4/9/2001 -------- [bash-2.05 released] 4/10 ---- redir.c - check return value of fclose() in write_here_document() for error returns; don't just rely on fwrite() failing support/bashbug.sh - set TMPDIR to /tmp if it's null or unset - use $TMPDIR in the TEMP tempfile name template - fixed the call to `mktemp', if it exists, to make it more portable jobs.c - if WCONTINUED is not defined, define it to 0 and add a define for WIFCONTINUED(wstatus) which expands to 0 - add WCONTINUED to the flags passed to waitpid(2) in waitchld() - don't increment children_exited if waitpid's status is WIFCONTINUED, since we don't want to call a SIGCHLD trap handler in this case - in waitchld(), we set child->running to 1 if WIFCONTINUED(status) is non-zero - make sure pretty_print_job doesn't check for the core dump bit if the process has been continued; it's only valid if the job is dead - in set_job_status_and_cleanup, set the job to JRUNNING if job_state is non-zero and the job was previously marked as JSTOPPED configure.in - add -DBROKEN_DIRENT_D_INO to interix LOCAL_CFLAGS lib/glob/glob.c - if BROKEN_DIRENT_D_INO is defined, define REAL_DIR_ENTRY to 1 jobs.c - in kill_pid, we only need to block and unblock SIGCHLD if the `group' argument is non-zero, since otherwise we just call `kill' on the pid argument version.c - update copyright date to 2001 bashline.c - prog_complete_return needs to take a `const char *' as its first argument - history_completion_generator needs to take a `const char *' as its first argument, and `text' needs to be a `const char *' 4/11 ---- redir.c - fixed a weird typo in redir_special_open, case RF_DEVFD, added call to all_digits before call to legal_number - fixed do_redirection_internal to call legal_number instead of atol(3) when translating r_duplicating_{in,out}put_word, so it handles overflow better - produce an error message in redirection_error for out-of-range file descriptors - change allocation strategy in redirection_error so we don't have to malloc scratch memory if redirection_expand() fails jobs.h - added defines for `running' member of a struct process general.c - fix legal_number to return 0 when strtol(3) reports overflow or underflow parse.y - changed read_token_word to call legal_number instead of atoi(3) input.c - return -1/EBADF from close_buffered_fd if fd is < 0 command.h - fixed bogus comment about IS_DESCRIPTOR in description of the REDIRECTEE struct print_cmd.c - change cprintf's 'd' modifier code to display negative numbers as an out-of-range value. We can do this only because the only use of %d is to output file descriptor numbers in redirections support/mksignames.c - need to include config.h to get a possible value for UNUSABLE_RT_SIGNALS 4/16 ---- lib/readline/doc/rluser.texinfo - corrected a small error in one description of M-DEL 4/17 ---- stringlib.c - need to initialize `ind' before calls to RESIZE_MALLOCED_BUFFER in strcreplace() support/bashversion.c - new file, prints bash version information Makefile.in - rules for building bashversion and linking it to version.o 4/24 ---- conftypes.h - new file with HOSTTYPE, OSTYPE, MACHTYPE, etc. defines from variables.h variables.h, version.c - include conftypes.h patchlevel.h - new file, contains define for PATCHLEVEL. Doing away with the old scheme of having the information in configure.in version.c - include patchlevel.h Makefile.in - run bashversion -p to find patch level rather than have configure substitute in a value - pass -S ${top_srcdir} to support/mkversion.sh support/mkversion.sh - don't put PATCHLEVEL define into version.h, but accept and ignore a -p option - take a new -S srcdir option - find the patch level by parsing it out of patchlevel.h configure.in - hard-code BASHVERS assignment instead of reading it from a file - remove BASHPATCH; don't substitute it _distribution,_patchlevel - removed 4/26 ---- shell.c - call init_noninteractive() in open_shell_script if forced_interactive is non-zero (the shell was started with -i) and fd_is_tty is 0 (the script file is a real file, not something like /dev/stdin), since it wasn't done earlier builtins/printf.def - change for POSIX.2 compliance when conversion errors are encountered when processing %d, %u, and floating point conversion operators (print a warning message, return the value accumulated at the time of the error -- which is always 0 -- and exit with a non-zero status) command.h - added CMD_COMMAND_BUILTIN for use by the `command' builtin and the code in execute_cmd.c builtins/command.def - add CMD_COMMAND_BUILTIN to the created command's flags 5/1 --- configure.in - add call to AC_C_CONST to test `const' compiler behavior - add call to AC_C_INLINE to test `inline' compiler behavior - add call to AC_C_STRINGIZE to test cpp #x stringizing operator config.h.in - add `#undef const' for configure to substitute - add `#undef inline' for configure to substitute - add `#undef HAVE_STRINGIZE' for configure to substitute include/stdc.h - remove code that defines or undefines `const' and `inline' - change the __STRING macro to be defined depending on the value of HAVE_STRINGIZE lib/malloc/malloc.c - change the __STRING macro to be defined depending on the value of HAVE_STRINGIZE lib/readline/{readline,rlprivate}.h - moved rl_get_termcap to readline.h, making it a public function lib/readline/readline.h - new #define, RL_READLINE_VERSION, hex-encoded library version number, currently set to 0x0402 - new public int variable, rl_readline_version lib/readline/readline.c - #define RL_READLINE_VERSION if it is not already defined (which it should be in readline.h) - initialize rl_readline_version to RL_READLINE_VERSION lib/readline/doc/rltech.texinfo - documented rl_get_termcap - documented rl_readline_version jobs.c - job_exit_status should return an int, not a WAIT (undetected before because on most POSIX-like systems a WAIT is really an int) builtins/evalfile.c - added FEVAL_REGFILE (file must be a regular file) to accepted _evalfile flags - fc_execute_file() adds FEVAL_REGFILE to _evalfile flags. This means that startup files and files read with `.' no longer need to be regular files 5/2 --- lib/termcap/Makefile.in - fix target for installed termcap library (normally unused) lib/tilde/Makefile.in - fix install target to install in $(libdir) (normally unused) Makefile.in - don't make $(man3dir) since there's nothing installed there Makefile.in,doc/Makefile.in - change `man1ext' to `.1', `man3ext' to `.3' - change appropriate install targets to use new values of man[13]ext - use `test ...' instead of `[...]' - add support for DESTDIR root installation prefix, for package building (installdirs, install, install-strip, uninstall targets) builtins/common.c - new function int get_exitstat(WORD_LIST *list) returns an eight-bit exit status value for use in return, exit, logout builtins builtins/common.h - extern declaration for get_exitstat() builtins/{exit,return}.def - call get_exitstat where appropriate builtins/printf.def - add support for "'" flag character as posix 1003.2-200x d6 says - fix core dump when user-supplied field width or precision is 0 - fix to printstr() to handle zero-length precision with `%b' format specifier (printf '%.0b-%.0s\n' foo bar) - fix to printstr() to treat a negative field width as a positive field width with left-justification - fix to mklong to avoid static buffers, which can always be overrun by someone sufficiently motivated bashline.c - change var in add_host_name to type `size_t' for passing to xrealloc 5/3 --- execute_cmd.c - change restore_signal_mask to accept a sigset_t *, since a sigset_t may not fit into a pointer, change call unwind_prot.c - use a union UWP in restore_variable when restoring a variable whose size is the same as sizeof(int), the reverse of the method used to store it in unwind_protect_int builtins/printf.def - use a #define LENMODS containing the length modifiers instead of testing against each possible modifier character, save any mod character found - add support for ISO C99 length specifiers `j', `t', and `z' - if `L' modifier is supplied with a floating point conversion char, pass a `long double' to printf if HAVE_LONG_DOUBLE is defined configure.in,config.h.in - call AC_C_LONG_DOUBLE to check for `long double'; define HAVE_LONG_DOUBLE if supported bashline.c - fix an inadvertantly-unclosed comment in attempt_shell_completion - make set_saved_history return a value - make dynamic_complete_history return a useful value {make_cmd,execute_cmd,shell,subst,trap,variables,input,unwind_prot,test, pcomplete}.c - removed some declared-but-unused variables builtins/{cd,enable,fc,set,setattr,type,umask,printf,complete}.def - removed some declared-but-unused variables lib/sh/{zread,netopen}.c - removed some declared-but-unused variables execute_cmd.c - in execute_arith_command, use a long variable to hold the result of evalexp(), since that's what it returns builtins/evalstring.c - make cat_file return -1 on a read or write error lib/sh/stringlib.c - make merge_stringlists() return the right value 5/7 --- pcomplete.c - remove typo that caused empty declaration (;;) parse.y - fix yyerror() to accept a single string argument; fix callers trap.c - cast pointer to long instead of int when printing message with internal_warning() in run_pending_traps() subst.c - fix process_substitute to handle stdin being closed test.c - change `while' to `if' in and() and or(), since the loop isn't actually performed -- there's an unconditional `return' in the loop body - check for integer overflow of arguments to `-t' lib/sh/netopen.c - change _getserv() to reject negative port/service numbers expr.c - fix strlong() to not convert the base specification from long to int before checking for overflow, since truncation on machines where sizeof(int) != sizeof(long) may mask errors builtins/{jobs,kill,wait}.def - use legal_number instead of atoi when converting strings to pid_t; check for numeric overflow input.c - fix for cygwin in b_fill_buffer -- off-by-one error when checking buffer for \r\n termination general.h - new #define INT_STRLEN_BOUND(t), computes max length of string representing integer value of type T, possibly including a sign character - include if it's present {execute_cmd,findcmd,test}.c - don't include , since general.h does it now {execute_cmd,lib/sh/itos,pcomplete,print_cmd,subst,variables}.c - use INT_STRLEN_BOUND instead of static array sizes when converting various strings to integer values shell.h - struct fd_bitmap now uses an `int' size, since it's bounded by the number of file descriptors, which must fit into an `int' execute_cmd.c - FD_BITMAP_DEFAULT_SIZE is now 32, not 32L - new_fd_bitmap takes an `int' size parameter, not a `long' execute_cmd.h - change prototype for new_fd_bitmap() test.c - fix test_stat to check for overflow when parsing the integer file descriptor number; return ENOENT instead of EBADF for files that are not open hashlib.c - don't discard the upper 32 bits of the random value, if present lib/readline/shell.c - use the same INT_STRLEN_BOUND mechanism to decide how much space to allocated in sh_set_lines_and_columns 5/8 --- aclocal.m4 - add check for libtinfo (termcap-specific portion of ncurses-5.2) to BASH_CHECK_LIB_TERMCAP - new macro, RL_LIB_READLINE_VERSION, checks version of installed readline library and (optionally) writes version #defines to config.h. Bash doesn't use the version defines configure.in - call RL_LIB_READLINE_VERSION instead of support/rlvers.sh execute_cmd.c - fix execute_shell_script and the WHITECHAR and STRINGCHAR macros to check array bounds before indexing into the sample string unwind_prot.[ch] - import new versions submitted by Paul Eggert with a couple of changes for backwards compatibility, so the rest of the source doesn't need to be changed yet jobs.c - use unwind_protect_var on last_made_pid in run_sigchld_trap builtins/bind.def - use unwind_protect_var on rl_outstream general.c - rework print_rlimtype to use INT_STRLEN_BOUND and handle the most negative number correctly expr.c - `tokval' should have been a `long', since all arithmetic is done as longs builtins/history.def - consolidate tests for valid history position in one block to avoid duplicate code and strings builtins/ulimit.def - fix check for overflow when setting limit to work when int is 32 bits and RLIMTYPE is 64 lib/sh/tmpfile.c - don't truncate the result of time(3) to int; just use time_t, since it's being assigned to an `unsigned long' mailcheck.c - use legal_number instead of atoi in time_to_check_mail() to catch more numeric errors; consolidate error checking in one block - last_time_mail_checked should be a time_t 5/9 --- builtins/set.def - recognize `set [-+]o nolog' if HISTORY is defined bashline.c - new variable `dont_save_function_defs', set by `set -o nolog'; currently ignored command.h - the `dest' member of a REDIRECTEE is now an `int' parse.y,redir.c - changed uses of `redir.test' (where redir is a REDIRECTEE) since it's now an int lib/readline/rlstdc.h - don't mess around with `const', rely on configure to supply a proper definition if the compiler doesn't support it lib/tilde/tilde.h - include if HAVE_CONFIG_H is defined - don't mess around with `const', rely on configure builtins/shopt.def - new read-only `shopt' option, login_shell, non-zero if shell is a login shell (as decided by shell.c) - new function set_login_shell(), sets shopt private value of login_shell builtins/common.h - new extern declaration for set_login_shell shell.c - call set_login_shell after setting value of login_shell (in main() and set_shell_name()) parse.y - added new `\A' prompt string escape sequence: time in 24-hour HH:MM format configure.in, config.h.in - check for , define HAVE_GRP_H if found builtins/complete.def - add new `-A group/-g' option to complete group names pcomplete.h - new define for CA_GROUP, used with group name completion pcomplete.c - add code to support CA_GROUP group name completion bashline.c - new function, bash_groupname_completion_function(), supports programmable completion of group names bashline.h - extern declaration for bash_groupname_completion_function lib/readline/bind.c - new inputrc variable, `match-hidden-files', controls completion matching files beginning with a `.' (on Unix) lib/readline/complete.c - new variable, _rl_match_hidden_files, mirrors `match-hidden-files' inputrc variable lib/readline/rlprivate.h - extern declaration for _rl_match_hidden_files builtins/hash.def - new `-t' option to list hash values for each filename argument builtins/read.def - alarm(3) takes an `unsigned int' argument, not int - check for arithmetic overflow with -t and -n options input.c - check for read error before doing \r\n translation on cygnus in b_fill_buffer - reset bp->b_used to 0 instead of leaving it at -1 on read error in b_fill_buffer builtins/shopt.def - new functions, shopt_setopt(name, mode) and shopt_listopt(name, mode) to give the rest of the shell an easy interface builtins/common.h - extern declarations for shopt_setopt and shopt_listopt shell.c - new invocation options -O and +O, to list or set/unset shopt options like +o/-o sets and unsets `set -o' options doc/{bash.1,bashref.texi} - document `set -o nolog' - document `login_shell' shopt option - document new `\A' prompt string escape sequence - document new `-t' option to `hash' - document new `[+-]O' invocation option doc/bashref.texi - add text to `Invoking Bash' section defining a login shell; text taken from man page doc/bash.1, lib/readline/doc/rluser.texinfo - documented new complete/compgen `-A group/-g' option lib/readline/doc/{rluser.texinfo,readline.3}, doc/bash.1 - documented new `match-hidden-files' inputrc variable 5/10 ---- configure.in - fix AC_CHECK_PROG(ar, ...) - add AC_CHECK_TYPE for ssize_t config.h.in - new #undef for ssize_t lib/sh/zread.c - int -> ssize_t fixes to mirror modern declarations of read and write - the `off' variable in zsyncfd should be an off_t since it computes a file offset - the local buffer `lbuf' is now char, since it's not nice to pass unsigned char * to read(2), and the values from it are assigned to a char anyway - lind and lused are now size_t, since they index into a buffer - set lused to 0 on read error lib/sh/zwrite.c - change second argument to type `char *', since ISO C says you have to pass a `char *' to `write' externs.h - fix extern declarations of zread, zread1, zreadc, and zwrite - prototype extern declaration of qsort_string_compare - add extern declaration for history_delimiting_chars() from parse.y input.h - b_used and b_inputp members ofr struct BSTREAM are now size_t builtins/evalstring.c - the number of chars read with zread in cat_file should be assigned to a variable of type ssize_t input.c - the number of chars read with zread in b_fill_buffer should be assigned to a variable of type ssize_t - `localbuf' is now type char[], since POSIX says you shouldn't pass unsigned char * to read(2) - in getc_with_restart(), use a variable of type unsigned char to get a value from the local buffer and return it - in ungetc_with_restart, explicitly return the