=head1 NAME
common::sense - save a tree AND a kitten, use common::sense!
=head1 SYNOPSIS
use common::sense;
# supposed to be the same, with much lower memory usage, as:
#
# use utf8;
# use strict qw(vars subs);
# use feature qw(say state switch);
# no warnings;
# use warnings qw(FATAL closed threads internal debugging pack substr malloc
# portable prototype inplace io pipe unpack regexp
# deprecated exiting glob digit printf layer
# reserved parenthesis taint closure semicolon);
# no warnings qw(exec newline unopened);
=head1 DESCRIPTION
This module implements some sane defaults for Perl programs, as defined by
two typical (or not so typical - use your common sense) specimens of Perl
coders. In fact, after working out details on which warnings and strict
modes to enable and make fatal, we found that we (and our code written so
far, and others) fully agree on every option, even though we never used
warnings before, so it seems this module indeed reflects a "common" sense
among some long-time Perl coders.
The basic philosophy behind the choices made in common::sense can be
summarised as: "enforcing strict policies to catch as many bugs as
possible, while at the same time, not limiting the expressive power
available to the programmer".
Two typical examples of this philosophy are uninitialised and malloc
warnings:
C<undef> is a well-defined feature of perl, and enabling warnings for
using it rarely catches any bugs, but considerably limits you in what you
can do, so uninitialised warnings are disabled.
Freeing something twice on the C level is a serious bug, usually causing
memory corruption. It often leads to side effects much later in the
program and there are no advantages to not reporting this, so malloc
warnings are fatal by default.
What follows is a more thorough discussion of what this module does,
and why it does it, and what the advantages (and disadvantages) of this
approach are.
=over 4
=item use utf8
While it's not common sense to write your programs in UTF-8, it's quickly
becoming the most common encoding, and the most convenient encoding
available (you can do really nice quoting tricks...). Experience has shown
that our programs were either all pure ascii or utf-8, both of which will
stay the same.
There are few drawbacks to enabling UTF-8 source code by default (mainly
some speed hits due to bugs in older versions of perl), so this module
enables UTF-8 source code encoding by default.
=item use strict qw(subs vars)
Using C<use strict> is definitely common sense, but C<use strict
'refs'> definitely overshoots its usefulness. After almost two
decades of Perl hacking, we decided that it does more harm than being
useful. Specifically, constructs like these:
@{ $var->[0] }
Must be written like this (or similarly), when C<use strict 'refs'> is in
scope, and C<$var> can legally be C<undef>:
@{ $var->[0] || [] }
This is annoying, and doesn't shield against obvious mistakes such as
using C<"">, so one would even have to write (at least for the time
being):
@{ defined $var->[0] ? $var->[0] : [] }
... which nobody with a bit of common sense would consider
writing: clear code is clearly something else.
Curiously enough, sometimes perl is not so strict, as this works even with
C<use strict> in scope:
for (@{ $var->[0] }) { ...
If that isn't hypocrisy! And all that from a mere program!
=item use feature qw(say state given)
We found it annoying that we always have to enable extra features. If
something breaks because it didn't anticipate future changes, so be
it. 5.10 broke almost all our XS modules and nobody cared either (or at
least I know of nobody who really complained about gratuitous changes -
as opposed to bugs).
Few modules that are not actively maintained work with newer versions of
Perl, regardless of use feature or not, so a new major perl release means
changes to many modules - new keywords are just the tip of the iceberg.
If your code isn't alive, it's dead, Jim - be an active maintainer.
But nobody forces you to use those extra features in modules meant for
older versions of perl - common::sense of course works there as well.
There is also an important other mode where having additional features by
default is useful: commandline hacks and internal use scripts: See "much
reduced typing", below.
=item no warnings, but a lot of new errors
Ah, the dreaded warnings. Even worse, the horribly dreaded C<-w>
switch: Even though we don't care if other people use warnings (and
certainly there are useful ones), a lot of warnings simply go against the
spirit of Perl.
Most prominently, the warnings related to C<undef>. There is nothing wrong
with C<undef>: it has well-defined semantics, it is useful, and spitting
out warnings you never asked for is just evil.
The result was that every one of our modules did C<no warnings> in the
past, to avoid somebody accidentally using and forcing his bad standards
on our code. Of course, this switched off all warnings, even the useful
ones. Not a good situation. Really, the C<-w> switch should only enable
warnings for the main program only.
