hunt for questions to answer (performed in Jan/Feb,
1993), I polled hundreds of people and I rephrased all sorts of text
found on the net. Between 2/93 and 10/93, I released about 20
versions, with corrections and additions from the info-cvs mailing
list and private correspondence.
Between 10/93 and 10/94 I extracted frequently asked questions from
the 1200 mail messages to the info-cvs mailing list, turned them into
focused questions and tried to answer them.
93/02/?? ~4000 lines 93/06/?? ~5000 lines 93/10/23 7839 lines 278K
94/10/29 9856 lines 360K 95/05/09 9981 lines 365K
Because there are so many posers of questions, I will list only those
who contribute answers or help significantly with the content and
structure of this document.
If I used someone else's text verbatim, I mentioned it in the given
answer. The people whose email postings have added to this document or
who have added to my understanding are:
Brian Berliner <berliner@sun.com>, CVS maintainer. Paul Eggert
<eggert@twinsun.com>, RCS maintainer.
Gray Watson <gray@antaire.com> Per Cederqvist <ceder@signum.se> Pete
Clark <pclark@is.com>
all of whom have sent me copies of their tutorials and local CVS
documentation.
Additional contributors, who have sent me ideas, text, corrections and
support include (in alphabetical order):
Per Abrahamsen <amanda@iesd.auc.dk> Donald Amby
<amby@mixcom.mixcom.com> Mark D Baushke <mdb@cisco.com> Jim Blandy
<jimb@cyclic.com> Tom Cunningham <tomc@bouwsma,sps.mot.com> Graydon
Dodson <grdodson@lexmark.com> Joe Drumgoole
<joed@splatter.demon.co.uk> Don Dwiggins <dwig@markv.com> Bryant
Eastham <bryant@ced.utah.edu> Dan Franklin <dan@diamond.bbn.com>
Michael Ganzberger <ganzbergermd@ES.net> Steve Harris
<vsh%etnibsd@uunet.uu.net> Erik van Linstee
<linstee@dutecaj.et.tudelft.nl> Jeffrey M Loomis <jml@world.std.com>
Barry Margolin <barmar@near.net> Mark K. Mellis <mkm@ncd.com> Chris
Moore <Chris.Moore@src.bae.co.uk> Gary Oberbrunner <garyo@avs.com>
Steve Turner <stevet@carrier.sps.mot.com> Dave Wolfe
<dwolfe@pffft.sps.mot.com> Dale Woolridge <dwoolridge@cid.aes.doe.ca>
Please send corrections. If I forgot you, remind me and I'll add your
name to the list.
Last modified: _6/13/1997_
Category: /Past__Future_/Development/
" + Development"
1. Where do I send bug reports?
First make sure it is a bug. Talk to your friends, coworkers and
anyone you know who uses CVS. Search this FAQ for related issues. Then
test it carefully. Try out variations to narrow down the problem. Make
sure it is repeatable. Look for workarounds so you can report them.
If you are still sure it's a bug and you tried to fix it, skip to the
next question. Otherwise, send a message to the info-cvs mailing list
containing one of the following:
If you have a good repeatable case and you think you know what is
going on, then describe the problem in detail. Include a workaround if
you have one.
If you have no idea what is going on, go ahead and send a question
to the info-cvs mailing list. Include any information you have
describing the symptoms.
Last modified: _6/13/1997_
2. Where do I send fixes and patches?
First make sure the "fix" does something useful. Have someone review
your fix. Spend a bit of one person's time in a detailed analysis of
your vast idea before displaying a half-vast idea to hundreds of
people.
If you tried to fix it and the patch is small, include the patch in
your message. Make sure the patch is based on the latest released
version of CVS.
If you tried to fix it and the patch is large, you should think about
why it is so large. Did you add a generally useful feature, or did it
grow out of hand?
If you still believe it is solid, produce a patch file using the CVS
commands "patch" or "diff -c". [[You *are* keeping CVS under CVS,
right?]] The patch should be based on the latest released version of
CVS. Then use the "cvsbug" program (provided with the CVS sources) to
send it to the CVS maintainers. A self-contained patch that provides a
single useful feature or correction might show up independently in the
patches directory of the FTP archive.
If careful testing reveals an RCS bug rather than a CVS bug, you can
send bug reports to: rcs-bugs@cs.purdue.edu
Last modified: _6/13/1997_
3. Where do I send ideas for future development?
If you have a bright idea, discuss it on the info-cvs mailing list. If
you have the time to implement something you can test, send the diffs
along too as described above.
Last modified: _6/13/1997_
4. What plans are there for new features?
A "rename" or "per-directory" database has been bandied about on
the net for years. Many of the goals of the rename database have
been achieved by the so-called "death support" in recent versions of
CVS (such as 1.9). For more information on what may remain to be
done, see item #189 in the TODO file of a development version of CVS.
CVS version 1.5 supports remote repository access, but Paul
Kunz has produced another version
(rCVS) that also runs remotely. Note that as far as I know there
are no advantages to rCVS over the remote CVS in CVS 1.5 and later,
and the rCVS user community has migrated to remote CVS.
rCVS is *not* a multisite CVS (see item #186 in TODO for more on
multisite). For more on rCVS, see
ftp://ftp.slac.stanford.edu/software/rcvs
kingdon@cyclic.com
Last modified: _9/6/1997_
5. I have some time and I'd like to help. What can I do for you?
You can review this document, correct errors and fill in any of
the incomplete sections.
You can write scripts or CVS add-ons and make them available by
web/FTP/etc.
You could work on the regression test suite (src/sanity.sh in the
CVS source distribution).
You can write specs for new features, fix bugs, review the
documentation or . . .
For more information, see the files HACKING and DEVEL-CVS in the
CVS source distribution or
http://www.cyclic.com/cyclic-pages/cvsdev.html
kingdon@cyclic.com
Last modified: _9/6/1997_
Category: /Past__Future_/Professional_Support/
" + Professional Support"
1. Doesn't Cygnus support CVS?
Cygnus is a company that supports free software such as the GCC
compiler. They have never sold support for CVS, however. They
do use CVS internally and have contributed much code to CVS over
the years (for which CVS users should be grateful).
kingdon@cyclic.com
Last modified: _9/6/1997_
2. What is Cyclic Software doing with CVS?
Cyclic Software exists to provide support for CVS. For details such
as prices and what this covers, see http://www.cyclic.com or ask
info@cyclic.com.
kingdon@cyclic.com
Last modified: _9/6/1997_
Category: /User_Tasks_/
" User Tasks "
Category: /User_Tasks_/Common_User_Tasks/
" + Common User Tasks"
1. What is the absolute minimum I have to do to edit a file?
Tell your Repository Administrator to create a module covering the
directory or files you care about. You will be told that your module
name is <module>. Then type:
cvs checkout <module>
cd <module>
emacs <file> # Isn't Emacs a synonym for edit?
cvs commit <file>
If you don't use modules (in my opinion, a mistake), you can check out
a directory by substituting its relative path within the Repository
for <module> in the example above.
To work on a single file, you'll have to change "cd <module>" to "cd
`dirname <module>`".
Last modified: _6/13/1997_
2. If I edit multiple files, must I type "commit" for each one?
No. You can commit a list of files and directories, including relative
paths into multiple directories. You can also commit every modified
file in the current directory or in all directories and subdirectories
from your current directory downward. See 3D.2.
Last modified: _6/13/1997_
3. How do I get rid of the <module> directory that "checkout" created?
Change your directory to be the same as when you executed the
"checkout" command that created <module>.
If you want to get rid of the CVS control information, but leave the
files and directories, type:
cvs release <module>
If you want to obliterate the entire directory, type:
cvs release -d <module>
("release -d" searches through the output of "cvs -n update" and
refuses to continue if the "update" command finds any modified files
or non-ignored foreign files. Foreign directories too.)
If you don't care about keeping "history", or checking for modified
and foreign files, you can just remove the whole directory. That's "rm
-rf <module>" under Unix.
Last modified: _6/13/1997_
4. How do I find out what has changed since my last update?
There are many ways to answer this.
To find out what you've changed in your current working directory
since your last checkout, update or commit, type:
cvs diff
To find out what other people have added (to your branch) since you
last checked out or updated, type:
cvs diff -r BASE -r HEAD
To look at a revision history containing the comments for all changes,
you can use the "log" command.
You can also use "history" to trace a wide variety of events.
Last modified: _6/13/1997_
5. I just created a new file. How do I add it to the Repository?
The "update" command will mark files CVS doesn't know about in your
working directory with a '?' indicator.
? <file>
To add <file> to the Repository, type:
cvs add <file>
cvs commit <file>
See 3A.[2-5] and 4C.8 for branch and merge considerations.
Last modified: _6/13/1997_
6. How do I merge changes made by others into my working directory?
If you are asking about other branches, see Section 4C on "Branching".
You will have to use the "update -j" command.
Retrieving changes made to the Repository on the *same* branch you are
working on is the main purpose of the "update" command. The "update"
command tries to merge work committed to the Repository by others
since you last executed "checkout", "update" or "commit" into your
working files.
For a single file, there are six possible results when you type the
"update" command:
If the file is lying in your working directory, but is not under
CVS, it will do nothing but print:
? <file>
If neither you nor anyone else has committed changes to <file>,
since your last "checkout", "update" or "commit", "update" will print
nothing and do nothing.