character arg passed to avoid relying on localbuf being unsigned char subst.c - the number of chars read with zread in read_comsub should be assigned to a variable of type ssize_t mksyntax.c - instead of casting to unsigned char * in addcstr, use a variable of type unsigned char and let the compiler do the work parse.y - instead of casting to unsigned char * in yy_readline_get, use a variable of type unsigned char and let the compiler do the work - ditto for yy_string_get and shell_getc (cast to unsigned char) subst.c - instead of casting to unsigned char when assigning to ifscmap in expand_word_internal, use a variable of type unsigned char and let the compiler do the work lib/sh/strtrans.c - instead of casting to unsigned char in ansic_quote, use a variable of type unsigned char and let the compiler do the work builtins/evalstring.c - remove extern declarations for zwrite and run_trap_cleanup; they're in externs.h - prototype cat_file forward declaration Makefile.in - remove -I$(includedir) from INCLUDES and SUBDIR_INCLUDES aclocal.m4 - change RL_LIB_READLINE_VERSION to set RL_PREFIX, RL_LIBDIR, and RL_INCLUDEDIR to what it used to test the installed readline library version for use by the caller - change RL_LIB_READLINE_VERSION to not compute ac_cv_rl_prefix if the caller has already assigned it a value - rename _rl_prefix -> ac_cv_rl_prefix, _rl_libdir -> ac_cv_rl_libdir, _rl_includedir -> ac_cv_rl_includedir configure.in - change testing of whether to use the value of $opt_with_installed_readline to be != no, to allow the user to specify a prefix where the installed readline library may be found - if --with-installed-readline=PREFIX is supplied, set ac_cv_rl_prefix to PREFIX before calling RL_LIB_READLINE_VERSION - if --with-installed-readline[=PREFIX] is supplied, don't set RL_LIBDIR and RL_INCLUDEDIR; let RL_LIB_READLINE_VERSION take care of it, set RL_INCLUDE=-I${RL_INCLUDEDIR} - if --with-installed-readline[=PREFIX] is supplied, and we're linking with the history library, assign $RL_LIBDIR to HIST_LIBDIR so we use the same version of the installed readline and history libraries Makefile.in, builtins/Makefile.in - have configure substitute RL_INCLUDEDIR, set RL_INCLUDEDIR variable doc/bashref.texi - updated description of --with-installed-readline configure option general.c - moved QSFUNC typedef here from builtins/common.c {alias,bashline,variables,lib/sh/stringvec}.c - cast fourth argument to qsort to (QSFUNC *) alias.c - prototype forward declaration of qsort_alias_compare bashhist.c - include for extern declaration of glob_pattern_p - remove extern declaration of history_delimiting_chars; it's now in externs.h - prototype forward declarations of histignore_item_func, maybe_add_history, and bash_add_history bracecomp.c - remove extern declaration for sh_backslash_quote; it's in externs.h braces.c - remove extern declaration for extract_command_subst; it's in subst.h - prototype forward declarations for expand_amble, array_concat, and brace_gobbler error.c - prototype extern declaration of give_terminal_to, fix bad call {execute_cmd,expr,findcmd,jobs,mailcheck,nojobs,pcomplete,print_cmd,redir, shell}.c - prototype all static forward function declarations pcomplete.c - changed some function parameters to `const char *' to avoid discarding const qualifier make_cmd.c - make_bare_word, make_word_flags, and make_word now take a `const char *' string argument make_cmd.h - changed extern declarations for make_bare_word and make_word print_cmd.c - cprintf now takes a `const char *' as its first argument, like xprintf and printf - the conditional define for xprintf should have been HAVE_VPRINTF, not HAVE_VFPRINTF shell.c - in isnetconn(), the return value of sizeof() is size_t aclocal.m4 - add inclusion of stddef.h if STDC_HEADERS is defined to 1 in BASH_CHECK_TYPE configure.in - add a call to BASH_CHECK_TYPE for socklen_t (type of third argument to getpeername(2)) 5/11 ---- lib/readline/bind.c - make `useq' a char array to pass to rl_macro_bind in rl_parse_and_bind lib/readline/{{bind,isearch}.c,rlprivate.h} - _rl_isearch_terminators is now a char *, not unsigned char * {subst,variables,lib/sh/tmpfile}.c - dollar_dollar_pid is now a `pid_t' instead of `int' variables.c - sbrand() now takes an `unsigned long' to set the seed value - changed last_random_value to type int, since it's always between 0 and 32767 - use strtoul to convert the value in assign_random instead of atoi - take out casts in any arguments to sbrand() - take out cast to int in call to inttostr in set_ppid() subst.c - don't cast last_asynchronous_pid when passing to itos() {sig,subst}.c - prototype all static forward function declarations 5/14 ---- {test,trap,variables}.c - prototype all static forward function declarations variables.c - free_variable_hash_data() now takes a PTR_T, a `generic pointer' builtins/{alias,bind,break,cd,complete,declare,enable,exit,fc,fg_bg,help, history,jobs,pushd,read,set,trap,umask, - prototype all static forward function declarations builtins/read.def - reset_eol_delim now takes a `char *' arg, since that's what the unwind_protect functions pass it, and it ignores its arguments anyway lib/readline/{histsearch,input,kill,rltty,search,vi_mode}.c - prototype all static forward function declarations lib/tilde/tilde.c - prototype all static forward function declarations - tilde_find_prefix, tilde_find_suffix, isolate_tilde_prefix, and glue_prefix_and_suffix now take `const char *' arguments where appropriate configure.in,config.h.in - check for vsnprintf, define HAVE_VSNPRINTF if found lib/readline/display.c - use vsnprintf() in rl_message if it's available; if we don't, at least set the last character in msg_buf to 0 to avoid overrun -- we really can't do anything about overflow at this point. if it's available, this fixes buffer overflow problems in rl_message 5/15 ---- lib/readline/histexpand.c - in get_history_word_specifier, allow any character to terminate a `:first-' modifier, not just `:' and null. This is what csh appears to do. This allows things like `!:0- xyzzy' to replace the last argument with xyzzy 5/18 ---- configure.in, config.h.in - check for , define HAVE_STDINT_H if found - check for intmax_t in , define intmax_t as long if not found 5/21 ---- builtins/kill.def - change to use strerror() for error message when kill(2) fails aclocal.m4 - new macro, BASH_C_LONG_LONG, check for `long long' configure.in, config.h.in - call BASH_C_LONG_LONG, define HAVE_LONG_LONG if found lib/sh/snprintf.c - new file, with implementations of snprintf, vsnprintf, asprintf, and vasprintf, derived from inetutils version Makefile.in, lib/sh/Makefile.in - add snprintf.c/snprintf.o configure.in, config.h.in - add checks for snprintf, asprintf, vasprintf, with appropriate cpp defines lib/readline/{rldefs,xmalloc}.h, lib/readline/xmalloc.c - xmalloc and xrealloc now take `size_t' arguments, like their bash counterparts externs.h,lib/sh/itos.c - inttostr and itos now take `long' arguments - inttostr takes a `size_t' argument for the buffer size {expr,lib/malloc/malloc,variables,general}.c - fixed calls to itos() by removing casts, etc. subst.[ch] - get_dollar_var_value now takes a long, not an int - sub_append_number now takes a long, not an int subst.c - in parameter_brace_expand_word, use a long and legal_number to translate ${N}, to avoid overflow - in parameter_brace_expand_length, use a long and legal_number to translate ${#N}, to avoid overflow - in do_array_element_assignment, array_expand_index, array_value_internal, use arrayind_t instead of int - let verify_substring_values take long * arguments for the return value of evalexp() - pass long * arguments to verify_substring_values in parameter_brace_substring - parameter_brace_expand_length now returns `long' - parameter_brace_expand now uses a long variable for the return value of parameter_brace_expand_length - param_expand now uses a long variable for the return value from evalexp - array_length reference now returns an `arrayind_t', since it can return the num_elements member of an array, which is of type arrayind_t subst.h - array_expand_index now returns an `arrayind_t' array.[ch] - array_subrange now takes arrayind_t arguments, not `int' - dup_array_subrange now uses arrayind_t local variable to do array indexing - use long to print array indices in print_element variables.c - null_array_assign, assign_dirstack, bind_array_variable now take arrayind_t arguments as array indices - assign_array_var_from_word_list, assign_array_var_from_string, unbind_array_element now use arrayind_t local variables for array indexing variables.h - change extern declaration of bind_array_variable builtins/common.[ch] - get_numeric_arg now returns a `long', since it usually returns the value of legal_number() builtins/{shift,break}.def - use long variables for the return value of get_numeric_arg builtins/history.def - convert string argument to int only if it's in range builtins/pushd.def - set_dirstack_element and get_dirstack_element now take `long' index arguments - get_dirstack_index now takes a `long' index argument, since it's passed the converted value from legal_number lib/sh/timeval.c - in print_timeval, don't assume that the number of minutes fits into an int, since it's just seconds/60. lib/sh/clock.c - ditto for print_clock_t 5/22 ---- shell.c - since the -O option settings may possibly be overridden by the normal shell initialization or posix initialization, save the invocation options on an alist (with add_shopt_to_alist) and process them after basic initialization (with run_shopt_alist) 5/23 ---- trap.h - new define, BASH_NSIG, all system signals plus special bash traps trap.c, builtins/trap.def - use BASH_NSIG for array bounds and loops where appropriate trap.c - change decode_signal to disallow numeric signal numbers above NSIG -- this means you can only reference special traps like DEBUG by name - new SPECIAL_TRAP(s) macro to test whether s is one of the special bash traps (currently DEBUG and EXIT) - change reset_or_restore_signal_handlers so command substitution doesn't inherit the debug trap (like ksh93), and child processes don't have to rely on initialize_traps being run to get rid of any debug trap support/mksignames.c - add extra "ERR" signal name, value NSIG+1, allocate space for it and write it out in signal_names[] trap.h - new define: ERROR_TRAP == NSIG+1, change BASH_NSIG to NSIG+2 - extern declarations for set_error_trap, run_error_trap - new define: TRAP_STRING(s), expands to trap_list[s] if signal S is trapped and not ignored, NULL otherwise trap.c - add ERROR_TRAP to SPECIAL_TRAPS define - initialize ERROR_TRAP stuff in initialize_traps - new function: set_error_trap(command), sets the ERR trap string - new function: run_error_trap(command), runs the ERR trap string - set trap string for ERROR_TRAP to NULL in free_trap_strings - change reset_or_restore_signal_handlers so child processes don't inherit the ERR trap - add case to call run_error_trap in maybe_call_trap_handler execute_cmd.c - in execute_command_internal, keep track of ERR trap and call it if necessary - use TRAP_STRING to get the value of debug and error traps - in execute_function, arrange things so the ERR trap is not inherited by shell functions, and is saved and restored like the DEBUG trap doc/{bash.1,bashref.texi} - documented new ERR trap tests/{trap.{tests,right},trap2.sub,trap2a.sub} - added ERR trap tests subst.c - on machines without /dev/fd, change the named pipe fifo list to a list of structs containing pathname and proc information - change unlink_fifo_list to kill the proc in the fifo list with signal 0 and not remove the fifo if the proc is still alive. This should fix the problem on those backward systems without /dev/fd where fifos were removed when a job using process substitution was suspended 5/24 ---- examples/loadables/getconf.h - new file, with basic defines needed to make getconf work minimally on POSIX systems without the necessary definitions examples/loadables/getconf.c - replacement functions for confstr, sysconf, pathconf for systems that lack them, providing a minimal posix interface - heavily augmented getconf, now supports all POSIX.1-200x, POSIX.2-200x, Solaris 7, AIX 4.2 getconf variables 5/29 ---- builtins/setattr.def - make `readonly', `export', and `declare' print `invisible' variables as just a command and variable name, without a value, when listing all variables (as POSIX.2-200x d6 requires) 5/30 ---- configure.in - upgraded to autoconf-2.50 on main development machine, so require autoconf-2.50 in preparation for using some if its new features - call AC_C_PROTOTYPES - remove call to AC_EXEEXT, which now does the wrong thing - changed AC_INIT to new flavor - added call to AC_CONFIG_SRCDIR - AC_CONFIG_HEADER -> AC_CONFIG_HEADERS - AC_RETSIGTYPE -> AC_TYPE_SIGNAL configure.in, aclocal.m4, config.h.in - removed call to BASH_LARGE_FILE_SUPPORT, use AC_SYS_LARGEFILE standard support, with new macros _FILE_OFFSET_BITS and _LARGE_FILES - removed definition of BASH_LARGE_FILE_SUPPORT doc/bashref.texi - document new `--enable-largefile' configure option lib/readline/readline.c - change rl_set_prompt to call rl_expand_prompt unconditionally, so local_prompt and local_prompt_prefix get set correctly 6/6 --- lib/readline/complete.c - don't append `/' or ` ' to a match when completing a symlink that resolves to a directory, unless the match doesn't add anything to the word. This means that a tab will complete the word up to the full name, but not add anything, and a subsequent tab will add a slash. Change to append_to_match; callers changed hashlib.c - new function, hash_table_nentries (table), returns the number of items in TABLE hashlib.h - extern declaration for hash_table_nentries configure.in - configure without bash malloc on openbsd; they claim it needs eight-bit alignment (which the bash malloc provides, but...) 7/2 --- stringlib.c - only call RESIZE_MALLOCED_BUFFER from strsub() if the replacement string length is > 0, avoid possible hangs if replacement is null subst.c - don't include input.h; no longer needed configure.in - remove calls to AC_SYS_RESTARTABLE_SYSCALLS and BASH_SYS_RESTARTABLE_SYSCALLS; the results are no longer used config.h.in - remove define for HAVE_RESTARTABLE_SYSCALLS aclocal.m4 - removed definition of BASH_SYS_RESTARTABLE_SYSCALLS; no longer used execute_cmd.c - changed select command so `return' no longer terminates the select command, so it can be used to return from an enclosing function. This is as ksh (88 and 93) does it lib/readline/vi_mode.c - fix trivial typo in declaration of vi_motion; `t' appears twice; the second instance should be `T' 7/3 --- configure.in - don't add -static to LDFLAGS on Solaris 2.x. This means that the auxiliary programs will be built as dynamic executables, but that should do no harm 7/5 --- lib/glob/fnmatch.c - fix the code that processes **(pattern) to short-circuit if the pattern is ill-formed or lacks a trailing `)' -- this fixes the segfault on **(/*) Makefile.in, builtins/Makefile.in - split CCFLAGS into CCFLAGS_FOR_BUILD and CFLAGS, to aid in cross-compilation - build programs that use $(CC_FOR_BUILD) using $(CCFLAGS_FOR_BUILD) configure.in, config.h.in - check for getaddrinfo(3), define HAVE_GETADDRINFO if found lib/sh/netopen.c - implemented a version of _netopen (_netopen6) that uses getaddrinfo(3) if available, use if HAVE_GETADDRINFO is defined. old _netopen is _netopen4; _netopen now calls either _netopen6 or _netopen4 as appropriate 7/9 --- builtins/exit.def - don't source ~/.bash_logout if subshell_environment is non-zero execute_command.c - in execute_until_or_while, handle the case where `breaking' is set in the loop test (e.g., by the job control code when a job is stopped with SIGTSTP), but the return value from the test is something that would cause the loop to break. Need to decrement `breaking' in this case 7/10 ---- execute_cmd.c - in execute_in_subshell, make sure a command of type cm_subshell inherits its `enclosing' command's CMD_IGNORE_RETURN flag variables.c - in maybe_make_export_env, don't allow restricted shells to put exported functions in the export environment 7/11 ---- lib/glob/strmatch.h - renamed old fnmatch.h - changed guard #ifdef to _STRMATCH_H - include system if HAVE_LIBC_FNM_EXTMATCH is defined lib/glob/strmatch.c - renamed old fnmatch.c - include "strmatch.h" - if HAVE_LIBC_FNM_EXTMATCH is defined, define a dummy version of strmatch() that just calls fnmatch(3) lib/glob/glob.c - include "strmatch.h" - fnmatch -> strmatch Makefile.in, lib/glob/Makefile.in - fnmatch -> strmatch {bashhist,execute_cmd,pathexp,pcomplete,shell,stringlib,subst,test}.c, pathexp.h,builtins/help.def - include - fnmatch -> strmatch execute_cmd.c - broke the code that parses the interpreter name from a #! line out from execute_shell_script to a new function, getinterp() - call getinterp from execute_shell_script - use return value from getinterp in error message about bad #! interpreter in shell_execve 7/12 ---- lib/readline/isearch.c - the last isearch string is now remembered in a