Funnily enough, L<perllexwarn> explicitly mentions C<-w> (and not in a
favourable way, calling it outright "wrong"), but standard utilities, such
as L<prove>, or MakeMaker when running C<make test>, still enable them
blindly.
For version 2 of common::sense, we finally sat down a few hours and went
through I<every single warning message>, identifiying - according to
common sense - all the useful ones.
This resulted in the rather impressive list in the SYNOPSIS. When we
weren't sure, we didn't include the warning, so the list might grow in
the future (we might have made a mistake, too, so the list might shrink
as well).
Note the presence of C<FATAL> in the list: we do not think that the
conditions caught by these warnings are worthy of a warning, we I<insist>
that they are worthy of I<stopping> your program, I<instantly>. They are
I<bugs>!
Therefore we consider C<common::sense> to be much stricter than C<use
warnings>, which is good if you are into strict things (we are not,
actually, but these things tend to be subjective).
After deciding on the list, we ran the module against all of our code that
uses C<common::sense> (that is almost all of our code), and found only one
occurence where one of them caused a problem: one of elmex's (unreleased)
modules contained:
$fmt =~ s/([^\s\[]*)\[( [^\]]* )\]/\x0$1\x1$2\x0/xgo;
We quickly agreed that indeed the code should be changed, even though it
happened to do the right thing when the warning was switched off.
=item much reduced typing
Especially with version 2.0 of common::sense, the amount of boilerplate
code you need to add to gte I<this> policy is daunting. Nobody would write
this out in throwaway scripts, commandline hacks or in quick internal-use
scripts.
By using common::sense you get a defined set of policies (ours, but maybe
yours, too, if you accept them), and they are easy to apply to your
scripts: typing C<use common::sense;> is even shorter than C<use warnings;
use strict; use feature ...>.
And you can immediately use the features of your installed perl, which
is more difficult in code you release, but not usually an issue for
internal-use code (downgrades of your production perl should be rare,
right?).
=item mucho reduced memory usage
Just using all those pragmas mentioned in the SYNOPSIS together wastes
<blink>I<< B<776> kilobytes >></blink> of precious memory in my perl, for
I<every single perl process using our code>, which on our machines, is a
lot. In comparison, this module only uses I<< B<four> >> kilobytes (I even
had to write it out so it looks like more) of memory on the same platform.
The money/time/effort/electricity invested in these gigabytes (probably
petabytes globally!) of wasted memory could easily save 42 trees, and a
kitten!
Unfortunately, until everybods applies more common sense, there will still
often be modules that pull in the monster pragmas. But one can hope...
=cut
package common::sense;
our $VERSION = '3.0';
sub import {
${^WARNING_BITS} ^= ${^WARNING_BITS} ^ "\xfc\x3f\x33\x00\x0f\xf3\xcf\xc0\xf3\xfc\x33\x00";
$^H |= 0x00800600;
$^H{feature_switch} =
$^H{feature_say} =
$^H{feature_state} = 1;
}
1;
=back
=head1 THERE IS NO 'no common::sense'!!!! !!!! !!
This module doesn't offer an unimport. First of all, it wastes even more
memory, second, and more importantly, who with even a bit of common sense
would want no common sense?
=head1 STABILITY AND FUTURE VERSIONS
Future versions might change just about everything in this module. We
might test our modules and upload new ones working with newer versions of
this module, and leave you standing in the rain because we didn't tell
you. In fact, we did so when switching from 1.0 to 2.0, which enabled gobs
of warnings, and made them FATAL on top.
Maybe we will load some nifty modules that try to emulate C<say> or so
with perls older than 5.10 (this module, of course, should work with older
perl versions - supporting 5.8 for example is just common sense at this
time. Maybe not in the future, but of course you can trust our common
sense to be consistent with, uhm, our opinion).
=head1 WHAT OTHER PEOPLE HAD TO SAY ABOUT THIS MODULE
apeiron
"... wow"
"I hope common::sense is a joke."
crab
"i wonder how it would be if joerg schilling wrote perl modules."
Adam Kennedy
"Very interesting, efficient, and potentially something I'd use all the time."
[...]
"So no common::sense for me, alas."