If you have made no changes to a working file, but you or others
have committed changes to the Repository since your last "checkout",
"update" or "commit" of this working file, CVS will remove your
working file and replace it with a copy of the latest revision of that
file in the Repository. It will print:
U <file>
You might want to examine the changes (using the CVS "diff" command)
to see if they mesh with your own in related files.
If you have made changes to a working file, but no one has changed
your BASE revision (the revision you retrieved from the Repository in
your last "checkout", "update" or "commit"), "update" will print:
M <file>
Nothing changes. You were told that you have a modified file in your
directory.
If you have made changes to your working file and you or others have
committed changes to the Repository, but in different sections of the
file, CVS will merge the changes stored in the Repository since your
last "checkout", "update" or "commit" into your working file. "update"
will print:
RCS file: /Repository/module/<file> retrieving revision 1.X retrieving
revision 1.Y Merging differences between 1.X and 1.Y into <file> M
<file>
If you execute "diff" before and after this step, you should see the
same output, since both the base file and your working file changed in
parallel. This is one of the few times the otherwise nonsensical
phrase "same difference" means something.
If both you and those who committed files (since your last checkout,
update or commit) have made changes to the same section of a file, CVS
will merge the changes into your file as in #5 above, but it will
leave conflict indicators in the file. "update" will print:
RCS file: /Repository/module/<file> retrieving revision 1.X retrieving
revision 1.Y Merging differences between 1.X and 1.Y into <file>
rcsmerge warning: overlaps during merge
cvs update: conflicts found in <file>
C <file>
This is a "conflict". The file will contain markers surrounding the
overlapping text. The 'C' conflict indicator is sticky -- subsequent
"update" commands will continue to show a 'C' until you edit the file.
You must examine the overlaps with care and resolve the problem by
analyzing how to retain the features of both changes. See 2D.7 and
3P.6 for more details on conflict resolution.
Last modified: _6/13/1997_
7. How do I label a set of revisions so I can retrieve them later?
To "tag" the BASE revisions (the ones you last checked out, updated,
or committed) you should "cd" to the head of the working directory you
want to tag and type:
cvs tag <tag>
It recursively walks through your working directory tagging the BASE
revisions of all files.
To "tag" the latest revision on the Main branch in the Repository, you
can use the following from anywhere: (No "cd" is required -- it works
directly on the Repository.)
cvs rtag <tag> <module>
Last modified: _6/13/1997_
8. How do I checkout an old release of a module, directory or file?
Module names and directories are simply ways to name sets of files.
Once the names are determined, there are 6 ways to specify which
revision of a particular file to check out:
By tag or symbolic name, via the "-r <tag>" option.
By date, via the "-D <date>" option.
By branch tag (a type of tag with a magic format), via the "-r
<branch_tag>" option.
By date within a branch, via the "-r <branch_tag>:<date>" option.
By an explicit branch revision number ("-r <rev>"), which refers to
the latest revision on the branch. This isn't really an "old"
revision, from the branch's perspective, but from the user's
perspective the whole branch might have been abandoned in the past.
An explicit revision number: "-r <rev>" Though this works, it is
almost useless for more than one file.
You type:
cvs checkout <option-specified-above> <module>
cd <module>
Last modified: _6/13/1997_
9. What do I have to remember to do periodically?
You should execute "cvs -n update" fairly often to keep track of what
you and others have changed. It won't change anything -- it will just
give you a report.
Unless you are purposely delaying the inclusion of others' work, you
should execute "update" once in a while and resolve the conflicts. It
is not good to get too far out of sync with the rest of the developers
working on your branch.
It is assumed that your system administrators have arranged for editor
backup and Unix temp files (#* and .#*) to be deleted after a few
weeks. But you might want to look around for anything else that is
ignored or hidden. Try "cvs -n update -I !" to see all the ignored
files.
If you are the Repository Administrator, see 4B.16 on Administrator
responsibilities.
Last modified: _6/13/1997_
Category: /User_Tasks_/General_Questions/
" + General Questions"
1. How do I see what CVS is trying to do?
The '-t' option on the main "cvs" command will display every external
command (mostly RCS commands and file deletions) it executes. When
combined with the '-n' option, which prevents the execution of any
command that might modify a file, you can see what it will do before
you let it fly. The '-t' option will *not* display every internal
action, only calls to external programs.
To see a harmless example, try typing:
cvs -nt update
Some systems offer a "trace" or "truss" command that will display all
system calls as they happen. This is a *very* low-level interface that
does not normally follow the execution of external commands, but it
can be useful.
The most complete answer is to read the source, compile it with the
'-g' option and step through it under a debugger.
Last modified: _6/13/1997_
2. If I work with multiple modules, should I check them all out and commit
them occasionally? Is it OK to leave modules checked out?
The simple answers are "Yes."
There is no reason to remove working directories, other than to save
disk space. As long as you have committed the files you choose to make
public, your working directory is just like any other directory.
CVS doesn't care whether you leave modules checked out or not. The
advantage of leaving them checked out is that you can quickly visit
them to make and commit changes.
Last modified: _6/13/1997_
3. What is a "sticky" tag? What makes it sticky? How do I loosen it?
When you execute "update -r <tag>", CVS remembers the <tag>. It has
become "sticky" in the sense that until you change it or remove it,
the tag is remembered and used in references to the file as if you had
typed "-r <tag>" on the command line.
It is most useful for a <branch_tag>, which is a sticky tag indicating
what branch you are working on.
A revision number ("-r <rev-number>") or date ("-D <date>") can also
become sticky when they are specified on the command line.
A sticky tag, revision or date remains until you specify another tag,
revision or date the same way. The "update -A" command moves back to
the Main branch, which has the side-effect of clearing all sticky
items on the updated files.
The "checkout" command creates sticky tags, revisions and dates the
same way "update" does.
Also, the '-k' option records a "sticky" keyword option that is used
in further "updates until "update -A" is specified.
Last modified: _6/13/1997_
4. How do I get an old revision without updating the "sticky tag"?
Use the '-p' option to "pipe" data to standard output. The command
"update -p -r <tag/rev>" sends the selected revision to your standard
output (usually the terminal, unless redirected). The '-p' affects no
disk files, leaving a "sticky tag" unaltered and avoiding all other
side-effects of a normal "update".
If you want to save the result, you can redirect "stdout" to a file
using your shell's redirection capability. In most shells the
following command works:
cvs update -p -r <tag/rev> filename > diskfile
Last modified: _6/13/1997_
5. What operations disregard sticky tags?
The functions that routinely disregard sticky tags are:
Those that work directly on the Repository or its administrative
files:
admin rtag log status remove history
Those that take Tags or revisions as arguments and ignore everything
else: (They also never *set* a sticky tag.)
rdiff import export
The "release" command itself ignores sticky tags, but it calls "cvs
-n update" (which *does* pay attention to a sticky tag) to figure out
what inconsistencies exist in the working directory. If no
discrepancies exist between the files you originally checked out
(possibly marked by a sticky tag) and what is there now, "release -d"
will delete them all.
The "tag" command works on the revision lying in the working
directory however it got there. That the revision lying there might
happen to have a sticky tag attached to it is not the "tag" command's
concern.
The main function that *does* read and write sticky tags is the
"update" command. You can avoid referring to or changing the sticky
tag by using the '-p' option, which sends files to your terminal,
touching nothing else.
The "checkout" command sets sticky tags when checking out a new module
and it acts like "update" when checking out a module into an existing
directory.
The "diff" and "commit" commands use the sticky tags, unless
overridden on the command line. They do not set sticky tags. Note that
you can only "commit" to a file checked out with a sticky tag, if the
tag identifies a branch.
There are really two types of sticky tags, one attached to individual
files (in the ./CVS/Entries file) and one attached to each directory
(in the ./CVS/Tag file). They can differ.
The "add" command registers the desire to add a new file. If the
"directory tag" (./CVS/Tag) file exists at the time of the "add", the
value stored in ./CVS/Tag becomes the "sticky tag" on the new file.
The file doesn't exist in the Repository until you "commit" it, but
the ./CVS/Entries file holds the sticky tag name from the time of the
"add" forward.
Last modified: _6/13/1997_
6. Is there a way to avoid reverting my Emacs buffer after committing a
file? Is there a "cvs-mode" for Emacs?
See Section 4F.1
Last modified: _6/13/1997_
7. How does conflict resolution work? What *really* happens if two of us
change the same file?
While editing files, there is no conflict. You are working on separate
copies of the file stored in the virtual "branch" represented by your
working directories. After one of you commits a file, the other may
not commit the same file until "update" has merged the earlier
committed changes into the later working file.
For example, say you both check out rev 1.2 of <file> and make change
to your working files. Your coworker commits revision 1.3. When you
try to commit your file, CVS says:
cvs commit: Up-to-date check failed for `<file>'
You must merge your coworker's changes into your working file by
typing:
cvs update <file>
which will produce the output described in 2B.6.
If a conflict occurs, the filename will be shown with a status of 'C'.
After you resolve any overlaps caused by the merging process, you may
then commit the file. See 3P.6 for info on "sticky conflicts".