new static variable, last_isearch_string - if ^R^R is typed, readline now searches for the remembered isearch string, if one exists 7/24 ---- pcomplete.h - extern declaration for completions_to_stringlist() 7/25 ---- builtins/complete.def - make compgen handle -o default option - make compgen return success only if sl->list_len is non-zero, indicating that there are items on the list 7/31 ---- execute_cmd.c - in execute_connection, force stdin to /dev/null for asynchronous commands if job control is not active, not just if the shell is running a shell script (since you can run `set -m' in a script) lib/readline/rltty.c - make sure _rl_tty_restore_signals resets `tty_sigs_disabled' on successful restoration of the terminal modes - make sure _rl_tty_disable_signals turns off IXON so that ^S and ^Q can be read by rl_quoted_insert 8/1 --- aclocal.m4 - new check for FNM_EXTMATCH being defined in , as Ullrich Drepper intends to do for new versions of GNU libc config.h.in - new definition for HAVE_LIBC_FNM_EXTMATCH configure.in - check for fnmatch, but don't define anything in config.h - call BASH_FUNC_FNMATCH_EXTMATCH to check for FNM_EXTMATCH 8/2 --- alias.h - remove bogus extern declaration for xmalloc() - include "stdc.h" - add prototype declarations for all extern function declarations xmalloc.c,lib/readline/xmalloc.c - fix xmalloc to return a PTR_T - fix xrealloc to return a PTR_T and take a PTR_T as first argument include/ansi_stdlib.h - extern declarations for malloc and realloc have them return PTR_T xmalloc.h - new file, with extern declarations for functions in xmalloc.c general.h - removed extern declarations for functions in xmalloc.c - include xmalloc.h Makefile.in,builtins/Makefile.in - update dependencies to include xmalloc.h parse.y,{alias,array,bashline,bracecomp,execute_cmd,findcmd,flags,general, hashcmd,locale,mailcheck,make_cmd,pathexp,pcomplete,print_cmd,stringlib, subst,unwind_prot,variables}.c builtins/{common,evalfile}.c builtins/{cd,command,enable,exec,printf,read,set}.def lib/sh/{makepath,netopen,pathphys,setlinebuf,shquote,snprintf,stringlist, strtrans,tmpfile}.c lib/readline/{util,terminal,shell,readline,macro,kill,isearch,input, histfile,histexpand,display,complete,bind}.c - make sure all calls to xmalloc are cast to the right return value siglist.c - include xmalloc.h parse.y,{alias,bashline,bracecomp,expr,make_cmd,nojobs,print_cmd,subst}.c builtins/{fc,printf,read}.def lib/sh/snprintf.c, lib/tilde/tilde.c lib/readline/{bind,display,histexpand,isearch,macro,util,vi_mode}.c - make sure all calls to xrealloc are cast to the right return value lib/sh/{netopen,setlinebuf,shquote,snprintf}.c, lib/tilde/tilde.c - include xmalloc.h, remove extern declaration of xmalloc lib/readline/xmalloc.h - xmalloc and xrealloc should return PTR_T lib/readline/rldefs.h - don't include an extern declaration for xmalloc 8/7 --- support/shobj-conf - fixed up commented-out stanzas for HP's unbundled C compiler on HP/UX support/bashbug.sh - force the subject to be changed from the default lib/readline/doc/{rluser.texinfo,readline.3}, doc/bash.1 - document that transpose-words swaps the last two words on the line if point is at the end of the line 8/9 --- stringlib.c - fix possible infinite recursion problem with null pattern in strsub() hashlib.c - new function copy_hash_table to copy a hash table using a caller- supplied function to copy item data (defaults to savestring()) hashlib.h - new extern declaration for copy_hash_table builtins/declare.def - changes so that declare [-a] var=value assigns `value' to element 0 of array variable `var' like ksh93 - change so that declare [-a] var[N]=value assigns `value' to element N of array variable `var' like ksh93 8/13 ---- arrayfunc.c - new file, for miscellaneous array functions arrayfunc.h - new file, extern declarations for functions in arrayfunc.c variables.c - move convert_var_to_array, bind_array_variable, assign_array_from_string, assign_array_var_from_word_list, assign_array_var_from_string, quote_array_assignment_chars, skipsubscript, unbind_array_element, print_array_assignment to arrayfunc.c shell.h - include arrayfunc.h after variables.h variables.h - remove above extern function declarations moved to arrayfunc.h - add extern declaration for var_lookup Makefile.in - add arrayfunc.c, arrayfunc.h in appropriate places - add arrayfunc.h to dependencies subst.c - move valid_array_reference, array_expand_index, array_variable_part, array_value_internal, array_value (now global), get_array_value, do_array_element_assignment to arrayfunc.c subst.h - extern declarations for functions above moved to arrayfunc.h arrayfunc.h - extern declarations for above functions from subst.c subst.[ch] - string_list_dollar_star and string_list_dollar_at are now global functions - quote_escapes is now a global function subst.c - maybe_expand_string -> expand_string_if_necessary - expand_string_to_string -> expand_string_to_string_internal - new functions: expand_string_to_string and expand_string_unsplit_to_string, which call expand_string_to_string_internal with expand_string and expand_string_unsplit as the FUNC arguments, respectively arrayfunc.c - change array_expand_index to call expand_string_to_string instead of maybe_expand_string 8/14 ---- shell.c - in execute_env_file, call expand_string_unsplit_to_string mailcheck.c - in check_mail, call expand_string_to_string variables.c - in assign_in_env, call expand_string_unsplit_to_string arrayfunc.c - new function, array_variable_name, splits an array reference into a name (which is returned as a new string) and subscript - change array_variable_part to just call array_variable_name and look up the string returned with find_variable - new function, find_or_make_array_variable (name, flags) which will look up an array variable and convert a string variable to an array if necessary. The FLAGS argument, if non-zero, says to check the readonly and noassign attributes and fail if either is set builtins/read.def - make `read -a aname' honor any readonly status of `aname' - read -a now calls find_or_make_array_variable with FLAGS value 1 arrayfunc.[ch], subst.c, builtins/{declare,read}.def - do_array_element_assignment -> assign_array_element 8/20 ---- parse.y - changed `for' command grammar to allow missing word list after `IN' token, like latest POSIX drafts require lib/sh/tmpfile.c - in sh_mktmpname(), check for filenum == 0 and init to non-zero number in this case. it can happen on arithmetic overflow support/mkversion.sh - added `[0-9].[0-9][0-9][a-z]' as an acceptable value for a distribution to allow for intermediate versions, like 2.05a support/config.guess - removed the addition of the output of `/usr/bin/objformat' when creating the canonical name on FreeBSD machines, so the canonical name is once again `freebsd4.2' instead of `freebsdelf4.2' 8/22 ---- lib/readline/{rlstdc,history,keymaps,readline,rldefs,rlprivate,rlshell, rltypedefs,xmalloc}.h lib/readline/{bind,compat,complete,display,funmap,histexpand,histsearch, input,isearch,kill,nls,parens,readline,rltty,search,shell,signals,vi_mode - changed __P to PARAMS lib/tilde/tilde.[ch] - changed __P to PARAMS {Makefile,configure}.in - changed the version number to 2.05a - changed the release status to `alpha1' 8/23 ---- support/shobj-conf - support for building shared libraries on Darwin/MacOS X siglist.h - extern declaration for strsignal() to compensate for lack of a definition in some system include files jobs.c - remove casts from strsignal() calls [bash-2.05a-alpha1 frozen] 8/27 ---- [bash-2.05a-alpha1 released] 8/27 ---- execute_cmd.c - fix eval_arith_for_expr to handle the case where the expanded word list is NULL, returning 0 in this case print_cmd.c - in print_function_def, make sure that func_redirects is assigned a value before being used 8/28 ---- alias.c - include for definition of isalpha() bashhist.h - add prototypes for extern function declarations flags.c - include bashhist.h for extern function declarations mksyntax.c - include if HAVE_UNISTD_H is defined in config.h parse.y - include test.h for extern function declarations externs.h - change extern declaration for setlinebuf to sh_setlinebuf stringlib.c - include for extern function declarations variables.h - add function prototypes for all of the sv_* functions builtins/common.h - add extern declarations for set_shellopts() and parse_shellopts() from builtins/set.def variables.c - include "hashcmd.h" for extern declaration for flush_hashed_filenames - include "pathexp.h" for extern declaration for setup_glob_ignore lib/malloc/malloc.c - cast to `long' instead of `int' in memalign for 64-bit machines {pcomplete,trap}.c - changed printf escape sequences used to print pointers to %p lib/readline/undo.c - include "xmalloc.h" for extern function declaration input.h - add function prototypes to extern declarations for getc_with_restart and ungetc_with_restart variables.[ch] - changed type of `function' member of `struct name_and_function' to `sv_func_t', which is defined and prototyped in variables.h - map_over now takes an `sh_var_map_func_t *' shell.h - start of a set of function pointer typedefs like those in lib/readline/rltypedefs.h hashlib.[ch] - second paramter to flush_hash_table is now an `sh_free_func_t *' trap.c - parameter to reset_or_restore_signal_handlers is now an `sh_resetsig_func_t *' pcomplete.h, pcomplib.c - function pointer argument to print_all_compspecs is now an `sh_csprint_func_t *' - function pointer `list_getter' element of an `ITEMLIST' is now prototyped with __P((...)) instead of using `Function *' jobs.[ch] - `j_cleanup' member of a JOB is now an `sh_vptrfunc_t *' alias.c - map_over_aliases now takes an `sh_alias_map_func_t *' - free_alias_data now takes a `PTR_T' pathexp.c - function pointer argument to ignore_globbed_names is now an `sh_ignore_func_t *' bashline.c - function pointer argument to _ignore_completion_names is now an `sh_ignore_func_t *' pathexp.h,{bashhist,bashline.c - `item_func' member of a `struct ignorevar' is now an `sh_iv_item_func_t *' builtins/evalfile.c - `errfunc' is now an `sh_vmsg_func_t *' jobs.c - map_over_job now takes an `sh_job_map_func_t *' as its first argument array.[ch] - function pointer argument to array_walk is now an `sh_ae_map_func_t *' general.c - tilde_expansion_preexpansion_hook has type `tilde_hook_func_t *', and so the assignment in tilde_initialize doesn't need a cast list.c - map_over_words now takes an `sh_icpfunc_t *' as its second argument input.h - the `getter' and `ungetter' function pointer members of a BASH_INPUT are now of types `sh_cget_func_t *' and `sh_cunget_func_t *' respectively - init_yy_io now takes an `sh_cget_func_t *' as its first argument and an `sh_cunget_func_t *' as its second parse.y - init_yy_io now takes an `sh_cget_func_t *' as its first argument and an `sh_cunget_func_t *' as its second - initialize_bash_input casts bash_input.getter and bash_input.ungetter appropriately builtins/mkbuiltins.c - make the extern function definitions written to builtext.h have prototypes with __P((...)) - include "stdc.h" - change Function to mk_handler_func_t - fixed comment_handler to take the right number of args - prototyped all the handler functions with __P((...)) builtins.h - the `function' member of a struct builtin is now of type `sh_builtin_func_t *' builtins/common.[ch] - last_shell_builtin, this_shell_builtin are now of type `sh_builtin_func_t *' - find_shell_builtin, builtin_address, find_special_builtin now return `sh_builtin_func_t *' builtins/exit.def, {execute_cmd,jobs,nojobs,variables}.c, parse.y - changed all declarations of last_shell_builtin and this_shell_builtin execute_cmd.c - execute_builtin, execute_builtin_or_function, execute_subshell_builtin_or_function now take an `sh_builtin_func_t *' instead of a `Function *' for argument - changed appropriate variables from `Function *' to `sh_builtin_func_t *' builtins/{bind,builtin,enable,read,setattr}.def - replaced uses of `Function *' in variable declarations with appropriate types (sh_builtin_func_t * or rl_command_func_t *) builtins/set.def - set_func and get_func members of binary_o_options are now of types `setopt_set_func_t *' and `setopt_get_func_t *', which are prototyped builtins/shopt.def - set_func member of shopt_vars is now of type `shopt_set_func_t *' bashline.c - enable_hostname_completion now returns `int' (the old value of perform_hostname_completion) [The only use of Function and VFunction now is for unwind-protects] 9/4 --- lib/sh/getcwd.c - use const define from config.h rather than `CONST' - use PTR_T define from xmalloc.h rather than `PTR' - include xmalloc.h for PTR_T - remove PATH_MAX define, rely on value from maxpath.h {general,mailcheck}.c, lib/sh/{pathcanon,pathphys}.c - don't include maxpath.h directly; it's already included by shell.h lib/sh/mailstat.c - new `mailstat()' implementation, to stat a mailbox file for mail checking. handles maildir-style mail directories with one file per message and creates a dummy stat struct from them lib/sh/Makefile.in - add mailstat.c and mailstat.o in the appropriate places lib/malloc/malloc.c - augmented implementation with wrapper functions that pass in file and line number information from cpp. currently unused, but a placeholder for future debugging and use tracking lib/malloc/shmalloc.h - new file, extern declarations for allocation wrapper functions for use by the shell (and others, I guess) xmalloc.[ch] - wrapper functions for xmalloc, xfree, xrealloc (sh_ prefixed) that pass cpp line number information through to the malloc functions, if USING_BASH_MALLOC is defined 9/5 --- lib/malloc/gmalloc.c - removed; no longer part of distribution lib/malloc/Makefile.in - removed references to gmalloc.[co] configure.in, doc/bashref.texi - removed references to `--with-glibc-malloc' configure option {configure,Makefile}.in - changed the way bash malloc is configured into the Makefile, making it more like how readline is configured. If the bash malloc is not configured in, nothing in lib/malloc will be built 9/6 --- lib/malloc/imalloc.h - new file, some internal malloc definitions lib/malloc/mstats.h - new file, definitions for malloc statistics structs and functions lib/malloc/trace.c - new file, malloc tracing functions (currently just print messages to stderr), code is #ifdef MALLOC_TRACE lib/malloc/stats.c - new file, moved malloc stats code from malloc.c to here lib/malloc/malloc.c - moved some definitions to imalloc.h - moved stats code to stats.c - malloc tracing calls added to internal_{malloc,realloc,free}, all #ifdef MALLOC_TRACE lib/malloc/Makefile.in, Makefile.in - added {imalloc,mstats}.h, {trace,stats}.c parse.y - changed decode_prompt_string to save and restore $? (last_command_exit_value) around calls to expand_prompt_string(), so command substitutions in PS1, etc. don't change $? {array,subst}.c - a couple more arrayind_t fixes from Paul Eggert configure.in - remove redundant check for wait3(2) redir.h - fixed a typo (stdin_redirs -> stdin_redirects) 9/10 ---- execute_cmd.c - remove check for \n and \r from WHITESPACE macro, since those chars are not whitespace as returned by the whitespace(c) macro - getinterp now takes a `char *' as first arg, not unsigned char * - execute_shell_script now takes a `char *' as first arg, not unsigned char * - fix typo in forward declaration for `initialize_subshell' general.