H.Merijn Brand
"Just one more reason to drop JSON::XS from my distribution list"
Pista Palo
"Something in short supply these days..."
Steffen Schwigon
"This module is quite for sure *not* just a repetition of all the other
'use strict, use warnings'-approaches, and it's also not the opposite.
[...] And for its chosen middle-way it's also not the worst name ever.
And everything is documented."
BKB
"[Deleted - thanks to Steffen Schwigon for pointing out this review was
in error.]"
Somni
"the arrogance of the guy"
"I swear he tacked somenoe else's name onto the module
just so he could use the royal 'we' in the documentation"
Anonymous Monk
"You just gotta love this thing, its got META.json!!!"
dngor
"Heh. '"<elmex at ta-sa.org>"' The quotes are semantic
distancing from that e-mail address."
Jerad Pierce
"Awful name (not a proper pragma), and the SYNOPSIS doesn't tell you
anything either. Nor is it clear what features have to do with "common
sense" or discipline."
acme
"THERE IS NO 'no common::sense'!!!! !!!! !!"
apeiron (meta-comment about us commenting^Wquoting his comment)
"How about quoting this: get a clue, you fucktarded amoeba."
quanth
"common sense is beautiful, json::xs is fast, Anyevent, EV are fast and
furious. I love mlehmannware ;)"
=head1 FREQUQNTLY ASKED QUESTIONS
Or frequently-come-up confusions.
=over 4
=item Is this module meant to be serious?
Yes, we would have put it under the C<Acme::> namespace otherwise.
=item But the manpage is written in a funny/stupid/... way?
This was meant to make it clear that our common sense is a subjective
thing and other people can use their own notions, taking the steam out
of anybody who might be offended (as some people are always offended no
matter what you do).
This was a failure.
But we hope the manpage still is somewhat entertaining even though it
explains boring rationale.
=item Why do you impose your conventions on my code?
For some reason people keep thinking that C<common::sense> imposes
process-wide limits, even though the SYNOPSIS makes it clear that it works
like other similar modules - only on the scope that uses them.
So, no, we don't - nobody is forced to use this module, and using a module
that relies on common::sense does not impose anything on you.
=item Why do you think only your notion of common::sense is valid?
Well, we don't, and have clearly written this in the documentation to
every single release. We were just faster than anybody else w.r.t. to
grabbing the namespace.
=item But everybody knows that you have to use strict and use warnings,
why do you disable them?
Well, we don't do this either - we selectively disagree with the
usefulness of some warnings over others. This module is aimed at
experienced Perl programmers, not people migrating from other languages
who might be surprised about stuff such as C<undef>. On the other hand,
this does not exclude the usefulness of this module for total newbies, due
to its strictness in enforcing policy, while at the same time not limiting
the expresive power of perl.
This module is considerably I<more> strict than the canonical C<use
strict; use warnings>, as it makes all its warnings fatal in nature, so
you can not get away with as many things as with the canonical approach.
This was not implemented in version 1.0 because of the daunting number
of warning categories and the difficulty in getting exactly the set of
warnings you wish (i.e. look at the SYNOPSIS in how complicated it is to
get a specific set of warnings - it is not reasonable to put this into
every module, the maintainance effort would be enourmous).
=item But many modules C<use strict> or C<use warnings>, so the memory
savings do not apply?
I am suddenly so sad.
But yes, that's true. Fortunately C<common::sense> still uses only a
miniscule amount of RAM.
=item But it adds another dependency to your modules!
It's a fact, yeah. But it's trivial to install, most popular modules have
many more dependencies and we consider dependencies a good thing - it
leads to better APIs, more thought about interworking of modules and so
on.
=item Why do you use JSON and not YAML for your META.yml?
This is not true - YAML supports a large subset of JSON, and this subset
is what META.yml is written in, so it would be correct to say "the
META.yml is written in a common subset of YAML and JSON".
The META.yml follows the YAML, JSON and META.yml specifications, and is
correctly parsed by CPAN, so if you have trouble with it, the problem is
likely on your side.
=item But! But!
Yeah, we know.
=back
=head1 AUTHOR
Marc Lehmann <schmorp@schmorp.de>
http://home.schmorp.de/
Robin Redeker, "<elmex at ta-sa.org>".
=cut