Even if you get a simple 'M', you should examine the differences
before committing the file. A smooth, error-free text merge is still
no indication that the file is in proper shape. Compile and test it at
least.
The answer to two obvious questions is "Yes".
Yes, the first one who commits avoids the merge. Later developers have
to merge the earlier changes into their working files before
committing the merged result. Depending on how difficult the merge is
and how important the contending projects are, the order of commits
and updates might have to be carefully staged.
And yes, between the time you execute "update" and "commit" (while you
are fixing conflicts and testing the results) someone else may commit
another revision of <file>. You will have to execute "update" again to
merge the new work before committing. Most organizations don't have
this problem. If you do, you might consider splitting the file. Or
hiring a manager.
Last modified: _6/13/1997_
8. How can I tell who has a module checked out?
If you "checkout" module names (not relative pathnames) and you use
the release command, the "history" command will display active
checkouts, who has them and where they were checked out. It is
advisory only; it can be circumvented by using the '-l' option on the
main "cvs" command.
Last modified: _6/13/1997_
9. Where did the .#<file>.1.3 file in my working directory come from?
It was created during an "update" when CVS merged changes from the
Repository into your modified working file.
It serves the same purpose as any "backup" file: saving your bacon
often enough to be worth retaining. It is invaluable in recovering
when things go wrong.
Say Developers A (you) and B check out rev 1.3 of file <file>. You
both make changes -- different changes. B commits first, so <file>,v
in the Repository contains revisions up through 1.4.
At this point, there are 5 (yes, five) versions of the file of
interest to you:
Revision 1.3 (What you originally checked out.)
Revision 1.4 (What you need from developer B.)
Your old working file. (Before the update.)
Your new working file. (After the merge caused by "update".)
Revision 1.5 (Which you will commit shortly.)
In the case where your working file was not modified, #1 and #3 will
be the same, as will #2 and #4. In this degenerate case, there is no
need to create #5. The following assumes that your working file was
modified.
If the merge executed by the "update" caused no overlaps, and you
commit the file immediately, #4 and #5 will be the same. But you can
make arbitrary changes before committing, so the difference between #4
and #5 might be more than just the correction of overlaps. In general,
though, you don't need #4 after a commit.
But #3 (which is the one saved as ".#<file>.1.3") holds all of your
work, independent of B's work. It could represent a major effort that
you couldn't afford to lose. If you don't save it somewhere, the merge
makes #3 *disappear* under a potential blizzard of conflicts caused by
overlapping changes.
I have been saved a few times, and others I support have been saved
hundreds of times, by the ability to "diff <original file> <original
file with only my work added>", which can be done in the example above
by the Unix shell command:
cvs update -p -r 1.3 <file> | diff - .#<file>.1.3
The assumption is that the ".#" files will be useful far beyond the
"commit" point, but not forever. You are expected to run the "normal"
Unix cleanup script from "cron", which removes "#*" and ".#*" files
older than a some period chosen by your sysadmin, usually ranging from
7 to 30 days.
A question was raised about the need for #3 after #5 has been
committed, under the assumption that you won't commit files until
everything is exactly as you like them.
This assumes perfect humans, which violates one of the Cardinal rules
of Software Engineering: Never assume any form of discipline on the
part of the users of software. If restrictions are not bound into the
software, then you, the toolsmith, have to arrange a recovery path.
In other words, I've seen every possible variety of screwup you can
imagine in #5. There is no way to make assumptions about what "should"
happen. I've seen #5 filled with zeros because of NFS failures, I've
seen emacs core dumps that leave #5 in an unreasonable state, I've
seen a foolish developer uppercase the whole file (with his "undo"
size set low so he couldn't undo it) and decide that it would be less
work to play with the uppercased file than to blow it away and start
over. I've even seen committed files with conflict markers still in
them, a sure sign of carelessness.
There are all sorts of scenarios where having #3 is incredibly useful.
You can move it back into place and try again.
Last modified: _6/13/1997_
10. What is this "ignore" business? What is it ignoring?
The "update" and "import" commands use collections of Unix wildcards
to skip over files and directories matching any of those patterns.
You may add to the built-in ignore list by adding lines of
whitespace-separated wildcards to the following places: (They are read
in this order.)
In a file named "cvsignore" in $CVSROOT/CVSROOT.
A Repository Administrator uses this to add site-specific files and
patterns to the built-in ignore list.
In a file named ".cvsignore" in your home directory.
For user-specific files. For example, if you use "__" as your default
junk file prefix, you can put "__*" in your .cvsignore file.
People who play around exclusively in directory trees where the
Makefiles are generated by "imake" or "configure" might want to put
"Makefile" in their ignore list, since they are all generated and
usually don't end up in the Repository.
In the CVSIGNORE environment variable.
For session-specific files.
Via the '-I' option on "import" or "update" commands.
For this-command-only files.
In a file named ".cvsignore" within each directory.
The contents of a ".cvsignore" file in each directory is temporarily
added to the ignore list. This way you can ignore files that are
peculiar to that directory, such as executables and other generated
files without known wildcard patterns.
In any of the places listed above, a single '!' character nulls out
the ignore list. A Repository administrator can use this to override,
rather than enhance, the built-in ignore list. A user can choose to
override the system-wide ignore list. For example, if you place "! *.o
*.a" in your .cvsignore file, only *.o *.a files, plus any files a
local-directory .cvsignore file, are ignored.
A variant of the ignore-file scheme is used internally during
checkout. "Module names" found in the modules file (or on the
"checkout" command line) that begin with a '!' are ignored during
checkout. This is useful to permanently ignore (if the '!' path is in
the modules file) or temporarily ignore (if the '!' path is on the
command line) a sub-directory within a Repository hierarchy. For
example:
cvs checkout !gnu/emacs/tests gnu/emacs
would checkout the module (or relative path within $CVSROOT) named
"gnu/emacs", but ignore the "tests" directory within it.
Last modified: _6/13/1997_
11. Is there a way to set user-specific configuration options?
User-specific configuration is available through use of a ".cvsrc"
file in your home directory.
CVS searches the first column of your ~/.cvsrc file for the cvs
command name you invoked. If the command is found, the rest of the
line is treated like a set of command line options, stuffed into the
command line before the arguments you actually typed.
For example, if you always want to see context diffs and you never
want to have to delete a file before you run "cvs remove", then you
should create a .cvsrc file containing the following:
diff -c
remove -f
which will add the given options to every invocation of the given
commands.
[[The rest of this will be removed someday, when CVS changes.]]
I would like to stop here with a comment that the command name to use
is the full, canonical one. But the command that the cvsrc support
uses is the string you typed on the command line, not the proper
command. So to get the full effect of the above example, you should
also add all the alternate command names:
di -c
dif -c
rm -f
delete -f
There are two other limitations that will probably be fixed when CVS
sprouts long option names:
It only affects options made available on the command line.
There is a limited number of short options. With long option names,
there is no problem. You can have as many long options as you like,
affecting anything that looks malleable.
The existing command line options do not come in on/off pairs, so
there is no easy way to override your ~/.cvsrc configuration for a
single invocation of a command.
Choosing a good set of long option pairs would fix this.
Last modified: _6/13/1997_
12. Is it safe to interrupt CVS using Control-C?
It depends on what you mean by "safe". ("Ah," said Arthur, "this is
obviously some strange usage of the word *safe* that I wasn't
previously aware of." -- Hitchhiker's Guide to the Galaxy)
You won't hurt the underlying RCS files and if you are executing a
command that only *reads* data, you will have no cleanup to do.
But you may have to hit Control-C repeatedly to stop it. CVS uses the
Unix "system" routine which blocks signals in the CVS parent process.
A single Control-C during "system" will only halt the child process,
usually some form of RCS command.
If you don't hit another Control-C while the CVS process has control,
it is likely to continue onto the next task assuming that the earlier
one did its job. It is not enough to hit two Control-C's. You might
simply kill two child processes and not interrupt CVS at all.
Depending on the speed of your processor, your terminal and your
fingers, you might have to hit dozens of Control-C's to stop the damn
thing.
Executing a CVS command, such as "commit" or "tag" that writes to the
files is a different matter.
Since CVS is not a full-fledged database, with what database people
call "commit points", merely stopping the process will not back out
the "transaction" and place you back in the starting blocks. CVS has
no concept of an "atomic" transaction or of "backtracking", which
means that a command can be half-executed.
Hitting Control-C will usually leave lock files that you have to go
clean up in the Repository.
Example1:
If you interrupt a multi-file "commit" in the middle of
an RCS checkin, RCS will leave the file either fully
checked-in or in its original state. But CVS might have
been half-way through the list of files to commit. The
directory or module will be inconsistent.
To recover, you must remove the lock files, then decide
whether you want to back out or finish the job.
To back out, you'll have to apply the "admin -o"
command, very carefully, to remove the newly committed
revisions. This is usually a bad idea, but is
occasionally necessary.
To finish, you can simply retype the same commit command.
CVS will figure out what files are still modified and
commit them. It helps that RCS doesn't leave a file in an
intermediate state.
Example2:
If you interrupt a multi-file "tag" command, you have a
problem similar, but not equivalent, to interrupting a
"commit". The RCS file will still be consistent, but
unlike "commit", which only *adds* to the RCS file, "tag"
can *move* a tag and it doesn't keep a history of what
revision a tag used to be attached to.