[ch] - check_binary_file now takes a (char *) argument, not unsigned char * - pass unsigned char to isspace and isprint because of ISO C fuckup - bash_tilde_expand now takes a `const char *' as its argument builtins/evalfile.c, shell.c - buffer passed to check_binary_file is char, not unsigned char parse.y - fix extern declaration for yyerror() - yyerror now takes a `const char *' as first arg {error,jobs}.c - fixes to printf-style functions to handle pids wider than an int lib/readline/{isearch,vi_mode}.c - fix call to rl_message in rl_display_search (remove extra arg) variables.c - fix missing argument to builtin_error in make_local_variable builtins/getopts.def - since getopts takes no options, change while loop calling internal_getopts to a simple `if' check builtins/printf.def - since printf takes no options, change while loop calling internal_getopts to a simple `if' check lib/readline/bind.c - remove _SET_BELL macro, expand code inline lib/readline/input.c - change _rl_input_available to use either select or FIONREAD, but not both lib/readline/readline.c - fix rl_digit_loop to remove unreachable code at end of loop {bashhist,bashline,expr,jobs,redir,shell}.c, builtins/fc.def, lib/sh/snprintf.c - bracket unused functions with #ifdef INCLUDE_UNUSED/#endif - remove some unused variables execute_cmd.c - remove #ifdef'd code that allowed `return' to terminate a select statement expr.c - remove some extraneous tests from strlong() array.h - arrayind_t is now a long, since shell arithmetic is performed as longs - remove second declaration of new_array_element builtins/printf.def - in mklong, xrealloc cannot return NULL, so don't check for it - remove some #if 0 code - fix core dump triggered by a format specification with more than one `*' - remove `foundmod', since its value mirrors `modchar != 0' - include "common.h" for builtin_{error,usage} declarations Makefile.in,builtins/Makefile.in - updated some dependencies due to new include files pcomplete.c - include "execute_cmd.h" for declaration of execute_shell_function arrayfunc.c - include for printf - include "builtins/common.h" for builtin_error declaration builtins/evalstring.c - include "../trap.h" for run_trap_cleanup declaration builtins/help.def - include "common.h" instead of locally declaring builtin_error and builtin_usage error.h - add extern declaration for itrace() - add prototype to extern declaration of get_name_for_error - file_error now takes a `const char *' as first argument externs.h - added prototype for sh_setlinebuf declaration, bracketed with NEED_SH_SETLINEBUF_DECL so we don't need stdio.h everywhere - add extern declaration for parse.y:return_EOF() shell.c - add NEED_SH_SETLINEBUF_DECL before including shell.h lib/readline/callback.c - include or "ansi_stdlib.h" for abort declaration quit.h - remove declaration of throw_to_top_level subst.c - remove unused extern declaration for getopts_reset lib/sh/netopen.c - include for legal_number, etc. - add prototype for inet_aton extern declaration lib/sh/clock.c - include for __P declaration - add extern declaration for get_clk_tck support/mkversion.sh - changed so that extern function declarations for functions in version.c (moved from externs.h) are in the generated version.h shell.h - include version.h version.c - various `char *' version variables are now `const char *' general.h - add prototype for same_file, bracketed with _POSIXSTAT_H #ifdef, since that's what include/posixstat.h defines builtins/common.[ch] - _evalfile, maybe_execute_file, source_file, and fc_execute_file now take a `const char *' as their first argument eval.c - removed extern declaration of yyparse; it's in externs.h parse.y - added prototypes to static forward function declarations - changed local `all_digits' variable in read_token_word () to all_digit_token to avoid clash with all_digits() function in general.c {bashhist,copy_cmd,make_cmd,hashlib,mailcheck}.c - added prototypes for static function declarations shell.h - add extern declarations for interactive, interactive_shell, changed c files with extern declarations pcomplete.c - changed it_init_aliases to avoid shadowing global variable `aliases' bashline.c,pathexp.c,general.h - sh_ignore_func_t is now a pointer to a function taking a `const char *'; users changed configure.in - test for config.h.in - add #undef HAVE_STRINGS_H bashansi.h - change like recommended in autoconf manual 9/11 ---- [a date which will live in infamy. prayers for the victims.] execute_cmd.c - don't use an absolute index into abuf in mkfmt, use sizeof(abuf) to compute last index builtins/common.c - fix read_octal to do a better job of detecting overflow while iterating through the string builtins/umask.def - change octal-print mode to print 4 digits, like other shells - cast umask to unsigned long to avoid problems on systems where it's wider than an int (POSIX doesn't guarantee that mode_t is no wider than an int, but real-world systems use int) builtins/printf.def - mklong can never return NULL (it uses xrealloc), so the mainline doesn't need to check for NULL returns - new function, getldouble (long double *), to get long doubles - mklong now takes a `char *' as its second argument, the modifier(s) to use - changed use of `modchar' to handle more than a single modifier character - changed to handle `long double' and `L' formats better, rather than discarding long double information - since printf now follows the POSIX.2 rules for conversion errors, we can dispense with the status returns from the get* functions - make the get* functions as similar in structure as possible, removing type casts, etc. lib/sh/timeval.c,execute_cmd.c - change some instances of `long' to `time_t', for systems where a time_t is bigger than a long jobs.c - include "posixtime.h" instead of config.h.in - add defines for HAVE_DECL_CONFSTR, HAVE_DECL_STRTOLD, HAVE_DECL_SBRK, HAVE_DECL_PRINTF - remove defines for SBRK_DECLARED and PRINTF_DECLARED - add _GNU_SOURCE define configure.in - add AC_CHECK_DECLS for strtold, confstr, sbrk, printf - remove call to BASH_FUNC_SBRK_DECLARED - remove call to BASH_FUNC_PRINTF xmalloc.c, lib/malloc/malloc.c - change check of SBRK_DECLARED to HAVE_SBRK_DECL print_cmd.c - change PRINTF_DECLARED to HAVE_DECL_PRINTF builtins/evalstring.c, builtins/common.h - parse_and_execute now takes a `const char *' as its second argument input.h,parse.y - with_input_from_* functions now take a `const char *' as their second argument - init_yy_io now takes a `const char *' as its fourth argument parse.y,externs.h - parse_string_to_word_list now takes a `const char *' as its second argument tests/builtins.right - change output to account for extra digit in umask output pcomplib.c - free_progcomp now takes a PTR_T argument builtins/bashgetopt.h - include - add prototypes to extern declarations builtins/shopt.def - add prototypes to static function declarations builtins/{fc,umask,wait}.def, builtins/{bashgetopt,common}.c - include for isdigit macro (referenced by `digit(x)') lib/readline/complete.c - added more static function declarations with prototypes 9/12 ---- lib/sh/tmpfile.c - use `^' instead of `*' in sh_mktmpname to make filenames a bit more random include/stdc.h,lib/readline/rldstdc.h - add __attribute__ definition builtins/common.h - add printf __attribute__ to declaration of builtin_error error.h - add printf __attribute__ to declaration of programming_error, report_error, parser_error, fatal_error, sys_error, internal_error, internal_warning lib/readline/readline.h - add printf __attribute__ to declaration of rl_message pcomplete.c - add printf __attribute__ to declaration of debug_printf print_cmd.c - add printf __attribute__ to declarations of cprintf, xprintf include/chartypes.h - new file, includes and defines macros that check for safe (ascii) arguments before calling the regular ctype macros {alias,bashline,execute_cmd,expr,findcmd,general,locale,mksyntax,stringlib,subst,variables}.c parse.y builtins/{bashgetopt,common}.c builtins/{fc,printf,umask,wait}.def lib/glob/strmatch.c lib/sh/{oslib,pathcanon,pathphys,snprintf,strcasecmp,strindex,stringvec,strtod,strtol,strtrans}.c examples/loadables/{head,sleep}.c - include "chartypes.h" or instead of Makefile.in,{builtins,lib/{glob,sh}}/Makefile.in - update dependencies to include chartypes.h lib/sh/inet_aton.c - use `unsigned char' instead of `char' to pass to ctype.h functions lib/sh/netopen.c - check for '0' <= host[0] <= '9' in _getaddr instead of using isdigit subst.c,lib/sh/shquote.c - change array subscripts into sh_syntaxtab from `char' to `unsigned char' {alias,bashline,execute_cmd,expr,general,subst}.c, parse.y builtins/{fc,printf,umask,wait}.def builtins/{bashgetopt,common}.c lib/sh/{pathcanon,pathphys,snprintf,strcasecmp,strindex,strtod,strtol,strtrans}.c examples/loadables/{head,sleep}.c - change to use some of the new macros in chartypes.h - remove old local macro definitions now provided by chartypes.h general.h - remove definition of isletter, ISOCTAL, digit, digit_value - change legal_variable_starter and legal_variable_char to use chartypes.h macros - change ABSPATH to use chartypes.h macros lib/readline/util.c - change to use Paul Eggert's FUNCTION_FOR_MACRO define to define function replacements for macros in chardefs.h lib/readline/chardefs.h - added some of the same macros as in chartypes.h - change _rl_lowercase_p, _rl_uppercase_p, _rl_digit_p, _rl_to_upper, _rl_to_lower to use new IS* macros - added _rl_isident macro from vi_mode.c:isident lib/readline/{bind,complete,nls}.c - change to use some of the new macros from chardefs.h lib/readline/vi_mode.c - isident -> _rl_isident - remove local defines of macros in chardefs.h lib/sh/strtol.c - updated to new version, modified from glibc 2.2.4 and sh-utils-2.0. This one can do strtoll and strtoull, if necessary 9/13 ---- builtins/ulimit.def - changed get_limit so it retrieves both hard and soft limits instead of one or the other - changed callers of get_limit - changed getmaxvm to take soft limit, hard limit as arguments - changed getmaxuprc to just take a single argument, the value - changed calls to printone() to pass soft limit or hard limit depending on `mode' instead of using old current_limit variable - moved check for out-of-range limits in ulimit_internal into the block that converts a string argument to a value of type rlim_t - changed RESOURCE_LIMITS struct to break the description into a description string and separate scale factor string - changed print_all_limits to print a single error message if get_limit fails, including limits[i].description now that the scale factor has been removed from the description string - removed DESCFMT define, since it's now used only in printone() - changed printone to print the option character associated with a particular limit if we're printing multiple limits - changed calls to builtin_error to print the description associated with a limit if setting or getting the limit fails - added support for new POSIX 1003.1-200x rlim_t values: RLIM_SAVED_CUR and RLIM_SAVED_MAX, which expand to the current soft and hard limits, whatever they are - changed printone to print `hard' or `soft' if the current limit is RLIM_SAVED_MAX or RLIM_SAVED_CUR, respectively - changed ulimit_internal to handle new `hard' and `soft' arguments - changed help text do describe the special limit arguments `hard', `soft', and `unlimited' doc/{bash.1,bashref.texi} - documented new `hard' and `soft' limit arguments to `ulimit' hashlib.[ch] - find_hash_item now takes a `const char *' is its first argument - hash_string now takes a `const char *' is its first argument - remove_hash_item now takes a `const char *' as its first argument pcomplib.c - removed cast from first argument to find_hash_item in find_compspec general.[ch] - absolute_program now takes a `const char *' as its argument - absolute_pathname now takes a `const char *' as its argument lib/glob/glob.[ch] - glob_pattern_p now takes a `const char *' as its argument bashline.c - removed cast from first argument to absolute_program in command_word_completion_function - removed cast from first argument to glob_pattern_p in attempt_shell_completion findcmd.[ch] - find_absolute_program, find_user_command, find_path_file, search_for_command, user_command_matches now take a `const char *' as their first argument - file_status, executable_file, is_directory, executable_or_directory now take a `const char *' as their argument - _find_user_command_internal, find_user_command_internal, find_user_command_in_path lib/sh/makepath.c, externs.h - changed sh_makepath so it takes `const char *' for its first two arguments hashcmd.[ch] - find_hashed_filename now takes a `const char *' as its first arg - remove_hashed_filename now takes a `const char *' as its first arg variables.[ch] - new_shell_variable, var_lookup, shell_var_from_env_string, find_name_in_env_array, bind_function, makunbound, bind_name_in_env_array, bind_tempenv_variable, bind_variable now take a `const char *' as their first arg - find_function, make_new_variable, find_tempenv_variable, find_variable_internal, find_variable, set_func_read_only, set_func_auto_export, all_variables_matching_prefix, assign_in_env, assignment, kill_local_variable, make_local_variable, unbind_variable now take a `const char *' as their arg - mk_env_string now takes `const char *' arguments arrayfunc.[ch] - skipsubscript now takes a `const char *' as its argument 9/17 ---- lib/readline/complete.c - attempt to preserve case of what the user typed in compute_lcd_of_matches if we're ignoring case in completion builtins/{let,pushd}.def,{execute_cmd,expr}.c - change some 0L constants to 0 and let the compiler sort it out 9/18 ---- lib/malloc/alloca.c - alloca now takes a `size_t' argument include/memalloc.h - if we're providing an extern function declaration for alloca, use `void *' and prototype if __STDC__ is defined - if HAVE_ALLOCA_H is defined, but C_ALLOCA is defined, don't define HAVE_ALLOCA 9/19 ---- subst.c - do_assignment_internal, do_assignment, and do_assignment_no_expand now take a `const char *' as their first argument general.h - a `sh_assign_func_t' is now a function taking a `const char *' and returning int hashcmd.c - free_filename_data now takes a `PTR_T' argument to agree with the typedef for `sh_free_func_t' lib/sh/snprintf.c - use TYPE_MAXIMUM define like strtol.c instead of huge constants 9/20 ---- lib/sh/snprintf.c - don't bother to compile the bulk of the body unless HAVE_SNPRINTF or HAVE_ASPRINTF is not defined 9/24 ---- flags.c - ignore `set -n' if the shell was started interactively lib/readline/readline.c - initialize readline_echoing_p to 0; let the terminal-specific code in rltty.c set it appropriately lib/malloc/malloc.c - changed internal_memalign() slightly to avoid compiler warnings about negating an unsigned variable (-alignment -> (~alignment + 1)) 9/27 ---- lib/readline/readline.c - changed rl_newline to set _rl_history_saved_point appropriately for the {previous,next}_history code lib/readline/rlprivate.h - extern declaration for _rl_history_preserve_point lib/readline/bind.c - new bindable variable, `history-preserve-point', sets value of _rl_history_preserve_point 10/1 ---- lib/malloc/table.c - new file, with a map of allocated (and freed) memory for debugging multiple frees, etc. Indexed by hash on values returned by malloc(); holds size, file and line number info for last alloc or free and a couple of statistics pointers lib/malloc/malloc.c - a few cleanups; added calls for registering allocations and frees if MALLOC_REGISTER is defined - replaced MALLOC_RETURN with explicit MALLOC_NOTRACE define - reordered fields in `struct...minfo' in `union mhead' to restore eight-byte alignment - added explicit checks for underflow in free and realloc since checking mh_magic2 is not sufficient to detect everything (it's no longer the last field in the struct, and thus not the bytes immediately preceding what's returned to the user) - new function, xbotch, for printing file and line number info for the failed assertion before calling botch() (programming_error()) configure.in - replaced call to BASH_C_LONG_LONG with call to AC_CHECK_TYPES([long long]) - moved the C compiler tests before the tests for various system types, so we can know whether we have `long long' before testing for 64-bit types - if we have `long long', check for sizeof(long long) and save value aclocal.m4 - changed BASH_TYPE_BITS64_T to check `long long' before `long', but after `double' 10/2 ---- lib/malloc/malloc.c - made malloc and realloc both agree on the rounding for a request of size N (round up to nearest multiple of 8 after adjusting for malloc overhead); uses new ALLOCATED_BYTES macro - realloc and free now use new IN_BUCKET macro for underflow checks execute_cmd.c - fixed time_command() to use `time_t' instead of `long' to hold time stamps lib/sh/clock.c - clock_t_to_secs now takes a `time_t *' second argument - fixed print_clock_t to call clock_t_to_secs with right arguments lib/sh/timeval.c - fixed print_timeval to make `minutes' a `long' and make its structure identical to print_clock_t redir.c - changed redirection_error to check for EBADF and use the file descriptor being redirected from in the error message if it is >= 0 Makefile.in - changed release status to `beta1' lib/glob/collsyms.h - added a few ASCII symbols to the posix_collsyms array 10/3 ---- aclocal.m4 - fixed typo in BASH_TYPE_BITS64_T configure.in - added check for unsigned chars with AC_C_CHAR_UNSIGNED config.h.in - added PROTOTYPES and __CHAR_UNSIGNED__ #defines general.h - if CHAR_MAX is not define by , provide a definition builtins/printf.def - change tescape() to mask \0 and \x escape sequences with 0xFF - change tescape() to process at most two hex digits after a `\x' lib/sh/strtrans.c - change strtrans() to mask \0 and \x escape sequences with 0xFF - change strtrans() to process at most two hex digits after a `\x'. This affects `echo -e' and $'...' processing lib/readline/bind.c - changed rl_translate_keyseq() to process at most two hex digits after a `\x' lib/readline/doc/{rluser.texinfo,readline.3}, doc/bash.1 - changed documentation for key binding escape sequences to specify that at most two hex digits after \x are translated - changed documentation for key binding to specify that the result of \nnn or \xhh escapes is an eight-bit value, not just ASCII doc/{bash.1,bashref.texi} - changed documentation of $'...' to specify that at most two hex digits after \x are translated - changed `echo' documentation to specify that at most two hex digits after \x are translated - changed documentation for `echo' and $'...' to specify that the result of \nnn or \xhh escapes is an eight-bit value, not just ASCII 10/4 ---- lib/malloc/malloc.c - changed interface for xbotch to