Normally, you have little choice but to re-execute the
command and allow it to tag everything consistently.
You might be able to recover by carefully re-applying the
tags via the "cvs admin -N" command, but you'll still have
to dig up from outside sources the information you use to
determine what tag was on what revision in what file.
the Repository, or by using the equivalent: "cvs admin".
Halting a new "checkout" should cause no harm. If you don't want it,
"release" (or rm -rf) it. If you do want it, re-execute the command. A
repeated "checkout" from above a directory acts like a repeated
"update -d" within it.
Halting "update" half-way will give you an unpredictable collection of
files and revisions. To continue, you can rerun the update and it
should move you forward into in a known state. To back out, you'll
have to examine the output from the first "update" command, take a
look at each file that was modified and reconstruct the previous state
by editing the ./CVS/Entries file and by using "cvs admin". Good Luck.
Last modified: _6/13/1997_
13. How do I turn off the "admin" command?
In the current revision, you'd have to edit the source code.
Last modified: _6/13/1997_
14. How do I turn off the ability to disable history via "cvs -l"?
In the current revision, you'd have to edit the source code.
Last modified: _6/13/1997_
15. How do I keep certain people from accessing certain directories?
If you don't try to run CVS set[ug]id, you can use Unix groups and
permissions to limit access to the Repository.
If you only want to limit "commit" commands, you can write a program
to put in the "commitinfo" file. In the "contrib" directory, there are
a few scripts that might help you out.
Last modified: _6/13/1997_
Category: /User_Tasks_/Getting_Started/
" + Getting Started"
1. What is the first thing I have to know?
Your organization has most likely assigned one or more persons to
understand, baby-sit and administer the CVS programs and the data
Repository. I call these persons Repository Administrators. They
should have set up a Repository and "imported" files into it.
If you don't believe anyone has this responsibility, or you are just
testing CVS, then *you* are the Repository Administrator.
If you are a normal user of CVS ask your Repository Administrator what
module you should check out.
Then you can work.
If you *are* the Repository Administrator, you will want to read
everything you can get your hands on, including this FAQ. Source
control issues can be difficult, especially when you get to branches
and release planning. Expect to feel stupid for a few days/weeks.
No tool in the universe avoids the need for intelligent organization.
In other words, there are all sorts of related issues you will
probably have to learn. Don't expect to dive in without any
preparation, stuff your 300 Megabytes of sources into CVS and expect
to start working. If you don't prepare first, you will probably spend
a few sleepless nights.
Last modified: _6/13/1997_
2. Where do I work?
Wherever you have disk space. That's one of the advantages of CVS: you
use the "checkout" command to copy files from the Repository to your
working directory, which can be anywhere you have the space.
Your local group might have conventions for where to work. Ask your
peers.
Last modified: _6/13/1997_
3. What does CVS use from my environment?
You must set two environment variables. Some shells share these
variables with local shell variables using a different syntax. You'll
have to learn how your shell handles them.
Variable Value (or action)
--------- ---------------------
CVSROOT Absolute pathname of the head of your Repository.
PATH Normally set to a list of ':'-separated directory
pathnames searched to find executables. You must
make sure "cvs" is in one of the directories.
If your CVS was built with the RCSBIN directory set
to null (""), and you don't set the RCSBIN
variable mentioned below, then the RCS commands
also must be somewhere in your PATH.
Optional variables: (Used if set, but ignored otherwise.)
Variable Value (or action)
--------- ---------------------
CVSEDITOR The name of your favorite fast-start editor
program. You'll be kicked into your editor to
supply revision comments if you don't specify them
via -m "Log message" on the command line.
EDITOR Used if CVSEDITOR doesn't exist. If EDITOR
doesn't exist, CVS uses a configured constant,
usually, "vi".
CVSREAD Sets files to read-only on "checkout".
RCSBIN Changes where CVS finds the RCS commands.
CVSIGNORE Adds to the ignore list. See Section 2D.
Other variables used by CVS that are normally set upon login:
Variable Value (or action)
--------- ---------------------
LOGNAME Used to find the real user name.
USER Used to find the real user name if no LOGNAME.
HOME Used to determine your home directory, if set.
Otherwise LOGNAME/USER/getuid() are used to find
your home directory from the passwd file.
TMPDIR Used during import. It might also be used if your
platform's version of mktemp(3) is unusual, or
you have changed the source to use tmpnam(3).
Last modified: _6/13/1997_
4. OK, I've been told that CVS is set up, my module is named "ralph" and I
have to start editing. What do I type?
cd <where you have some space to work>
cvs checkout ralph
cd ralph
And hack away.
Last modified: _6/13/1997_
5. I have been using RCS for a while. Can I convert to CVS without losing
my revision history? How about converting from SCCS?
If you are asking such questions, you are not a mere user of CVS, but
one of its Administrators! You should take a look at Section 4A,
"Installing CVS" and Section 4B, "Setting up and Managing the
Repository".
Last modified: _6/13/1997_
Category: /User_Tasks_/Less_Common_User_Tas/
" + Less Common User Tasks"
1. Can I create non-CVS sub-directories in my working directory?
Yes. Unless the directory exists in the Repository, "update" will skip
over them and print a '?' the way it does for files you forgot to add.
You can avoid seeing the '?' by adding the name of the foreign
directory to the ./.cvsignore file, just ask you can do with files.
If you explicitly mention a foreign directory on the "update" command
line, it will traverse the directory and waste a bit of time, but if
any directory or sub-directory lacks the ./CVS administrative
directory, CVS will print an error and abort.
Last modified: _6/13/1997_
2. How do I add new sub-directories to the Repository?
The "add" command will work on directories. You type:
mkdir <dir>
cvs add <dir>
It will respond:
Directory /Repos/<dir> added to the repository
and will create both a matching directory in the Repository and a
./CVS administrative directory within the local <dir> directory.
Last modified: _6/13/1997_
3. How do I remove a file I don't need?
(See the questions in Section 4B on removing files from the
Repository.)
You type:
rm <file>
cvs remove <file>
CVS registers the file for removal. To complete the removal, you must
type:
cvs commit <file>
CVS moves the file to the Attic associated with your working
directory. Each directory in the Repository stores its deleted files
in an Attic sub-directory. A normal "checkout" doesn't look in the
Attic, but if you specify a tag, a date or a revision, the "checkout"
(or "update") command will retrieve files from the Attic with that
tag, date or revision.
Last modified: _6/13/1997_
4. How do I rename a file?
CVS does not offer a way to rename a file in a way that CVS can track
later. See Section 4B for more information.
Here is the best (to some, the only acceptable) way to get the effect
of renaming, while preserving the change log:
Copy the RCS (",v") file directly in the Repository.
cp $CVSROOT/<odir>/<ofile>,v $CVSROOT/<ndir>/<nfile>,v
By duplicating the file, you will preserve the change history and the
ability to retrieve earlier revisions of the old file via the "-r
<tag/rev>" or "-D <date>" options to "checkout" and "update".
Remove the old file using CVS.
cd <working-dir>/<odir> rm <ofile>
cvs remove <ofile>
cvs commit <ofile>
This will move the <ofile> to the Attic associated with <odir>.
Retrieve <nfile> and remove all the Tags from it.
By stripping off all the old Tags, "checkout -r" and "update -r" won't
retrieve revisions Tagged before the renaming.
cd <working-dir>/<ndir>
cvs update <nfile>
cvs log <nfile> # Save the list of Tags
cvs tag -d <tag1> <nfile>
cvs tag -d <tag2> <nfile>
. . .
This technique can be used to rename files within one directory or
across different directories. You can apply this idea to directories
too, as long as you apply the above to each file and don't delete the
old directory.
Of course, you have to change your build system (e.g. Makefile) in
your <working-dir> to know about the name change.
Warning: Stripping the old tags from the copied file will allow "-r
<tag>" to do the right thing, but you will still have problems with
"-D <date>" because there is no place to store the "deletion time".
See 5B.3 for more details.
Last modified: _6/13/1997_
5. How do I make sure that all the files and directories in my working
directory are really in the Repository?
A "cvs update", or "cvs -n update" (which won't modify your working
directory) will display foreign elements, which have no counterpart in
the Repository, preceded by a '?'. To register foreign directories,
you can use "cvs add". To register foreign files, you can use "cvs
add" followed by "cvs commit".
You could also checkout your module, or the Repository directory
associated with your working directory, a second time into another
work area and compare it to your working directory using the (non-CVS)
"diff -r" command.
By default many patterns of files are ignored. If you create a file
named "core" or a file ending in ".o", it is usually ignored. If you
really want to see all the files that aren't in the Repository, you
can use a special "ignore" pattern to say "ignore no files". Try
executing: (You may have to quote or backwhack (i.e. precede by '\')
the '!' in your shell.)
cvs -n update -I !
The above command will display not only the normal modified, update
and conflict indicators ('M', 'U', and 'C' respectively) on files
within the Repository, but it will also display each file not in the
Repository preceded by a '?' character.
The '-n' option will not allow "update" to alter your working
directory.
Last modified: _6/13/1997_
6. How do I create a branch?
Type this in your working directory:
cvs tag -b <branch_tag>
and you will create a branch. No files have real branches in them yet,
but if you move onto the branch by typing:
cvs update -r <branch_tag>
and commit a file in the normal way:
cvs commit <file>
then a branch will be created in the underlying <file>,v file and the
new revision of <file> will appear only on that branch.
See Section 4C, on Branching.
Last modified: _6/13/1997_
7. How do I modify the modules file? How about the other files in the
CVSROOT administrative area?
A module named "modules" has been provided in the default modules
file, so you can type:
cvs checkout modules
cd modules
Another module named CVSROOT has been provided in the default modules
file, covering all the administrative files. Type:
cvs checkout CVSROOT
cd CVSROOT
Then you can edit your files, followed by:
cvs commit
If you start with the provided template for the "modules" file, the
CVSROOT and the "modules" module will have the "mkmodules" program as
a "commit helper". After a file is committed to such a module,
"mkmodules" will convert a number of standard files (See 4B.2) in the
CVSROOT directory inside the Repository into a form that is usable by
CVS.
Last modified: _6/13/1997_
8. How do I split a file into pieces, retaining revision histories?
If you and a coworker find yourselves repeatedly committing the same
file, but never for changes in the same area of the file, you might
want to split the file into two or more pieces. If you are both
changing the same section of code, splitting the file is of no use.
You should talk to each other instead.
If you decide to split the file, here's a suggestion. In many ways, it
is similar to multiple "renamings" as described in 2C.4 above.
Say you want to split <fileA>, which already in the Repository, into
three pieces, <fileA>, <fileB> and <fileC>.
Copy the RCS (",v") files directly in the Repository, creating the
new files, then bring readable copies of the new files into the
working directory via "update".
cp $CVSROOT/<path>/<fileA>,v $CVSROOT/<path>/<fileB>,v cp
$CVSROOT/<path>/<fileA>,v $CVSROOT/<path>/<fileC>,v
cvs update <fileB> <fileC>
Then remove all the <tags> from the new files by using:
cvs log <fileB> <fileC> # Save the list of <tag?>
cvs tag -d <tag1> <fileB> <fileC>
cvs tag -d <tag2> <fileB> <fileC>
. . .
Edit each file until it has the data you want in it. This is a
hand-editing job, not something CVS can handle. Then commit all the
files.
[From experience, I'd suggest making sure that only one copy of each
line of code exists among the three files, except for "include"
statements, which must be duplicated. And make sure the code
compiles.]
emacs <fileA> <fileB> <fileC>
cvs commit <fileA> <fileB> <fileC>
As in the "rename" case, by duplicating the files, you'll preserve the
change history and the ability to retrieve earlier revisions.
Of course, you have to alter your build system (e.g. Makefiles) to
take the new names and the change in contents into account.
Last modified: _6/13/1997_
Category: /What_is_CVS_/
" What is CVS? "
Category: /What_is_CVS_/How_does_CVS_differ_/
" + How does CVS differ from other, similar software?"
1. How does CVS differ from RCS?
CVS uses RCS to do much of its work and absolutely all the work of
changing the underlying RCS files in the Repository.
RCS comprises a set of programs designed to keep track of changes to
individual files. Of course, it also allows you to refer to multiple
files on the command line, but they are handled by iterating over
individual files. There is no pretense of coordinated interaction
among groups of files.
CVS's main intent is to provide a set of grouping functions that allow
you to treat a collection of RCS files as a single object. Of course,
CVS also has to do a lot of iteration, but it tries its best to hide
that it is doing so. In addition, CVS has some truly group-oriented
facets, such as the modules file and the CVS administrative files that
refer to a whole directory or module.
One group aspect that can be a bit confusing is that a CVS branch is
not the same as an RCS branch. To support a CVS branch, CVS uses
"tags" (what RCS calls "symbols") and some local state, in addition to
RCS branches.
Other features offered by CVS that are not supported directly by RCS
are
Automatic determination of the state of a file, (e.g. modified,
up-to-date with the Repository, already tagged with the same string,
etc.) which helps in limiting the amount of displayed text you have to
wade through to figure out what changed and what to do next.
A copy-modify-merge scheme that avoids locking the files and allows
simultaneous development on a single file.
Serialization of commits. CVS requires you to merge all changes
committed (via "update") since you checked out your working copy of
the file. Although it is still possible to commit a file filled with
old data, it is less likely than when using raw RCS.
Relatively easy merging of releases from external Vendors.
Last modified: _6/13/1997_
2. How does CVS differ from SCCS?
SCCS is much closer to RCS than to CVS, so some of the previous entry
applies.
You might want to take a look at Walter Tichy's papers on RCS, which
are referred to in the RCS man pages.
[[More info here?]]
Last modified: _6/13/1997_
3. How does CVS differ from ClearCase?
ClearCase is a distributed client-server version control system.
ClearCase is a variant DSEE tools, formerly available on Apollo
platforms. The ClearCase tool set includes a few X-based interface
tools, a command-line interface, and C programmer API. It is currently
available on Sun, HP, SGI and OSF/1 platforms.
ClearCase uses a special Unix filesystem type, called "mvfs" for
"multi-version file system". Conceptually, mvfs adds another dimension
to a regular Unix filesystem. The new axis is used to store the
different versions of files and to provide a tree-hierarchical view of
a collection of objects that might be scattered across any number of
separate hosts on your local network.
Each user acquires a "view" into the file database by creating a
special mvfs mount point on their machine. Each view has a
"configuration spec" containing a set of selection rules that specify
the particular version of each file to make visible in that view. You
can think of a "view" as a work area in CVS, except that the files
don't really exist on your local disk until you modify them. This
technique conserves disk space because it doesn't keep private copies
of read-only files.
Another advantage is that a view is "transparent" in the sense that
all of the files in a "view" appear to be regular Unix files to other
tools and Unix system calls. An extended naming convention allows
access to particular versions of a file directly:
"test.cc@@/main/bugfix/3" identifies the third version of test.c on
the bugfix branch.
ClearCase supports both the copy-modify-merge model of CVS (by using
what are called "unreserved checkouts" and the checkin/checkout
development model with file locking. Directories are
version-controlled objects as well as files. A graphical merge tool is
provided. Like RCS, ClearCase supports branches, symbolic tags, and
delta compression. ASCII as well as binary files are supported, and
converters from RCS, SCCS, DSEE formats are also included.
A make-compatible build facility is provided that can identify common
object code and share it among developers. A build auditing feature
automatically records file dependencies by tracking every file that is
opened when producing a derived object, thus making explicit
dependency lists unnecessary. Pre- and post-event triggers are
available for most ClearCase operations to invoke user programs or
shell scripts. User-defined attributes can be assigned to any version
or object. Hyper-links between version controlled objects can record
their relationship.
For more information, contact:
Atria Software, Inc. 24 Prime Park Way Natick, MA 01760 info@atria.com
(508) 650-1193 (phone) (508) 650-1196 (fax)
Originally contributed by Steve Turner
Edited by the author of this FAQ.
Last modified: _6/13/1997_
4. How does CVS differ from TeamWare/SparcWorks?
TeamWare is a configuration management tool from Sun Microsystems, a
part of SparcWorks. It uses the same copy and merge model as CVS. The
central abstraction is a workspace, which corresponds to either a CVS
branch or a checked out module. TeamWare allows you to manipulate
workspaces directly, including moving and merging code between
workspaces. You can put your workspace on tape and continue to work
with it at home, just like you can with CVS. TeamWare is built upon
and compatible with SCCS.
TeamWare provides both a command line interface and a graphical
interface. The CodeManager tool will display the project as a tree of
workspaces, and allows you to manipulate them with drag and drop. The
other tools are VersionTool that displays and manipulates a dag with a
version history of a single file, CheckPoint that will create symbolic
tags, MakeTool, a make compatible tool with a GUI, and FileMerge which
will interactively merge files when needed (like emerge for emacs). If
you have a sun, you can try /usr/old/mergetool for an old SunView
version of FileMerge.
Email: sunprosig@sun.com
Originally extracted from TeamWare
Marketing literature by Per Abrahamsen.
Edited by the author of this FAQ.
For more information, contact:
SunExpress, Inc. P.O. Box 4426 Bridgeton, MO 63044-9863 (800)873-7869
Last modified: _6/13/1997_
5. How does CVS differ from Aegis?
Aegis appears to be a policy-setting tool that allows you to use other
sub-programs (make, RCS, etc.) to implement pieces of the imposed
policy.
The initial document seems to say that most Unix tools are inadequate
for use under Aegis.
It is not really similar to CVS and requires a different mindset.
[[Need more info here.]]
Last modified: _6/13/1997_
6. How does CVS differ from Shapetools?
Shapetools includes a build mechanism (called Shape, not surprisingly)
that is aware of the version mechanism, and some dependency tracking.
It is based on a file system extension called Attributed File System,
which allows arbitrary-sized "attributes" to be associated with a
file. Files are version controlled in a manner similar to RCS.
Configurations are managed through the Shapefile, an extension of the
Makefile syntax and functionality. Shape includes version selection
rules to allow sophisticated selection of component versions in a
build.
Shapetools' concurrency control is pessimistic, in contrast to that of
CVS. Also, there's very limited support for branching and merging. It
has a built-in policy for transitioning a system from initial
development to production.