pass memory address and error code as two additional arguments - call mregister_describe_mem from xbotch to get the last allocation or free before the botch configure.in - call AC_CHECK_DECLS([strsignal]) config.h.in - add HAVE_DECL_STRSIGNAL siglist.h - make declaration of strsignal() dependent on !HAVE_DECL_STRSIGNAL 10/5 ---- support/texi2html - upgraded to version 1.64 10/9 ---- aclocal.m4 - added check for `long long' to BASH_TYPE_PTRDIFF_T configure.in - replaced call to BASH_HAVE_TIOCGWINSZ with AC_HEADER_TIOCGWINSZ aclocal.m4 - replaced body of BASH_STRUCT_TERMIOS_LDISC with call to AC_CHECK_MEMBER(struct termios.c_line, ...) - replaced body of BASH_STRUCT_TERMIO_LDISC with call to AC_CHECK_MEMBER(struct termios.c_line, ...) [bash-2.05a-beta1 frozen] 10/10 ----- lib/sh/snprintf.c - fixed exponent() to not smash the trailing zeros in the fraction when using %g or %G with an `alternate form' - fixed exponent() to handle the optional precision with %g and %G correctly (number of significant digits before the exponent) 10/11 ----- expr.c - fixed strlong() to correct the values of `@' and `_' when translating base-64 constants (64#@ == 62 and 64#_ == 64), for compatibility with ksh lib/sh/itos.c - added a slightly more flexible fmtlong() function that takes a base argument and flags (for future use) - rewrote itos and inttostr in terms of fmtlong lib/sh/fmtulong.c - new file, converts unsigned long to string. hooks for `unsigned long long' in the future. unused as yet 10/15 ----- lib/readline/rltty.c - change the SET_SPECIAL macro to avoid possible (but highly unlikely) negative array subscripts error.h - add __attribute__ to extern declaration of itrace (even though the function isn't defined in released versions of bash) bashansi.h - include if HAVE_STRINGS_H is defined, to get any extra function declarations provided therein copy_cmd.c - fix typo in forward declaration for copy_arith_for_command lib/malloc/stats.c - make the accumulators in _print_malloc_stats be `unsigned long' instead of `int' externs.h, sig.h - add `__noreturn__' gcc attribute to exit_shell and jump_to_top_level declarations lib/sh/mailstat.c, support/bashversion.c - include for some string function declarations lib/malloc/shmalloc.h - added extern declarations of functions that do malloc debugging lib/readline/{isearch,readline,vi_mode}.c - make sure we index into _rl_keymap with a non-negative index parse.y - make sure we index into sh_syntaxtab with a non-negative index lib/readline/vi_mode.c - bound the vi_mark_chars array with the number of characters between 'a' and 'z' rather than using a fixed amount - don't use _rl_lowercase_p when deciding whether the char read by rl_vi_set_mark is a valid mark; just use 'a' <= char <= 'z' lib/readline/chardefs.h - conditionally include memory.h and strings.h as in general.h - replace ISASCII with IN_CTYPE_DOMAIN like other GNU software - add defines for ISPRINT(c), ISLOWER(c) and ISUPPER(c) - fix defines for _rl_lowercase_p, _rl_uppercase_p, _rl_digit_p, _rl_pure_alphabetic, ALPHABETIC, _rl_to_upper, _rl_to_lower, and _rl_isident to work on systems with signed chars include/chartypes.h - replace ISASCII with IN_CTYPE_DOMAIN like other GNU software lib/sh/{strcasecmp,strtod,strtol}.c - don't pass possibly-negative characters to tolower() or toupper() lib/glob/strmatch.c - don't bother testing for isupper in FOLD; rely on TOLOWER macro from to do it - don't use local definitions of isblank, et al.; rely on macros from lib/readline/{display,readline}.c, mksyntax.c - use new ISPRINT macro instead of isprint() builtins/{kill.def,mkbuiltins.c},{error,execute_cmd,jobs,nojobs,subst}.c - don't assume that a pid_t fits into an int for printing and other uses variables.[ch] - the unused put_gnu_argv_flags_into_env now takes a `long' pid argument configure.in, config.h.in - call AC_STRUCT_ST_BLOCKS, define HAVE_STRUCT_STAT_ST_BLOCKS if found - check for strtoull(), define HAVE_STRTOULL if found - check for uintmax_t, define to `unsigned long' if not found lib/sh/mailstat.c - don't use st_blocks member of struct stat unless HAVE_STRUCT_STAT_ST_BLOCKS is defined; otherwise use the st_nlink field to return the total number of messages in a maildir-style mail directory general.h,{alias,expr,general,subst,variables}.c builtins/{printf,read}.def lib/readline/{bind,complete,nls}.c lib/sh/{pathcanon,pathphys,shquote,snprintf,strindex,strtod,strtol,strtrans}.c - cast args to ctype macros to unsigned char for systems with signed chars; other fixes for signed chars lib/sh/{fmtullong,strtoull.c} - new files, more support for `long long' Makefile.in, lib/sh/Makefile.in - make fmtullong.o and strtoull.o part of libsh lib/sh/itos.c - remove local copy of fmtlong; use fmtulong instead - new functions: uitos, uinttostr work on `unsigned long' lib/sh/snprintf.c - fixes to make `unsigned long long' work (%llu) - fixes to make unsigned formats not print the sign when given an unsigned long that is greater than LONG_MAX externs.h - extern declarations for fmtulong, fmtulloing, strtoull - extern declarations for uitos, uinttostr 10/16 ----- configure.in - move header checks before function checks - move c compiler tests before header checks - check for with BASH_HEADER_INTTYPES - change type checks for intmax_t, uintmax_t to not attempt to include - check for strtoimax, strtoumax, strtoll, strtol, strtoull, strtoul with BASH_CHECK_DECL (for declarations in header files) and AC_REPLACE_FUNCS (for availability and LIBOBJS substitution) - remove check for have_long_long around sizeof check for long long (since autoconf will give it a size of 0 if the type isn't found) config.h.in - add a define for HAVE_INTTYPES_H - add a define for HAVE_UNSIGNED_LONG_LONG - add defines for HAVE_STRTOIMAX, HAVE_STRTOUMAX, HAVE_STRTOLL aclocal.m4 - new func, BASH_HEADER_INTTYPES, which just calls AC_CHECK_HEADERS on ; separate so it can be AC_REQUIREd - AC_REQUIRE([BASH_HEADER_INTTYPES]) in BASH_CHECK_TYPE - include in BASH_CHECK_TYPE if HAVE_INTTYPES_H is defined - change AC_DEFINE to AC_DEFINE_UNQUOTED in BASH_CHECK_TYPE - new `long long' checking macros: BASH_TYPE_LONG_LONG and BASH_TYPE_UNSIGNED_LONG_LONG - new BASH_CHECK_DECL lib/sh/{strto[iu]max,strtoll}.c, lib/sh/Makefile.in, Makefile.in - new files externs.h - extern declarations for strtoll, strtoimax, strtoumax lib/malloc/alloca.c - include for size_t builtins/printf.def - new functions: getllong, getullong, getintmax, getuintmax; return long long, unsigned long long, intmax_t, uintmax_t respectively - builtin printf now handles `ll' and `j' length modifiers directly lib/sh/Makefile.in - use LIBOBJS to decide whether or not the strto* functions are needed 10/17 ----- configure.in - call AC_REPLACE_FUNCS(rename) - move getcwd, strpbrk, strcasecmp, strerror, strtod from AC_CHECK_FUNCS to AC_REPLACE_FUNCS - only call BASH_FUNC_GETCWD if $ac_func_getcwd == "yes" - call BASH_CHECK_SYS_SIGLIST - if we don't have vprintf but have _doprnt, call AC_LIBOBJ(vprint) lib/sh/Makefile.in - remove rename, getcwd, inet_aton, strpbrk, strcasecmp, strerror, strtod, vprint from OBJECTS; picked up from LIBOBJS aclocal.m4 - change BASH_FUNC_GETCWD to call AC_LIBOBJ(getcwd) if the libc getcwd(3) calls popen(3) - change BASH_FUNC_INET_ATON to call AC_LIBOBJ(inet_aton) if it's not found in libc or as a #define even with the special includes - BASH_KERNEL_RLIMIT_CHECK -> BASH_CHECK_KERNEL_RLIMIT - BASH_DEFAULT_MAILDIR -> BASH_SYS_DEFAULT_MAILDIR - BASH_JOB_CONTROL_MISSING -> BASH_SYS_JOB_CONTROL_MISSING - BASH_REINSTALL_SIGHANDLERS -> BASH_SYS_REINSTALL_SIGHANDLERS - BASH_SIGNAL_CHECK -> BASH_SYS_SIGNAL_VINTAGE - BASH_DUP2_CLOEXEC_CHECK -> BASH_FUNC_DUP2_CLOEXEC_CHECK - BASH_PGRP_SYNC -> BASH_SYS_PGRP_SYNC - BASH_RLIMIT_TYPE -> BASH_TYPE_RLIMIT - BASH_FUNC_PRINTF -> BASH_DECL_PRINTF - BASH_FUNC_SBRK_DECLARED -> BASH_DECL_SBRK - BASH_MISC_SPEED_T -> BASH_CHECK_SPEED_T - BASH_CHECK_SOCKLIB -> BASH_CHECK_LIB_SOCKET - new macro, BASH_CHECK_SYS_SIGLIST, encapsulates all the checks for sys_siglist, _sys_siglist, and strsignal(), sets SIGLIST_O to siglist.o if appropriate Makefile.in - use SIGLIST_O variable to decide whether or not we need siglist.o {execute_cmd,subst}.c - change a couple of instances of ISDIGIT to DIGIT, where we really, really only want ascii digits ansi_stdlib.h - don't need a declaration for atol() 10/18 ----- aclocal.m4 - new macro, BASH_FUNC_PRINTF_A_FORMAT, checks for printf support for %a, %A conversion specifiers, defines HAVE_PRINTF_A_FORMAT if successful configure.in - call AC_CHECK_FUNCS for isascii - call BASH_FUNC_PRINTF_A_FORMAT config.h.in - add a define for HAVE_ISASCII - add a define for HAVE_PRINTF_A_FORMAT lib/sh/snprintf.c - for long double output, fall back to sprintf using ldfallback() function for floating point formats - support %a, %A using dfallback() or ldfallback() if HAVE_PRINTF_A_FORMAT is defined - fix bug in vasprintf that returned wrong value in its first argument if the buffer holding the result string got reallocated - fixed PUT_CHAR macro to increment the counter even if we've exceeded the buffer size, for the return value from vsnprintf/snprintf - fix vsnprintf_internal to not use counter < length as a loop condition, but always process the entire format string (for the return value from vsnprintf/snprintf) builtins/printf.def - support %a, %A if HAVE_PRINTF_A_FORMAT is defined include/typemax.h - new file, with the TYPE_MAXIMUM stuff that's duplicated in several files in lib/sh lib/sh/{fmtulong,strtol,snprintf}.c - include instead of having the definitions in each file lib/sh/Makefile.in - updated dependencies for typemax.h 10/22 ----- configure.in - call AC_CHECK_FUNCS on ctype.h functions/macros that bash redefines in chartypes.h config.h.in - defines for HAVE_IS{ASCII,BLANK,GRAPH,PRINT,SPACE,XDIGIT} include/chartypes.h, lib/glob/strmatch.c, lib/readline/chardefs.h - don't redefine some is* ctype macros/functions if HAVE_ISXXX is defined (meaning that an appropriate function, but not a macro, exists) lib/sh/strtrans.c - new function, ansic_shouldquote, returns 1 if argument string contains non-printing chars that should be quoted with $'...' externs.h - new declaration for ansic_shouldquote() variables.c - change print_var_value to ansi C quote the string if we're not in posix mode and the variable's value contains non-printing chars, to use the regular shell single quoting if the value contains shell meta-characters, and to just output the string otherwise lib/sh/shquote.c - add `break' to `case '~':' to avoid fallthrough and extra test doc/bashref.texi - note that in POSIX mode, `set' displays variable values that include nonprinting characters without quoting, unless they contain shell metacharacters builtins/printf.def, lib/sh/snprintf.c - handle `F' conversion specifier as equivalent to 'f' parse.y, {nojobs,variables}.c - a couple of cleanups for when building a minimal configuration nojobs.c - new function: stop_making_children(), just sets already_making_children to 0 (like stop_pipeline) subst.c - call stop_making_children from subst.c:command_substitute if JOB_CONTROL is not defined. This fixes the bug where the wrong process is waited for (and its status returned) when using command substitution in a null command in a shell function builtins/printf.def - new variable `tw' used to keep track of the total number of characters written by a single call to `printf' -- to be used for the `%n' conversion, which will be added later. It gets reset each time we reuse the format string, which is what ksh93 seems to do 10/23 ----- variables.c - new function, bind_var_to_int (char *var, long val) variables.h - extern declaration for bind_var_to_int lib/sh/netopen.c - use gai_strerror() for error messages when getaddrinfo() fails - use PF_INET if DEBUG is defined, since IPv6 doesn't work for me Makefile.in - pass DEBUG=${DEBUG} down to makes in some subdirectories {builtins,lib/{glob,sh}}/Makefile.in - append ${DEBUG} to LOCAL_CFLAGS value, passed by top-level Makefile builtins/printf.def - added support for %n format conversion char (number of chars printed so far from current format string) 10/24 ----- variables.c - if posixly_correct is set, the default value of $MAILCHECK is 600 - use legal_number instead of atoi in adjust_shell_level - treat non-numeric assignments to SECONDS as 0 in assign_seconds - new function, init_funcname_var; sets FUNCNAME as a dynamic variable if it's not set in the initial environment - new function, init_groups_var; sets GROUPS as a dynamic array variable if it's not set in the initial environment - new function, init_dirstack_var; sets DIRSTACK as a dynamic array variable if it's not set in the initial environment - new function, init_seconds_var; sets SECONDS as a dynamic variable using any valid integer value in the initial environment as the initial value, as if an assignment had been performed - call init_funcname_var, init_groups_var, init_dirstack_var, init_seconds_var from initialize_dynamic_variables - non-numeric values assigned to LINENO are treated as 0 - change initialize_shell_variables to not auto-export PATH or TERM - change set_home_var to not auto-export HOME - change set_shell_var to not auto-export SHELL - broke the code that sets HOSTNAME, HOSTTYPE, MACHTYPE, OSTYPE out into a separate function, set_machine_vars; none of those variables are auto-exported - bash no longer un-exports SSH_CLIENT or SSH2_CLIENT shell.c - changed isnetconn() to check SSH_CLIENT and SSH2_CLIENT only if SSH_SOURCE_BASHRC is defined in config-top.h config-top.h - added a commented-out definition for SSH_SOURCE_BASHRC 10/25 ----- Makefile.in - changed RELSTATUS to `rc1' (release candidate 1) 10/29 ----- locale.c - fixed an `=' vs. `==' typo in set_locale_var when parsing LC_NUMERIC doc/{bash.1,bashref.texi} - document what bash does with $POSIXLY_CORRECT doc/builtins.1 - some updates builtins/psize.sh - some mktemp(1) changes lib/readline/readline.c - change rl_backward to check for rl_point < 0 and reset to 0 if so lib/readline/util.c - don't compile in _rl_strpbrk if HAVE_STRPBRK is defined lib/readline/rlprivate.h - remove extern declaration of _rl_strpbrk lib/readline/rldefs.h - #define _rl_strpbrk as strpbrk if HAVE_STRPBRK is define, otherwise add extern declaration of _rl_strpbrk from rlprivate.h {mailcheck,shell,variables}.c - make sure to include posixtime.h to get any prototype for time(3) in scope {array,eval,execute_cmd,mksyntax,subst}.c, parse.y builtins/common.c lib/sh/pathcanon.c - a few changes as the result of `gcc -Wall' patches from solar designer builtins/read.def, parse.y - change some calls to free() to xfree() builtins/set.def - make sure unset_builtin() resets unset_array to 0 each time through the loop, because it's set (and used) depending on the current argument shell.h - new define, USE_VAR, to force the compiler to not put a particular variable in a register -- helpful if registers are not restored by setjmp/longjmp builtins/{evalfile.c,{read,wait}.def}, {eval,execute_cmd,shell,test}.c - use USE_VAR for some variables subst.c - fixed a case in expand_word_internal where a NULL pointer could have been passed to free() (though free() should ignore it) - fixed a case at the end of expand_word_internal where LIST could have been used uninitialized (it makes gcc happy, though it doesn't happen in practice) test.c - give test_syntax_error(), beyond(), and integer_expected_error() the `__noreturn__' attribute for gcc unwind_prot.c - in clear_unwind_protect_list(), convert `flags' to `long' (via assignment to a `long' variable) before casting to `char *', in case pointers and longs are 64 bits and ints are 32 (makes no difference on 32-bit machines) 10/30 ----- print_cmd.c - fixed cprintf to avoid gcc warning about assigning const pointer to non-const (discarding type qualifier) {make_cmd,pcomplete,test}.c,parse.y - some minor changes to shut up gcc warnings lib/sh/tmpfile.c - fixed sh_mktmpfp to avoid file descriptor leaks in the case that sh_mktmpfd succeeds but fdopen fails for some reason - change sh_mktmpfd to use the same scheme for computing `filenum' as sh_mktmpname - change get_sys_tmpdir to prefer P_tmpdir if P_tmpdir is defined - changed sh_mktmpname and sh_mktmpfd to avoid trying to assign to `nameroot' if `nameroot == 0' (duh) - add code to sh_mktmpfd to use mkstemp(3) if USE_MKSTEMP is defined - add code to sh_mktmpname to use mktemp(3) if USE_MKTEMP is defined support/{fixlinks,mkclone} - use mktemp if it's available for the symlink test - use $TMPDIR instead of hardcoding /tmp; default to /tmp - use a better filename for the symlink test instead of `z' support/bashbug.sh - more changes inspired by a patch from solar designer lib/malloc/Makefile.in - new target `alloca', which builds libmalloc.a with alloca.o only (for systems without alloca that are configured --without-bash-malloc) configure.in - if we don't have a working alloca and are not configured to build the bash malloc library, make a malloc library containing only alloca.o aclocal.m4 - slight change to RL_LIB_READLINE_VERSION to deal with