Contributed by Don Dwiggins
Last modified: _6/13/1997_
7. How does CVS differ from TeamNet?
TeamNet is a configuration management tool from TeamOne.
For more information, contact:
TeamOne 710 Lakeway Drive, Ste 100 Sunnyvale, CA 94086 (800) 442-6650
Contributed by Steve Turner
Last modified: _6/13/1997_
8. How does CVS differ from ProFrame?
ProFrame is a new system integration framework from IBM. ProFrame is
compliant with the CFI (CAD Framework Initiative) industry standards,
including the Scheme extension language.
ProFrame consists of three major components: (1) the Process Manager
that automates your local design methodology (2) the Design Data
Manager handles configuration management, and (3) Inter-tool
Communication to provide a communication path among tools running on
heterogeneous servers.
The Design Data Manager(2) is probably the appropriate component to
compare to CVS. The Design Data Manager provides version control with
checkin/checkout capability, configuration management, and data
dependency tracking. A graphical data selection interface is provided.
Using this interface, you may create and manipulate objects and
hierarchy structures, view the revision history for an object, and
view and assign attributes to a design object.
The ProFrame server currently runs only on RS6000, but clients may be
a wide variety of Unix platforms. Contact IBM for the latest platform
information.
For more information, contact:
IBM EDA Marketing and Sales P.O. Box 950, M/S P121 Poughkeepsie, NY
12602 (800) 332-0066
Contributed by Steve Turner
[extracted from the ProFrame 1.1.0 datasheet]
Last modified: _6/13/1997_
9. How does CVS differ from CaseWare/CM?
CaseWare/CM is a software configuration management product from
CaseWare, Inc. CaseWare/CM may be customized to support a wide variety
of methodologies, including various phases of the software lifecycle,
and different access rights for users.
A GUI is provided to view version histories and configurations. A
merge tools is also included. CaseWare supports type-specific
lifecycles, which allows different types of files to move through
different lifecycles. Also provided is a build facility to support
automatic dependency analysis, parallel, distributed, and remote
builds, and variant releases.
CaseWare/CM has been integrated with other CASE tools, including
FrameMaker, ALSYS Ada, CodeCenter/Object Center, HP SoftBench, and
Software Through Pictures. CaseWare also offers CaseWare/PT, a problem
tracking system to integrate change requests with configuration
management.
Multiple vendors and operating systems are supported.
For more information, contact:
CaseWare, Inc. 108 Pacifica, 2nd Floor Irvine, CA 92718-3332 (714)
453-2200 (phone) (714) 453-2276 (fax)
Contributed by Steve Turner
[extracted from the CaseWare/CM data sheet]
Last modified: _6/13/1997_
10. How does CVS differ from Sublime?
Produced by AT&T. Sablime uses SCCS as the underlying source code
control system. It uses some other control system (called sbcs I
think) for managing binary files. It uses lock, edit, comit, unlock
mechanism. It has a motif based GUI and curses based GUI (that works
only with ksh, not tcsh, or bash) to do more common tasks. It has even
a command line interface.
Changing source happens as a result of MR. A testing person or a
developer assigns an MR (modification request) to a group of people.
They are allowed to take out files under that MR and change them and
check them back in. You can set up dependencies between and MR and do
release management to say "I want the sources to include these MRs"
etc. It is a reasonably good maintanance system. It is bit heavy
weight though, and the interface is not too polished and does not work
on windows (though that may have changed). rama@savera.com
Last modified: _12/12/1997_
11. How does CVS differ from PVCS?
PVCS works on single files like RCS and SCCS, CVS works on complete
subsystems. PVCS has a make utility (called a configuration builder),
CVS does not. PVCS has a GUI interface for Unix, DOS, OS/2, and MS
Windows.
Intersolv, Inc.
1700 NW 167th Place
OR 97006
Contributed by Per Abrahamsen
[Extracted from Intersolv Marketing literature.]
Last modified: _6/13/1997_
12. How does CVS differ from CMVC?
CMVC is an IBM Configuration Management and Version Control system.
(Though I'm not certain that's the right acronym expansion.) It runs
on Suns, HPs, RS6000s, OS/2 and Windows.
Other than revision control, it apparently has features to manage
releases, bug tracking and the connection between alterations and
reported bugs and feature requests. It is a client/server system,
based on a choice of commercial Relational Database systems, and it
provides a Motif or command line interface.
Unlike CVS, it uses a strict locking protocol to serialize source code
alterations.
Last modified: _6/13/1997_
Category: /What_is_CVS_/What_do_you_mean_by_/
" + What do you mean by . . .? (Definitions)"
1. What are "The Repository", "$CVSROOT" and "CVSROOT"?
The Repository is a directory tree containing the CVS administrative
files and all the RCS files that constitute "imported" or "committed"
work. The Repository is kept in a shared area, separate from the
working areas of all developers.
Users of CVS must set their "CVSROOT" environment variable to the
absolute pathname of the head of the Repository. Most command line
interpreters replace an instance of "$CVSROOT" with the value of the
"CVSROOT" environment variable. By analogy, in this document
"$CVSROOT" is used as shorthand for "the absolute pathname of the
directory at the head of the Repository".
One of the things found in $CVSROOT is a directory named CVSROOT. It
contains all the "state", the administrative files, that CVS needs
during execution. The "modules", "history", "commitinfo", "loginfo"
and other files can be found there. See 4B.2 for more information
about CVSROOT files.
Last modified: _6/13/1997_
2. What is an RCS file?
An RCS file is a text file containing the source text and the revision
history for all committed revisions of a source file. It is stored
separately from the working files, in a directory hierarchy, called
the Repository.
RCS is the "Revision Control System" that CVS uses to manage
individual files. RCS file names normally end in ",v", but that can be
altered (via the RCS -x option) to conform to file naming standards on
platforms with unusual filename limitations.
Last modified: _6/13/1997_
3. What is a working file?
A working file is a disk file containing a checked-out copy of a
source file that earlier had been placed under CVS. If the working
file has been edited, the changes since the last committed revision
are invisible to other users of CVS.
Last modified: _6/13/1997_
4. What is a working directory (or working area)?
A working directory is the place where you work and the place from
which you "commit" files.
The "checkout" command creates a tree of working directories, filling
them with working files. Each working directory contains a
sub-directory named ./CVS containing three administrative files, which
are created by "checkout" and are always present:
./CVS/Entries
contains information about working files.
./CVS/Repository
contains the location of the directory within the
Repository that was used to create the working directory.
./CVS/Root
contains the value of $CVSROOT at the time you created
the working directory.
Other files may also appear in ./CVS depending on the state of your
working directory:
./CVS/Tag
contains the "sticky tag" associated with the whole
directory. See 3A.2 for its main purpose.
[Created by "checkout" or "update" when using "-r <tag>".]
[Deleted by "checkout" or "update" when using '-A'.]
./CVS/Entries.Static
contains a fixed list of working files. If this file
exists, an "update" doesn't automatically bring newly
added files out of the Repository.
[Created and maintained by hand.]
./CVS/Checkin.prog
contains a program to run whenever anything in the
working directory is committed.
[Created by checkout if "-i <prog>" appears in the
modules file for the checked-out module.]
./CVS/Update.prog
contains a program to run whenever anything in the
working directory is updated.
[Created by checkout if "-u <prog>" appears in the
modules file for the checked-out module.]
./CVS/<file>,p ./CVS/<file>,t
contain (possibly zero-length) state information about an
"add" that has not been committed.
[Created by "add".]
[Deleted by "commit" or "remove".]
Last modified: _6/13/1997_
5. What is "checking out"?
"Checking out" is the act of using the "checkout" command to copy a
particular revision from a set of RCS files into your working area.
You normally execute "checkout" only once per working directory (or
tree of working directories), maintaining them thereafter with the
"update" command.
See section 3C on the "checkout" command.
Last modified: _6/13/1997_
6. What is a revision?
A "revision" is a version of a file that was "committed" ("checked
in", in RCS terms) some time in the past. CVS (and RCS) can retrieve
any file that was committed by specifying its revision number or its
"tag" ("symbolic name", in RCS terms).
In CVS, a "tag" is more useful than a revision number. It usually
marks a milestone in development represented by different revision
numbers in different files, all available as one "tagged" collection.
Sometimes the word "revision" is used as shorthand for "the file you
get if you retrieve (via "checkout" or "update") the given revision
from the Repository."
Last modified: _6/13/1997_
7. What is a "Tag"?
A "Tag" is a symbolic name, a synonym or alias for a particular
revision number in a file. The CVS "tag" command places the same "Tag"
on all files in a working directory, allowing you to retrieve those
files by name in the future.
The CVS "Tag" is implemented by applying RCS "symbols" to each
individual file. The Tags on a file (or collection of files) may be
displayed using the "log" command.
Last modified: _6/13/1997_
8. What are "HEAD" and "BASE"?
HEAD and BASE are built-in tags that don't show up in the "log" or
"status" listings. They are interpreted directly by CVS.
"HEAD" refers to the latest revision on the current branch in the
Repository. The current branch is either the main line of development,
or a branch in development created by placing a branch tag on a set of
files and checking out that branch.
"BASE" refers to the revision on the current branch you last checked
out, updated, or committed. If you have not modified your working
file, "BASE" is the committed revision matching it.
Most of the time BASE and HEAD refer to the same revision. They can
become different in two ways:
Someone else changed HEAD by committing a new revision of your file
to the Repository. You can pull BASE up to equal HEAD by executing
"update".
You moved BASE backward by executing "checkout" or "update" with the
option "-r <rev/tag>" or "-D <date>". CVS records a sticky tag and
moves your files to the specified earlier revision. You can clear the
sticky tag and pull BASE up to equal HEAD again by executing "update
-A".
Last modified: _6/13/1997_
9. What is a Branch?
In general, a branch is any mechanism that allows one or more
developers to modify a file without affecting anyone other than those
working on the same branch.
There are four kinds of "branch" CVS can manage:
The Vendor Branch.
A single vendor branch is supported. The "import" command takes a
sequence of releases from a source code vendor (called a "vendor" even
if no money is involved), placing them on a special "Vendor" branch.
The Vendor branch is considered part of the "Main line" of
development, though it must be merged into locally modified files on
the RCS Main branch before the "import" is complete.
See Section 3H ("import").
Your Working directory.
A checked-out working directory, can be treated like a private branch.
No one but you can touch your files. You have complete control over
when you include work committed by others. However, you can't commit
or tag intermediate versions of your work.
A Development branch.
A group of developers can share changes among the group, without
affecting the Main line of development, by creating a branch. Only
those who have checked-out the branch see the changes committed to
that branch. This kind of branch is usually temporary, collapsing
(i.e. merge and forget) into the Main line when the project requiring
the branch is completed.
You can also create a private branch of this type, allowing an
individual to commit (and tag) intermediate revisions without changing
the Main line. It should be managed exactly like a Development Branch
-- collapsed into the Main line (or its parent branch, if that is not
the Main Branch) and forgotten when the work is done.
A Release branch.
At release time, a branch should be created marking what was released.
Later, small changes (sometimes called "patches") can be made to the
release without including everything else on the Main line of
development. You avoid forcing the customer to accept new, possibly
untested, features added since the release. This is also the way to
correct bugs found during testing in an environment where other
developers have continued to commit to the Main line while you are
testing and packaging the release.
Although the internal format of this type of branch (branch tag and
RCS branches) is the same as in a development branch, its purpose and
the way it is managed are different. The major difference is that a
Release branch is normally Permanent. Once you let a release out the
door to customers, or to the next stage of whatever process you are
using, you should retain forever the branch marking that release.
Since the branch is permanent, you cannot incorporate the branch fixes
into the Main line by "collapsing" (merging and forgetting) the
release branch. For large changes to many files on the release branch,
you will have to perform a branch merge using "update -j <rev> -j
<rev>". (See 4C.7)
The most common way to merge small changes back into Main line
development is to make the change in both places simultaneously. This
is faster than trying to perform a selective merge.
See 1D.12 (merges) and Section 4C, on Branching for more info.
Last modified: _6/13/1997_
10. What is "the trunk"?
Another name for the RCS Main Branch. The RCS Main Branch is related,
but not equivalent, to both the CVS Main branch and what developers
consider to be the Main line of development. See 3H.3 and Section 4C
on Branching.
Last modified: _6/13/1997_
11. What is a module?
In essence, a module is a name you hand to the "checkout" command to
retrieve one or more files to work on. It was originally intended to
be a simple, unique name in the "modules" file attached to a directory
or a subset of files within a directory.
The module idea is now a somewhat slippery concept that can be defined
in two different ways:
* A module is an argument to "checkout". There are three types:
1. An entry in the modules file. A "module" name as described in
'B.' below.
2. A relative path to a directory or file in the Repository.
3. A mixed-mode string of "modulename/relative-path". Everything
up to the first slash ('/') is looked up as a module. The
relative path is appended to the directory associated with
the module name and the resulting path is checked out as in
#2 above.
* A module is a unique (within the file) character string in the
first column of the modules file. There are five types:
1. A name for a directory within the Repository that allows you
to ignore the parent directories above it.
Example:
emacs gnu/emacs
2. A name for a subset of the files within such a directory.
Example:
ls unix/bin Makefile ls.c
The 2nd through Nth strings in the above can be files,
directories or module substitutions. No relative paths.
A module substitution occurs when you use a '&module-name'
reference. The module-name referred to is logically
substituted for the '&module-name' string.
3. A relative pathname to a directory within the Repository
which, when checked out, creates an image of part of the
Repository structure in your current directory.
Example:
gnu/emacs -o /bin/emacs.helper gnu/emacs
The files checked out are exactly the same as the files
"checkout" would retrieve if the path weren't even in the
modules file. The only reason to put this kind of relative
pathname into the modules file is to hook one of the helper
functions onto it.
4. A relative pathname to a single file within the Repository
which, when checked out, creates something you probably don't
want: It creates a directory by the name of the file and puts
the file in it.
Example:
gnu/emacs/Makefile -o /bin/emacs.helper gnu/emacs Makefile
The file checked out is the same as what you would get if you
handed the relative pathname to the "checkout" command. But
it puts it in a strange place. The only reason to do this is
to hook a helper function onto a specific file name.
5. An alias consisting of a list of any of the above, including
other aliases, plus exceptions.
Example:
my_work -a emacs !emacs/tests gnu/bison unix/bin/ls.c
The exception "!emacs/test" above is functionally equivalent
to specifying "!emacs/tests" on the "checkout" command line.
Another way to look at it is that the modules file is simply another
way to "name" files. The hierarchical directory structure provides
another. You should use whatever turns out to be simplest for your
development group.
See 4G.2 for some specific ideas about how to use the modules file.
Last modified: _11/12/1997_
12. What does "merge" mean?
A merge is a way of combining changes made in two independent copies
of a common starting file. Checking out an RCS revision produces a
file, so for the purposes of a merge "file" and "revision" are
equivalent. So, we can say there are always three "files" involved in
a merge:
The original, starting, "base" or "branch point" file.
A copy of the base file modified in one way.
Another copy of the base file modified in a different way.
Humans aren't very good at handling three things at once, so the
terminology dealing with merges can become strained. One way to think
about it is that all merges are performed by inserting the difference
between a base revision and a later revision (committed by someone
else) into your working file. Both the "later" revision and your
working file are presumed to have started life as a copy of the "base"
revision.
In CVS, there are three main types of "merge":
The "update" command automatically merges revisions committed by
others into your working file. In this case, the three files involved
in the merge are:
Base: The revision you originally checked out. Later: A revision
committed onto the current branch after you checked out the Base
revision. Working: Your working file. The one lying in the working
directory containing changes you have made.
The "update -j <branch_tag> {optional files}" command merges changes
made on the given branch into your working files, which is presumed to
be on the Main line of development.
See 4C.6
The "update -j <rev> -j <rev> {optional files}" command merges the
difference between two specified revisions into files in your working
directory. The two revisions <rev> are usually on the same branch and,
when updating multiple files, they are most useful when they are Tag
names rather than numeric revisions.
See 4C.7
Last modified: _6/13/1997_
Category: /What_is_CVS_/What_is_CVS_Whats_it/
" + What is CVS? What's it for? Why CVS?"
1. What does CVS stand for? Can you describe it in one sentence?
"CVS" is an acronym for the "Concurrent Versions System".
CVS is a "Source Control" or "Revision Control" tool designed to keep
track of source changes made by groups of developers working on the
same files, allowing them to stay in sync with each other as each
individual chooses.
Last modified: _6/13/1997_
2. What is CVS for? What does it do for me?
CVS is used to keep track of collections of files in a shared
directory called "The Repository". Each collection of files can be
given a "module" name, which is used to "checkout" that collection.
After checkout, files can be modified (using your favorite editor),
"committed" back into the Repository and compared against earlier
revisions. Collections of files can be "tagged" with a symbolic name
for later retrieval.
You can add new files, remove files you no longer want, ask for
information about sets of files in three different ways, produce patch
"diffs" from a base revision and merge the committed changes of other
developers into your working files.
Last modified: _6/13/1997_
3. How does CVS work?
CVS saves its version-control information in RCS files stored in a
directory hierarchy, called the Repository, which is separate from the
user's working directory.
Files in the Repository are stored in a format dictated by the RCS
commands CVS uses to do much of its real work. RCS files are standard
byte-stream files with an internal format described by keywords stored
in the files themselves.
To begin work, you execute a "checkout" command, handing it a module
name or directory path (relative to the $CVSROOT variable) you want to
work on. CVS copies the latest revision of each file in the specified
module or directory out of the Repository and into a directory tree
created in your current directory. You may specify a particular branch
to work on by symbolic name if you don't want to work on the default
(main or trunk) branch.
You may then modify files in the new directory tree, build them into
output files and test the results. When you want to make your changes
available to other developers, you "commit" them back into the
Repository.
Other developers can check out the same files at the same time. To
merge the committed work of others into your working files you use the
"update" command. When your merged files build and test correctly, you
may commit the merged result. This method is referred to as
"copy-modify-merge", which does not require locks on the source files.
At any time, usually at some milestone, you can "tag" the committed
files, producing a symbolic name that can be handed to a future
"checkout" command. A special form of "tag" produces a branch in
development, as usually happens at "release" time.
When you no longer plan to modify or refer to your local copy of the
files, they can be removed.