minor version numbers with a letter appended (like 4.2a) 10/31 ----- doc/{bash.1,bashref.texi} - slight change to note that only interactive shells resend a SIGHUP to all jobs before exiting externs.h - declare strto[ui]max only if NEED_STRTOIMAX_DECL is defined. This keeps picky compilers from choking because intmax_t is not defined (MacOS X 10.1) builtins/printf.def - #define NEED_STRTOIMAX_DECL before including shell.h 11/1 ---- general.c - check in bash_tilde_expand() for an unquoted tilde-prefix; don't bother passing the string to tilde_expand unless the prefix is unquoted shell.c - fix a problem with $LINENO when executing commands supplied with the -c invocation option when ONESHOT is defined [bash-2.05a-rc1 frozen] builtins/printf.def - fix the %n conversion to require that the variable name supplied be a valid shell identifier variables.c - improve random number generator slightly by using the upper 16 bits of the running random number instead of the lower 16, which are incrementally more random 11/2 ---- configure.in - if RL_INCLUDEDIR ends up being /usr/include, don't put -I$(RL_INCLUDEDIR) into CFLAGS 11/5 ---- doc/{bash.1,bashref.texi} - correct description of POSIXLY_CORRECT to note that the shell enters posix mode *before* the startup files are read if POSIXLY_CORRECT is in the initial environment variables.c - fix function prologues for init_dirstack_var and init_groups_var to agree with caller (no arguments) jobs.c - fix forward function declarations for pipe_read and pipe_close subst.c - removed `inline' attribute from skip_double_quoted because it can potentially be called recursively bashline.c - quick fix to bashline.c:attempt_shell_completion programmable completion code to just punt if the end of the command word found by find_cmd_end is <= the start found by find_cmd_start (the bug is probably in find_cmd_start -- fix later) pcomplete.c - fix gen_matches_from_itemlist to return if the stringlist is null after any cleaning or initialization, before trying to use it - fix GEN_COMPS to only bother to try to append the STRINGLIST returned by gen_matches_from_itemlist to `glist' if it's non-NULL lib/sh/stringlist.c - make copy_stringlist return NULL if the STRINGLIST * passed as an argument is NULL - make append_stringlist call copy_stringlist only if M2 is non-NULL; otherwise just return NULL if m1 is NULL - make word_list_to_stringlist return 0 immediately if the passed LIST argument is NULL - make realloc_stringlist call alloc_stringlist if the passed STRINGLIST argument (`sl') is 0, just like realloc calls malloc subst.c - in skip_to_delim(), if we have an unclosed ${, and it's at the end of the string (string[i] == '{', string[i+1] == '{' and string[i+2] == 0, return si (i +2) immediately without bothering to call extract_dollar_brace_string or extract_delimited_string - in skip_to_delim(), if string[i] is 0 after a call to extract_dollar_brace_string or extract_delimited_string (meaning we have an unclosed ${ or other expansion, return i immediately without doing a `continue' (which will increment i past the end of string) - in split_at_delims, don't increment te by 1 if it's pointing to a delimiter. this has the effect of skipping the first delimiter char in a possibly multi-character delimiter, and ignoring single-char delimiters like `>' configure.in - use AC_CHECK_MEMBERS([struct stat.st_blocks]) instead of a call to AC_STRUCT_ST_BLOCKS to avoid configure changing LIBOBJS if the test fails general.c - introduce two new variables: bash_tilde_{prefixes,suffixes}, set to the additional prefixes and suffixes bash wants to pass to the tilde expansion code (reserved for post-bash-2.05a fix) aclocal.m4 - add missing `test' in BASH_CHECK_SYS_SIGLIST 11/7 ---- lib/readline/vi_mode.c - fix rl_vi_goto_mark to explicitly check that the desired mark is between 'a' and 'z', since some locales have lowercase letters outside that range, which could cause a negative subscript include/chartypes.h - remove superfluous `#undef ISASCII' lib/sh/strto[iu]max.c - changes from Paul Eggert to work around buggy compilers and catch configuration errors at compile time aclocal.m4 - new macro, BASH_C_LONG_DOUBLE, identical to AC_C_LONG_DOUBLE but with a fix for Irix 5.3 (not called, since I'm not sure it's the right thing to do -- the C standard allows double and long double to be the same size) lib/sh/snprintf.c - only try to write the trailing NUL in vsnprintf_internal if data->length is >= 0, since if it's not, we probably don't have a buffer Makefile.in - changed RELSTATUS to `release' 11/8 ---- lib/sh/strtol.c - make sure chars passed to toupper are cast to unsigned unwind_prot.c - change clear_unwind_protect_list to not require a cast from `int' to `char *' lib/readline/chardefs.h - make _rl_digit_p succeed only for ascii digits, since that's what most callers assume 11/13 ----- doc/bashref.texi - added `ERR' trap and [-+]O invocation option to section listing differences from the Bourne shell 11/15 ----- [bash-2.05a released] 11/19 ----- include/stdc.h - new define, INLINE, defined as `inline' for gcc and empty otherwise subst.c - make skip_double_quoted, sub_append_string have INLINE attribute trap.c - use BASH_NSIG as upper limit for signal names in signal_name() lib/readline/bind.c - use RL_COMMENT_BEGIN_DEFAULT in output for rl-comment-begin value error.c - fix sys_error to save value of errno around calls to fprintf doc/Makefile.in - added rules to create PDF files from postscript and dvi input MANIFEST.doc - added {article,bash,bashref,rose94}.pdf doc/bash.1 - rearranged some `.PD 0' and `.TP' directives so man2html will handle them better (shouldn't affect groff output) support/man2html.c - small fix to handle quoted string arguments to directives like `.BR' without mangling the output 11/20 ----- {arrayfunc,variables}.c - changed calling sequence for dynamic array variable `assign' functions to (SHELL_VAR *self, char *value, arrayind_t ind) - changed calling sequence for dynamic variable assign functions to the same as array variable assign_func. Now this can be prototyped variables.h - the assign_func member of a `struct variable' is now of type `sh_var_assign_func_t', which is prototyped - the dynamic_value member of a `struct variable' is now of type `sh_var_value_func_t', which is prototyped variables.c - changed to use `sh_var_assign_func_t' and `sh_var_value_func_t' builtins/cd.def - when in posix mode, if the new directory name formed by PWD and the argument passed by the user cannot be canonicalized, and the -P option has not been supplied, return failure immediately - if canonicalization failed, but the fallback to the directory name specified by the user succeeds, reset the current working directory lib/readline/{input.c,rlprivate.h} - renamed rl_unget_char to _rl_unget_char; made library global lib/readline/{{bind,readline}.c,{keymaps,rlprivate}.h} - support for `key subsequences'; allows a key sequence and a function mapped to a subsequence of that key sequence. Primarily to allow arrow keys to be bound in readline vi insert mode, while preserving the ESC function to switch to command mode. lib/readline/{input.c,rlprivate.h} - new function, _rl_input_queued(T), does a check with select or FIONREAD with a timeout of `T' (which is generally 0) lib/readline/readline.c - change _rl_dispatch_subseq to test for input in the queue if we get ESC while in vi insertion mode if the keymap entry type for ESC is ISKMAP. If _rl_input_queued returns non-zero, we assume that an arrow key sequence has been pressed and go ahead with the subsequence. If it returns zero, we assume that the user pressed ESC to switch into command mode, and dispatch to that right away. This avoids forcing the user to press another key before switching into command mode 11/21 ----- lib/readline/readline.c - bind common arrow key sequences in vi insertion keymap lib/readline/terminal.c - bind termcap definition's arrow keys in vi insertion keymap lib/readline/bind.c - check for rl_vi_movement_mode in _rl_bind_if_unbound, so binding the arrow keys can work lib/readline/readline.c - since _rl_bind_if_unbound does the check of what's currently bound to the key sequence, the check in bind_arrow_keys_internal was redundant - bind_arrow_keys_internal now takes a Keymap argument and handles saving and restoring _rl_keymap; changed bind_arrow_keys accordingly builtins/fc.def - fix from Paul Eggert to substitute the nearest history number in range if an out-of-range value is supplied. POSIX requires this lib/sh/pathcanon.c - fix from Corrina Vinschen for the special `cygdrive' prefix on Cygwin bashhist.c - split the history adding code into more pieces: check_history_control (char *line) checks LINE against the value of HISTCONTROL, returning 1 if LINE should be saved and 0 if not check_add_history (char *line) calls check_history_control and history_should_ignore (line) and saves the line with bash_add_history if the checks indicate that it should be saved maybe_add_history just calls check_add_history to set the value of first_line_saved bashhist.h - extern declaration for check_add_history() shell.c - don't call load_history() from the interactive shell startup code if history_lines_this_session is > 0, indicating that we've already saved some lines in the history and that we probably don't want to overwrite them builtins/history.def - call check_add_history from push_history, so `history -s xx' works even when in a compound command whose first line has not been saved. (Caveat: in a compound command when the first line has been saved, the line supplied to history -s will become part of the compound command's history entry. Of course, the delete_history call could remove the compound command from the history entirely) bashline.c - use sh_makepath instead of xmalloc/sprintf in command_word_completion_function lib/readline/complete.c - get_y_or_n now takes an int FOR_PAGER argument; caller changed If FOR_PAGER is non-zero, get_y_or_n returns appropriate values for a more-like pager: `newline' or `return' return 2; `q' or `Q' return 0 - there is now a mini internal more-like pager for displaying a list of completions that exceeds the screen height (new function _rl_internal_pager, called from rl_display_match_list) 11/24 ----- command.h - new flag, W_TILDEEXP, says to do tilde expansion on an assignment word execute_cmd.c - fix_assignment_words now sets W_TILDEEXP for assignment word arguments to `assignment builtins' general.c - bash_tilde_expand now takes a second argument indicating whether or not it's being invoked in an `assignment context' general.h - change extern declaration for bash_tilde_expand {bashline,execute_cmd,findcmd,general,variables}.c builtins/evalfile.c lib/sh/makepath.c - fix callers of bash_tilde_expand appropriately subst.c - fix callers of bash_tilde_expansion appropriately - add (currently commented-out) code that would tilde expand assignment statement arguments to assignment builtins (W_TILDEEXP flag set) even when the shell is in posix mode bashline.c - fix attempt_shell_completion to turn off rl_filename_completion_desired when doing command name completion, so no slash gets appended to the name if there happens to be a directory with the same name in the current directory 11/26 ----- lib/readline/rltech.texinfo - a couple of additions to the rl_stuff_char description parse.y - turn off echo_input_at_read in parse_string_to_word_list, so `set -v' doesn't give extra lines of output when doing compound array assignment subst.c - fix split_at_delims to handle skipping over a `\n' if it's a delimiter (use spctabnl(c) instead of whitespace(c)) 11/27 ----- support/config.{guess,sub} - updated (with bash changes) to latest version from gnu.org sig.h - add prototype for set_signal_handler declaration builtins/setattr.def - add prototype to extern declaration of declare_builtin builtins/times.def - add no_options call, since times takes no options lib/sh/spell.c - add prototypes to forward declarations for midist and spdist lib/sh/strtrans.c - add explicit int return type to ansic_shouldquote declaration lib/readline/rldefs.h, lib/readline/{macro,readline,util,undo}.c - move define for SWAP to rldefs.h, removed from various C files lib/readline/vi_mode.c - removed define for exchange(), changed to use SWAP instead lib/readline/bind.c - added some static forward function declarations - find_boolean_var, find_string_var now take a `const char *' argument lib/readline/signals.c - added static forward declaration for rl_maybe_set_sighandler lib/readline/readline.c - add some common key bindings for the HOME and END keys in bind_arrow_keys_internal lib/readline/terminal.c - fetch the `@7' termcap string; it's sent by the END key - attempt to bind the terminal's END key to rl_end_of_line in bind_termcap_arrow_keys; I don't know why I was using `kH' instead of `@7' doc/builtins.1 - remove `case', `for', `if', `until', `while' from NAME section; those are not shell builtins 11/28 ----- stringlib.c - new function, find_token_in_alist, takes a token value and an ALIST argument, and returns the string correspoinding to the token if found in the alist externs.h - new extern declaration for find_token_in_alist() subst.c - string_list_internal is no longer static subst.h - new extern declaration for string_list_internal() parse.y - new alist array of other tokens returned by read_token which are not reserved words in word_token_alist[] - reworked error reporting: new functions print_offending_line, which prints the line containing the syntax error, error_token_from_token, which takes the current token and tries to figure out its textual representation, and error_token_from_text, which does the old job of finding the bad token by analyzing the text of shell_input_line at the current index - report_syntax_error now tries to figure out the token that caused the syntax error by first looking at current_token and falling back to the old method of textual analysis if that fails - report_syntax_error doesn't say the token resulting from the textual analysis of the input line is an `unexpected token'; it just says there is a `syntax error near xxx' - changed conditional command error reporting to use the value returned by error_token_from_token if it's not null instead of just using the token value in the message, since current_token ends up being set to -1, and the text of the message from report_syntax_error might not be exactly right - change parse_string_to_word_list to set current_token to the offending token returned by read_token before calling yyerror() to make the error reporting do the right thing aclocal.m4 - fixed typo in BASH_CHECK_LIB_TERMCAP configure.in - add check for isinf(3); define HAVE_ISINF_IN_LIBC if found config.h.in - add define for HAVE_ISINF_IN_LIBC lib/sh/snprintf.c - check for Inf and NaN, using isinf and isnan if they're found in libc - use the current locale for thousands separator and decimal point - recognize "'" flag; not implemented yet - fix for snprintf/vsnprintf with length of 0 and string argument of 0 with non-zero length builtins/read.def - TMOUT is now the default timeout for `read' (and select) if set, like ksh93 when reading from the terminal - edit_line (called by read -e) now just does readline's filename completion by setting rl_attempted_completion_function to NULL, since e.g., doing command completion for the first word on the line wasn't really useful execute_cmd.c - changed select_command to return failure status if select_query returns NULL, indicating that read_builtin returned EXECUTION_FAILURE doc/{bash.1,bashref.texi} - documented new TMOUT behavior - slight change to the description of the test `-ef' option doc/bashref.texi - added item to posix mode section describing failure behavior of cd when invoked in logical mode and the pathname formed by combining $PWD and the directory argument does not refer to an existing directory 11/29 ----- execute_cmd.c - fix execute_function to call dispose_function_env after merge_function_env if the shell is in posix mode (fixes debian bash bug #117673) lib/readline/readline.c - rl_forward -> rl_forward_char; rl_forward function for compatibility - rl_backward -> rl_backward_char; rl_forward function for compatibility - new functions, rl_forward_byte, rl_backward_byte, for future use lib/readline/readline.h - extern declarations for rl_forward_char, rl_backward_char, rl_forward_byte, rl_backward_byte lib/readline/{emacs_keymap,funmap,vi_keymap,vi_mode - rl_forward -> rl_forward_char - rl_backward -> rl_backward_char lib/readline/funmap.c - new bindable names, `backward-byte' and `forward-byte' aclocal.m4 - new function, BASH_CHECK_MULTIBYTE, encapsulates checks for multibyte code config.h.in - add necessary defines for multibyte include files and functions configure.in - add call to BASH_CHECK_MULTIBYTE config-bot.h - add code to define HANDLE_MULTIBYTE if prerequisites are met lib/sh/xstrchr.c - new file, xstrchr() is strchr(3) that handles multibyte characters bashhist.c - first_line_saved -> current_command_first_line_saved; variable is now global bashhist.h - extern declaration for current_command_first_line_saved 11/30 ----- bashhist.c - break the code that actually calls add_history out of bash_add_history into a new function, really_add_history; bash_add_history now calls really_add_history - check_add_history takes a second `force' argument telling it whether to call bash_add_history (force == 0) or really_add_history (force != 0) builtins/history.def - in push_history, call delete_last_history if the current command has more than one line, the first line was saved, and command-oriented history is active. This takes care of deleting the right history element if `history -s' is used within a compound or multiline command - in push_history, call check_add_history with second argument of 1 to skip check of current_command_line_count and add the arguments to history -s as a single separate history entry 12/3 ---- lib/readline/complete.c - append a slash to completed names which are symlinks to directories if the new variable _rl_complete_mark_symlink_dirs is non-zero lib/readline/rlprivate.h - extern declaration for _rl_complete_mark_symlink_dirs lib/readline/bind.c - new bindable variable, `mark-symlinked-directories', mirrors the value of _rl_complete_mark_symlink_dirs doc/bash.1, lib/readline/doc/{readline.3,rluser.texinfo} - documented new `mark-symlinked-directories' variable 12/4 ---- variables.