Last modified: _6/13/1997_
4. What is CVS useful for?
CVS is intended to handle source control for files in three major
situations:
Multiple developers working on the same files.
The major advantage of using CVS over the simpler tools like RCS or
SCCS is that it allows multiple developers to work on the same sources
at the same time.
The shared Repository provides a rendezvous for committed sources that
allows developers a fair amount of flexibility in how often to publish
(via the "commit" command) changes or include work committed by others
(via the "update" command).
Tracking a stream of releases from a source vendor.
If you are making changes to sources distributed by someone else, the
CVS feature, called the Vendor Branch, allows you to combine local
modifications with repeated vendor releases.
I have found this most useful when dealing with sources from three
major classes of source vendor:
Large companies who send you tapes full of the latest release (e.g.
Unix OS vendors, database companies).
Public Domain software which *always* requires work.
Pseudo-Public sources which may require work. (e.g. GNU programs, X,
CVS itself, etc.)
Branching development.
Aside from the "Vendor Branch", there are three kinds of "branches in
development" that CVS can support:
Your working directory can be treated as a private branch.
A Development branch can be shared by one or more developers.
At release time, a branch is usually created for bug fixes.
(See 1D.9 and Section 4C for more info on branches.)
CVS's branch support is a bit primitive, but it was designed to allow
you to create branches, work on them for while and merge them back
into the main line of development. You should also be able to merge
work performed on the main branch into the branch you are working on.
Arbitrary sharing and merging between branches is not currently
supported.
Last modified: _6/13/1997_
5. What is CVS *not* useful for?
CVS is not a build system.
Though the structure of your Repository and modules file interact with
your build system (e.g. a tree of Makefiles), they are essentially
independent.
CVS does not dictate how you build anything. It merely stores files
for retrieval in a tree structure you devise.
CVS does not dictate how to use disk space in the checked out working
directories. If you require your Makefiles or build procedures to know
the relative positions of everything else, you wind up requiring the
entire Repository to be checked out. That's simply bad planning.
If you modularize your work, and construct a build system that will
share files (via links, mounts, VPATH in Makefiles, etc.), you can
arrange your disk usage however you like.
But you have to remember that *any* such system is a lot of work to
construct and maintain. CVS does not address the issues involved. You
must use your brain and a collection of other tools to provide a build
scheme to match your plans.
Of course, you should use CVS to maintain the tools created to support
such a build system (scripts, Makefiles, etc).
CVS is not a substitute for management.
You and your project leaders are expected to plan what you are doing.
Everyone involved must be aware of schedules, merge points, branch
names, release dates and the range of procedures needed to build
products. (If you produce it and someone else uses it, it is a
product.) CVS can't cover for a failure to manage your project.
CVS is an instrument for making sources dance to your tune. But you
are the piper and the composer. No instrument plays itself or writes
its own music.
CVS is not a substitute for developer communication.
When faced with conflicts within a single file, most developers manage
to resolve them without too much effort. But a more general definition
of "conflict" includes problems too difficult to solve without
communication between developers.
CVS cannot determine when simultaneous changes within a single file,
or across a whole collection of files, will logically conflict with
one another. Its concept of a "conflict" is purely textual, arising
when two changes to the same base file are near enough to spook the
merge command into dropping conflict markers into the merged file.
CVS is not capable of figuring out distributed conflicts in program
logic. For example, if you change the arguments to function X defined
in file A and, at the same time, edit file B, adding new calls to
function X using the old arguments. You are outside the realm of CVS's
competence.
Acquire the habit of reading specs and talking to your peers.
CVS is not a configuration management system.
CVS is a source control system. The phrase "configuration management"
is a marketing term, not an industry-recognized set of functions.
A true "configuration management system" would contain elements of the
following:
* Source control.
* Dependency tracking.
* Build systems (i.e. What to build and how to find
things during a build. What is shared? What is local?)
* Bug tracking.
* Automated Testing procedures.
* Release Engineering documentation and procedures.
* Tape Construction.
* Customer Installation.
* A way for users to run different versions of the same
software on the same host at the same time.
CVS provides only the first.
Last modified: _6/13/1997_
Category: /What_is_CVS_/Where_do_I_find_CVS_/
" + Where do I find CVS? Where can I find Help?"
1. How do I get more information about CVS?
The first thing I would do is to read the Info file that comes with
the CVS sources under "doc". You can format and read the cvs.texinfo
file in two ways: 1. Use TeX to format it and a "dvips" command to
print it and 2. Install the cvs.info files that are created by the
Makefile and read them online using the Emacs "info-mode" or a
stand-alone "info" reader.
Then I'd run "cvsinit" to set up a Repository and read the man page
while trying out the commands.
Type "cvs -H" for general help or "cvs -H command" for
command-specific help.
For background, you can read the original CVS paper (in the source
tree, under "doc"). It describes the purpose of CVS and some of how it
was designed. Note that the emphasis of the document (especially on
multiple vendors providing the same sources) is somewhat out of date.
For more detailed information about "internals", read the man pages
for RCS. If you are a programmer, you can also read the source code to
CVS.
Other information and tutorials may be available in the "doc"
directory of the FTP archive described below.
For current information, and a fair amount of detail, join the
info-cvs mailing list described below.
Last modified: _6/13/1997_
2. Is there an archive of CVS material?
An anonymous FTP area has been set up. It contains many of the CVS
files you might want, including extra documentation, patches and a
copy of the latest release.
ftp ftp.delos.com
>>> User: anonymous
>>> Passwd: <Your Internet address>
cd /pub/cvs
get README
get Index
The README has more (and more up-to-date) information. The Index
contains a terse list of what is in the archive.
A WWW home page is also available at http://www.delos.com/cvs.
Last modified: _6/13/1997_
3. How do I get files out of the archive if I don't have FTP?
Use one of the FTP<->Email servers. These are the ones I've been told
about:
FTPMAIL service is available from the same host as the FTP server
described above. Send mail to "ftpmail@delos.com" containing "help" in
the body of the message. For example, on most Unix systems, you can
type:
echo help | Mail ftpmail@delos.com
The FTPMAIL server will respond with a document describing how to use
the server. If the "Mail" command doesn't exist on your system, try
"mailx", "/usr/ucb/mail" or "/bin/mail".
If you are on BITNET, use Princeton's BITFTP server. Type
echo 'send help' | Mail bitftp@pucc.princeton.edu
(It is likely that only BITNET addresses can use this one.)
Other possibilities I've heard of from the net: (Try the one closest
to you.)
ftpmail@decwrl.dec.com ftpmail@sunsite.unc.edu ftpmail@cs.arizona.edu
ftpmail@cs.uow.edu.au ftpmail@doc.ic.ac.uk
Last modified: _6/13/1997_
4. How do I get a copy of the latest version of CVS?
The latest released version of CVS and all the programs it depends on
should be available through anonymous FTP on any FSF archive. The main
FSF archive is at "prep.ai.mit.edu". There are mirrors of the FSF
archive on UUNET and other large Internet sites.
Program(s) Suggested revision
----------- -----------------------
CVS 1.5
RCS 5.7 (latest version available today)
GNU diff 2.7 (or later) [contained in diffutils-2.7]
GDBM 1.5 (or later) [optional]
The GNU version of diff is suggested by both the RCS and CVS
configuration instructions because it works better than the standard
version.
It is a good idea not to accept the versions of CVS, RCS or diff you
find lying on your system unless you have checked out their
provenance. Using inconsistent collections of tools can cause you more
trouble than you can probably afford.
The FTP archive mentioned above should contain the latest official
release of CVS, some official and unofficial patches and possibly
complete patched versions of CVS in use somewhere.
Last modified: _6/13/1997_
5. Is there a mailing list devoted to CVS? How do I find it?
An Internet mailing list named "info-cvs" grew out of the private
mailing list used by the CVS 1.3 alpha testers in early 1992.
Throughout 1994, the list received an average of 100 messages per
month.
You can add yourself to the mailing list by sending an Email message
to:
info-cvs-request@prep.ai.mit.edu
(Don't forget the "-request" or you'll send a message to the whole
list, some of whom are capable of remote execution.)
Mail to the whole list should be sent to:
info-cvs@prep.ai.mit.edu
An archive of the mailing list is maintained in the FTP archive
mentioned above.
Last modified: _6/13/1997_
6. What happened to the CVS Usenet newsgroup I heard about?
A Usenet newsgroup named "gnu.cvs.info" was announced in April
1993, with an expected creation date of August, 1993. However,
nothing came of this.
If you want to discuss CVS on usenet, the correct group is
comp.software.config-mgmt (which also covers other configuration
management systems). Someday it might be possible to create a
comp.software.config-mgmt.cvs, but only if there is sufficient
CVS traffic on comp.software.config-mgmt.
kingdon@cyclic.com
Last modified: _9/6/1997_
_________________________________________________________________
[Add an answer to this category]
[Category /]
_________________________________________________________________
_Search the FAQ-O-Matic:_ ____________________ ______
[matching all words]
Or look for questions modified in the last: [7.] ____
_________________________________________________________________
The FAQ-O-Matic lives at http://gille.loria.fr:7000/cgi-bin/faqomatic.
The code was written by Jon Howell, and the content by folks from all
over the web.