[ch] - set_pipestatus_array now takes a second argument with the number of processes in the array - changed set_pipestatus_array to just modify the value in place if the existing array has one element and the new array has one element, and to modify existing values in place if new array has more elements than existing array variables.c, jobs.c - changed set_pipestatus_array callers jobs.c - moved call to setjstatus() from set_job_status_and_cleanup to wait_for, since set_job_status_and_cleanup is part of the SIGCHLD signal handler call path, and race conditions accessing the PIPESTATUS array will result for things like while true; do date; done | cat > /dev/null 12/5 ---- xmalloc.h - don't redefine xmalloc, xrealloc, and xfree if DISABLE_MALLOC_WRAPPERS is #defined config.h.in - #undef for DISABLE_MALLOC_WRAPPERS configure.in - define DISABLE_MALLOC_WRAPPERS if the --with-purify option is supplied lib/malloc/trace.c - new function, malloc_trace_bin(N), traces allocations and frees to bucket N (uses the same type of bitmap as `busy') lib/malloc/table.c - fix wraparound search problem in find_entry when searching for a free entry when the table is full 12/6 ---- lib/malloc/table.c - keep an `overflow bucket' around to use when the table is full, so find_entry always returns a valid pointer when FIND_ALLOC is set - new static variable to keep a count of the number of MT_ALLOC entries in the mem_table lib/sh/{oslib,clktck}.c - if HAVE_LIMITS_H is defined, include lib/sh/oslib.c - new function, getmaxgroups() returns max number of simultaneous groups - new function, getmaxchild(), returns max number of simultaneous user processes general.c - removed forest of #defines for getmaxgroups() externs.h - new extern declaration for getmaxgroups() - new extern declaration for getmaxchild() - new extern declaration for isnetconn() lib/sh/netconn.c,shell.c - new file, isnetconn() from shell.c moved here Makefile.in, lib/sh/Makefile.in - necessary changes for netconn.c builtins/ulimit.def - changed getmaxuprc() to just call getmaxchild() and massage the return value appropriately {jobs,nojobs}.c - use the value returned by getmaxchild() in mark_dead_jobs_as_notified instead of static CHILD_MAX jobs.c - new function, compact_jobs_list, removes some number of jobs from the jobs table and reallocates the table, copying the jobs that are left from the old table to the new. Compaction happens from the beginning of the list and removes dead jobs, and we make sure to keep the last CHILD_MAX jobs as POSIX.2 requires - call compact_jobs_list from stop_pipeline if we're in a subshell, there are no free jobs in the jobs table, and the jobs table is at or above some maximum limit execute_cmd.c - change eval_arith_for_expr to set this_command_name to `((' before calling evalexp, since it might be changed by evaluating the loop body between evalexp calls trap.c - change reset_signal to turn off the SIG_TRAPPED flag for the given signal, so shell builtins and functions running in command substitutions don't run the signal handlers (traps are not supposed to be inherited by command substitutions) parse.y - changed parse_string_to_word_list to turn off alias expansion while parsing the array assignment 12/9 ---- alias.c - fix add_alias so that redefining an alias's value also resets the EXPANDNEXT flag 12/10 ----- parse.y - new function, token_is_assignment, called to check whether the text before `=' makes up a valid assignment token before trying to parse a compound assignment statement - new function, parse_compound_assignment, to parse a compound assignment statement instead of using parse_matched_pair; handles comments and error reporting in the parser instead of waiting until expansion time - changed parse_compound_assignment and parse_string_to_word_list to allow reserved words in compound array assignments lib/readline/doc/rltech.texinfo - changed the documentation for rl_callback_read_char and rl_callback_handler_remove to say what happens to the terminal settings and what needs to be done to reset them 12/11 ----- bashline.c - add emacs_edit_and_execute_command, bound to C-xC-e, like vi-mode `v' command - add bindable command name `edit-and-execute-command', bound to run emacs_edit_and_execute_command() lib/glob/strmatch.c - add support for ksh93-like [:word:] character class (isalnum + `_') doc/{bash.1,bashref.texi} - add note to section describing lists to clarify that a sequence of one or more newlines may be used to delimit a command, equivalent to a semicolon - document new [:word:] pattern matching character class doc/bash.1, lib/readline/doc/rluser.texinfo - document `edit-and-execute-command' and its default emacs-mode binding include/chartypes.h - add defines for TOCTRL and UNCTRL if they're not already defined lib/readline/chardefs.h - #undef UNCTRL if it's defined to avoid cpp redefinition warnings lib/sh/strtrans.c - add \cX (Control-X) escape for $'...' to ansicstr() - change ansic_quote() to allocate at least four chars for each char in the string argument, to account for \0xx octal values - change ansic_quote() to no longer call sprintf for non-printable characters; just translate the string to octal directly print_cmd.c - change xtrace_print_word_list to call ansic_quote() if ansic_shouldquote() indicates that there are nonprinting characters in a word builtins/type.def - changed deprecated long option parsing to just replace the word in the list with the equivalent short option (-type -> -t) instead of removing words from the list - changed describe_command to take a single flags argument instead of two int args; changed caller - type now has two new options: -f suppresses function lookup (like command), and -P forces a PATH search for the name(s) builtins/common.h - flags for describe_command are here - changed extern declaration of describe_command builtins/command.def - changed call to describe_command to use flags from common.h, and the right number of arguments doc/{bash.1,bashref.texi} - documented new -f and -P options to `type' 12/12 ----- lib/readline/rldefs.h - fixed prototype for _rl_strnicmp execute_cmd.c - select_query now takes a new argument, an int flag saying whether or not to print the menu the first time through the loop. An empty line in response to the prompt will always cause the menu to be reprinted - changed execute_select_command to cause select_query to reprint the menu only if REPLY is set to NULL, if KSH_COMPATIBLE_SELECT is defined config-top.h - define KSH_COMPATIBLE_SELECT, with a comment about its meaning lib/readline/readline.c - change rl_insert_comment to toggle if given an explicit numeric argument: if the first characters on the line don't specify a comment, insert one; if they do, delete the comment text doc/bash.1, lib/readline/doc/{readline.3,rluser.texinfo} - documented new behavior of insert-comment with a numeric argument 12/13 ----- lib/malloc/watch.c - new file, implements watchpoint functions lib/malloc/watch.h - new file, define some `events' for watchpoints and extern function and variable declarations for watchpoint code lib/malloc/imalloc.h - #define MALLOC_WATCH if MALLOC_DEBUG is defined - add __P define as in include/stdc.h if not already defined lib/malloc/malloc.c - remove __P define, now in imalloc.h - include watch.h if MALLOC_WATCH is defined - added calls to _malloc_ckwatch in internal_malloc, internal_free, and internal_realloc include/stdc.h - augment __P define to allow prototypes if PROTOTYPES is defined lib/readline/rlstdc.h - augment PARAMS define to allow prototypes if PROTOTYPES is defined lib/malloc/Makefile.in, Makefile.in necessary changes to include watch.c in libmalloc lib/readline/readline.c - fix rl_delete_text to make sure that the starting position is >= 0 - _rl_init_line_state (called by readline via readline_initialize) now sets rl_mark to 0 - rl_get_{next,previous}_history set rl_mark to 0 if rl_point is at the end of the line and rl_end otherwise in emacs mode lib/readline/kill.c - rl_yank_nth_arg_internal and rl_paste_clipboard now set the mark at point before calling rl_insert_text, like rl_yank - rl_kill_full_line now resets rl_mark to 0 - rl_kill_line and rl_backward_kill_line now set rl_mark to the point after the kill in emacs mode - rl_kill_word and rl_backward_kill_word now set rl_mark to the point after the kill in emacs mode - rl_unix_word_rubout and rl_unix_line_discard now set rl_mark to the point after the kill in emacs mode lib/readline/search.c - noninc_search saves and restores the mark, since it can be changed while reading the search string - noninc_dosearch sets the mark at the end of the line, making the region bound the `inserted' text since rl_point is set to 0 - rl_history_search_internal sets the mark at the end of the line, for the same reason lib/readline/isearch.c - rl_search_history now saves and restores the mark - if no matching lines are found at all when doing an isearch, leave point where it was instead of moving it to the end of the line 12/17 ----- lib/readline/rlmbutil.h - new file, place for multi-byte character defines and extern declarations lib/readline/{bind.c,readline.c,rlprivate.h} - new bindable variable, `byte-oriented', tracks value of rl_byte_oriented variable lib/readline/mbutil.c - new file, with multibyte char utility functions lib/readline/{complete,display,readline,util,vi_mode}.c - new code for multibyte characters, derived from IBM patch 12/18 ----- lib/sh/tmpfile.c - include posixtime.h for time() extern declaration support/bashversion.c - include if it's available lib/readline/{histexpand,input,isearch,search}.c - new code for multibyte characters, derived from IBM patch lib/readline/readline.h - include rltypedefs.h 12/19 ----- lib/readline/complete.c - slight change to mark-directories code to avoid adding a slash if point is at the end of the line (rl_line_buffer[rl_point] == '\0') and the previous character was a slash - change printable_part to not return empty pathnames, which could happen when completing filenames and a filename with a trailing slash was passed as the argument. If the portion following the trailing slash is NULL, ignore it and look for a previous slash. If there's no previous slash, just return the filename argument - new variable, rl_completion_mark_symlink_dirs, mirrors the value of (user-settable with a variable) _rl_complete_mark_symlink_dirs but may be modified by application-specific completion functions when appropriate (set in rl_complete_internal and rl_menu_complete) lib/readline/readline.h - extern declaration for rl_completion_mark_symlink_dirs pcomplete.c - if one of the actions is CA_DIRECTORY, set rl_completion_mark_symlink_dirs to indicate that we want the trailing slash (might have to relax this) lib/readline/doc/rltech.texinfo - documented rl_completion_mark_symlink_dirs variable lib/readline/doc/rluser.texinfo, doc/bash.1 - documented the fact that `complete -d' and `complete -o dirnames' force readline to append a slash to symlinks to directories builtins/enable.def - changed enable_shell_builtin to disallow enabling disabled builtins in a restricted shell doc/{bash.1,bashref.texi} - documented new enable behavior in restricted shells doc/Makefile.in - new rule to make an `RBASH' file documenting the restrictions imposed by a restricted shell expr.c - broke the code that evaluates variables and returns results out of readtok() into a new function: expr_streval() - expr_streval() now performs the standard unset variable error behavior if `set -u' has been executed and it's asked to look up an unset variable - broke the code that frees up the expression context stack into a new function: expr_unwind() variables.c - fixed bind_int_variable so it handles array element assignment, so expressions like `b[7]++' and `b[0] = 42' work right - new function, get_variable_value, returns the string value of the SHELL_VAR * passed as an argument - get_string_value now calls get_variable_value with a non-null result from find_variable 12/20 ----- lib/readline/rlmbutil.h, mbutil.c - combined _rl_find_next_mbchar and _rl_find_next_nonzero_mbchar into a single function - combined _rl_find_prev_mbchar and _rl_find_prev_nonzero_mbchar into a single function lib/readline/{display,readline,vi_mode}.c - changed callers of _rl_find_next_mbchar and _rl_find_next_nonzero_mbchar lib/readline/{complete,display,histexpand,readline,vi_mode}.c - changed callers of _rl_find_prev_mbchar and _rl_find_prev_nonzero_mbchar 12/20 ----- lib/sh/mktime.c - new file, from glibc/gawk, compiled in if system doesn't have a working mktime(3) lib/sh/strftime.c - new file, from gawk, compiled in if system doesn't have a working strftime(3) lib/sh/Makefile.in, Makefile.in - changes for mktime.c, strftime.c configure.in - call AC_FUNC_MKTIME, AC_STRUCT_TM, AC_STRUCT_TIMEZONE - call AC_REPLACE_FUNC(strftime) config.h.in - add defines for TM_IN_SYS_TIME, HAVE_TZSET, HAVE_TM_ZONE, HAVE_STRUCT_TM_TM_ZONE, HAVE_STRFTIME externs.h - provide an extern declaration for strftime if HAVE_STRFTIME is not defined and NEED_STRFTIME_DECL is lib/tilde/tilde.h - header files should not include parse.y - replace code in decode_prompt_string that chops up value returned by ctime(3) with calls to strftime -- as a result, the expansion of \@ has changed slightly (since it depends on the locale) - added new \D{format} prompt string escape; `format' is passed to strftime(3). Empty format is the same as `%X' (locale-specific representation of the current time) - combined cases for '\\', '\a', '\e', and '\r' in same case branch in decode_prompt_string doc/{bash.1,bashref.texi} - documented new \D{format} prompt string expansion builtins/printf.def - use ISO C PRIdMAX instead of INTMAX_CONV - pass length of format modifiers to mklong instead of computing it with strlen() lib/sh/{fmtulong,fmtullong}.c - changes from Paul Eggert to make more general arrayfunc.c - when converting a variable to an array, make sure to unset the dynamic_value and assign_func members of the struct variable, since they're not valid anymore 12/27 ----- configure.in - use AC_HELP_STRING in AC_ARG_WITH and AC_ARG_ENABLE - remove AC_ARG_ENABLE for largefile, since AC_SYS_LARGEFILE adds one 1/2/2002 -------- {alias,bashline,execute_cmd,general,shell,subst,variables,arrayfunc}.c,general.h - changed some calls to strchr to calls to xstrchr for multibyte characters include/shmbutil.h - add extern declaration for xstrchr to avoid including externs.h where it's not appropriate {braces,make_cmd,pathexp,subst,arrayfunc}.c, lib/sh/xstrchr.c - include shmbutil.h {stringlib,subst}.c, {externs,subst}.h - moved substring() from subst.c to stringlib.c, moved declaration from subst.h to externs.h lib/sh/xmbsrtowcs.c - new file, replacement function for mbsrtowcs lib/sh/Makefile.in - add entries for xmbsrtowcs.c Makefile.in - add dependencies on shmbutil.h to appropriate object files lib/glob/strmatch.c - break character-class testing out into separate function: is_cclass, in prep for multibyte changes {braces,make_cmd}.c - changes for multibyte characters builtins/printf.def - changes from Paul Eggert to just use intmax_t everywhere an int/long/quad is needed and print with "%ld" if the number fits in a long and %PRIdMAX otherwise - remove getlong, getulong, getllong, getullong, since they're no longer needed - use a new type `floatmax_t' to print floating point numbers, the widest-available floating point type (like `intmax_t'); new function `getfloatmax' that calls strtold or strtod as appropriate - remove getdouble, getldouble, since they're no longer needed lib/sh/fmtumax.c - new file, string-to-[u]intmax_t conversion, just includes fmtulong.c with the right defines Makefile.in, lib/sh/Makefile.in - additions for fmtumax.c bashtypes.h - include if it's available expr.c - arithmetic is now in intmax_t instead of long externs.h - extern declaration for fmtumax - change extern declarations for evalexp, itos, inttostr, uitos, uinttostr since they now return or use intmax_t instead of long {execute_cmd,general,mailcheck,subst,variables}.c, parse.y {array,general,subst,test,variables}.h lib/sh/{itos,netopen}.c builtins/{bashgetopt,common}.c, builtins/common.h builtins/{break,fc,history,jobs,let,printf,pushd,read,shift,wait}.def - changes for intmax_t shell arithmetic conversion doc/{bashref.texi,bash.1} - documented long->intmax_t shell arithmetic conversion sig.c - in initialize_terminating_signals, if we've already trapped a terminating signal, don't reset the signal handler for it 1/3 --- {arrayfunc,pathexp}.c, parse.y - changes for multibyte chars parse.y, lib/sh/strtrans.c - moved ansiexpand from parse.y to lib/sh/strtrans.c parse.y, locale.c - moved mk_msgstr and localeexpand from parse.y to locale.c parse.y - new function, yy_input_name, returns name of input file from bash_input.name - broke the code that parses ((...)) constructs out of read_token into a new function, parse_dparen() externs.h - new extern declaration for ansiexpand(), mk_msgstr(), and localeexpand() input.h - new extern declaration for yy_input_name() {error,locale}.c - use yy_input_name for error and other messages execute_cmd.c - change shell_execve to make sure that the file is executable before looking at the interpreter to find out why the execve() failed (avoids misleading error message) lib/glob/glob.c - move code that matches leading `.' and skips those filenames into a separate function: skipname(), so there can be unibyte and multibyte versions of that function 1/7 --- subst.c - more changes for multibyte characters print_cmd.c - change semicolon() so it doesn't output a `;' immediately after a newline, since that results in a null command, which is a syntax error variables.c - fix indirection_level_string to turn off set -x while evaluating PS4 1/8 --- builtins/set.def - make -o options into one struct, instead of separate structs for option names corresponding to flags and non-flag option names. This has the side effect of sorting the option names in output lib/glob/glob.c - new function, mbskipname(), multibyte char version of skipname() - removed all #ifndef SHELL code, this will never be used outside the shell include/posixdir.h - move REAL_DIR_ENTRY define here from lib/glob/glob.c lib/glob/glob_loop.c - new file, included in glob.c for unibyte and multibyte versions of glob_pattern_p - added some forward static function declarations with prototypes - more changes for multibyte character handling lib/glob/Makefile.in - make glob.c depend on glob_loop.c - changes for xmbsrtowcs.[co] lib/glob/xmbsrtowcs.c - moved here from lib/sh, since the matching functions use it, and libglob.a is linked after libsh.a 1/9 --- lib/glob/smatch.c - new file, with strmatch (now xstrmatch) and associated functions, with changes for multibyte chars lib/glob/sm_loop.c - new file, included by smatch.c, with `generic' versions of matching functions that are compiled twice: once each for single-byte and multibyte characters lib/glob/strmatch.c - strip out everything except strmatch(), which either calls fnmatch (if HAVE_LIBC_FNM_EXTMATCH is defined) or xstrmatch lib/glob/collsyms.c - changes for multibyte chars lib/glob/Makefile.in, Makefile.in - changes for new source files 1/10 ---- lib/readline/complete.c - new function, rl_completion_mode (rl_command_func_t *func), returns the appropriate value to pass to rl_complete_internal depending on FUNC and the value of `show-all-if-ambiguous'. This allows application completion functions to present the same interface as rl_complete lib/readline/readline.h - new extern declaration for rl_completion_mode() lib/readline/doc/rltech.texinfo - documented rl_completion_mode lib/readline/readline.[ch] - bumped the version number to 4.3, changing the relevant cpp defines configure.in - require that an installed readline version be at least readline-4.3 bashline.c - converted bash-specific completion functions to use rl_completion_mode instead of passing TAB unconditionally builtins/bashgetopt.c - the `#' option specifier now means a required numeric argument, not an optional one builtins/type.def - when converting [-]-{path,type,all} to -[pta], don't bother freeing and reallocating the option string; just change opt[1] and null opt[2] lib/sh/snprintf.c - support %ls/%S and %lc/%C for wide strings and characters, respectively, if HANDLE_MULTIBYTE is defined mailcheck.c - don't print a message about new mail if the file has not grown, even if the access time is less than the modification time 1/14 ---- lib/readline/readline.c - new function, rl_replace_line, to replace the readline line buffer with the text supplied as an argument - new function, rl_replace_from_history, replaces readline line buffer with text from history entry passed as argument (undocumented, not in readline.h because it requires a definition of HIST_ENTRY for the prototype) lib/readline/readlne.h - new extern declaration for rl_replace_line lib/readline/doc/rltech.texinfo - documented rl_replace_line lib/readline/{isearch,readline,search}.c - use rl_replace_line and rl_replace_from_history where appropriate lib/readline/readline.c - broke the code that sets point after moving through the history (_rl_history_preserve_point and _rl_history_saved_point) out into a separate function, _rl_history_set_point() lib/readline/{complete.c,rlprivate.h} - find_completion_word -> _rl_find_completion_word - free_match_list -> _rl_free_match_list lib/readline/complete.c - postprocess_matches and _rl_free_match_list now return immediately if passed a null match list variables.c - new function, find_local_variable, finds a local variable by name at the current variable context - in find_variable_internal, call find_local_variable before searching any of the temporary environments if variable_context > 0 (meaning we're in a shell function). This lets a local variable override a variable whose value was passed in the `function environment' 1/15 ---- variables.h, execute_cmd.c - declare variables describing the temporary environments in variables.h instead of in C files findcmd.c, builtins/setattr.def - instead of calling find_tempenv_variable, use find_variable_internal and check whether the returned SHELL_VAR * has the tempvar attribute variables.c - tentative change to lookup order in find_variable_internal so that function local variables are found before variables in function_env when executing a shell function - change make_local_variable to handle making a local variable when a variable with the same name already appears in one of the temporary environments - broke the body of make_var_array out into a new function: static char **make_env_array_from_var_list (SHELL_VAR **vars) - new function, make_var_array_internal, takes a hash table to look in and a pointer to a mapping function and returns a char ** environment-style list - make_var_array now just calls make_var_array_internal - new mapping function, local_and_exported, returns all local variables in the current variable context with the export attribute set - new function, make_local_export_array, returns an environment-style char ** array of exported local variables in current context - change environment creation order in maybe_make_export_env to add variables to the environment in opposite order that find_variable_internal uses. This means that local variables in shell functions override variables with the same name in the function_env - change make_local_variable to set the initial value of the variable it creates to NULL to make the `is set' and `is null' tests that the expansion code does work right - change make_local_variable to inherit the value of a variable with the same name from the temporary enviroment 1/16 ---- Makefile.in - link bashversion with buildversion.o instead of version.o, for cross-compiling. version.o is for the target system; buildversion.o is for the build system error.c - add line numbers to internal_error() messages if the shell is not interactive and running a shell script or a -c command - report_error now prints non-zero line numbers for non-interactive shells test.c - test_syntax_error now calls builtin_error() instead of printing its own messages builtins/common.c - builtin_error now prints line numbers if a non-interactive shell is running a shell script or a -c command print_cmd.c - in cprintf, remove free_argp, since it's not used builtins/history.def - make `history -n' increment the number of history lines in this session by the number of lines read from the history file arrayfunc.c - fix array_value_internal to expand the subscript even if the variable is unset, so side effects produced by the arithmetic evaluation will take place lib/readline/doc/{rluser,rltech}.texinfo - some fixes for printing in @smallbook format from Brian Youmans 1/17 ---- jobs.h - new PRUNNING, PSTOPPED, PDEADPROC defines for PROCESSes, analogous to RUNNING, STOPPED, and DEADJOB defines for jobs jobs.c - use PS_RUNNING, PS_DONE, PS_STOPPED values for `running' field of a PROCESS - find_pipeline and find_job now take an additional flags argument that, if non-zero, means to find only running processes; changed all callers - changed calls to find_pipeline and find_job made from waitchld to find only running processes - find_pipeline takes a third argument: an int *. If it looks in the jobs list to find the pid, and the arg is non-null, it passes the job index back to the caller. Used to avoid calls to find_pipeline immediately followed by find_job with the same PID nojobs.c - a couple of changes to make sure that set_pid_status is never called with a pid argument of 0 or -1 trap.c - change trap_handler to longjmp to wait_intr_buf (set by wait_builtin) if a signal is received for which a trap has been set during execution of the wait builtin (need to include builtins.h and builtins/builtext.h and declare some extern variables for the right things to check) - new variable to keep track of which signal caused the longjmp to wait_intr_buf, set by trap_handler (wait_signal_received) builtins/wait.def - set the return value of wait when a longjmp(wait_intr_buf, 1) is done to 128 + wait_signal_received {jobs,nojobs}.c - set wait_signal_received to SIGINT in wait_sigint_handler before the longjmp(wait_intr_buf, 1) 1/18 ---- bashline.c - turn off rl_filename_completion_desired when completing a command name with a single match only if the first char of that match is not a `/' - if there are multiple identical matches for a command name in attempt_shell_completion, turn off rl_filename_completion_desired if the first char is not a `/' to avoid readline appending a slash if there's a directory with the same name in the current directory 1/22 ---- lib/readline/complete.c - new variable, _rl_page_completions, to control whether we want to run the internal pager when listing completions (defaults to 1) lib/readline/rlprivate.h - extern declaration for _rl_page_completions lib/readline/bind.c - new bindable variable, `page-completions', controls value of _rl_page_completions lib/readline/doc/{rluser.texinfo,readline.3}, doc/bash.1 - documented `page-completions' variable Makefile.in - use $(INSTALL_SCRIPT) instead of $(INSTALL_PROGRAM) to install `bashbug' aclocal.m4 - fix small quoting problem in RL_LIB_READLINE_VERSION macro lib/readline/terminal.c - fetch and save terminal's `vs' and `ve' cursor control attributes - fetch and save terminal's `kI' attribute (string sent by Insert) - new function, _rl_set_cursor, sets cursor to normal (insert mode) or very visible (overwrite mode) lib/readline/readline.c - new global variable, rl_insert_mode - new function to toggle overwrite mode, rl_overwrite_mode - each new line starts in insert mode - switching to vi mode or emacs mode resets to insert mode - reset cursor to normal before returning line - _rl_replace_text now returns the number of characters inserted, the return value from rl_insert_text - new function, _rl_insert_or_replace_text (const char *string, int insert), either inserts STRING or replaces the number of chars in STRING with STRING starting at rl_point, depending on value of INSERT - renamed rl_insert to _rl_insert_char, rl_insert just calls _rl_insert_char with the same arguments when in insert mode - new function, _rl_overwrite_char, handles self-insert in overwrite mode. Does multibyte chars by reading an entire multibyte character before entering overwrite loop - new function, _rl_overwrite_rubout, handles RUBOUT when in overwrite mode, called from rl_rubout - new function, _rl_rubout_char, old body of rl_rubout; rl_rubout calls this when not in overwrite mode lib/readline/readline.h - extern declarations for rl_insert_mode and rl_overwrite_mode() lib/readline/rldefs.h - define constants for values of rl_insert_mode lib/readline/rlprivate.h - extern declarations for _rl_set_cursor and _rl_set_insert_mode - change type of _rl_replace_text to return int - extern declarations for _rl_insert_char, _rl_rubout_char lib/readline/funmap.c - new bindable name `overwrite-mode', bound to rl_overwrite_mode lib/readline/rlconf.h - define CURSOR_MODE if you want the cursor to show insert or overwrite mode (only available if both `vs' and `ve' capabilities are present) lib/readline/{complete,parens,readline,search,vi_mode}.c - change calls to rl_insert to _rl_insert_char lib/readline/{readline,search}.c - change calls to rl_rubout to _rl_rubout_char to avoid overwrite mode problems lib/readline/vi_mode.c - fix rl_vi_overstrike to just call _rl_overwrite_char, which handles multibyte chars lib/readline/doc/{rluser.texinfo,readline.3}, doc/bash.1 - document new `overwrite-mode' command 1/23 ---- lib/readline/readline.c - return 0 immediately from rl_insert_text if the string to insert is NULL or "" bashline.c - if a numeric argument is given to one of the bash-specific glob pattern completion functions (including TAB), append a `*' to the word before generating matches - in attempt_shell_completion, when doing glob completion, only set the match list to NULL if rl_completion_type == TAB and there is more than one completion. This permits listing completions with double tabs and displaying ambiguous completions - new function, bash_glob_complete_word, appends a `*' to the word to be completed and then globs it. It uses a new filename quoting function (bash_glob_quote_filename) to avoid quoting globbing characters in the filename if there are no matches or multiple matches lib/readline/complete.c - set completion_changed_buffer to 0 in rl_complete_internal if no matches were produced by the completion generator function - new variable, rl_completion_suppress_append, suppresses appending of rl_completion_append_character. Settable by application completion functions, always 0 when application completion functions are called (set to 0 by rl_complete_internal and rl_menu_complete) - broke the code that assigns default values to readline completion variables out of rl_complete_internal and rl_menu_complete into a new function, set_completion_defaults (int what_to_do) lib/readline/readline.h - extern declaration for rl_completion_suppress_append lib/readline/doc/rluser.texinfo, doc/bash.1 - documented behavior of glob-expand-word and glob-list-expansions when supplied a numeric argument - documented glob-complete-word lib/readline/doc/rltech.texinfo - documented rl_completion_suppress_append 1/24 ---- lib/readline/text.c - new file, text and character handling functions from readline.c lib/readline/misc.c - new file, miscellanous bindable functions and their supporting code from readline.c Makefile.in, lib/readline/Makefile.in - changes for text.c, misc.c lib/readline/bind.c - change ISKMAP case of rl_invoking_keyseqs_in_map to output ESC as "\M-" instead of "\e" -- it's closer to the documentation - change _rl_get_keyname to output ESC as \e instead of \C-[ (it's easier to understand) pcomplete.h - new flag, COPT_NOSPACE builtins/complete.def - new `-o nospace' option for complete and compgen (though it doesn't really do anything for compgen, since that doesn't hand anything off to readline) bashline.c - if a programmable completion specifies COPT_NOSPACE, set rl_completion_suppress_append = 1 lib/readline/doc/rluser.texinfo - documented new `-o nospace' option to complete and compgen doc/{bash.1,bashref.texi} - documented $'\cX' escape sequence (forgot to before) 1/28 ---- variables.c - make_new_variable now takes the HASH_TABLE * as its second argument; changed callers - new function, bind_variable_in_table, takes the HASH_TABLE * as its third paramter; bind_variable calls bind_variable_in_table with shell_variables as third argument variables.h - new struct var_context, variable context (per-scope -- global, function local, etc.) variables.[ch],builtins/common.[ch] - moved functions that push and pop a variable context from builtins/common.c to variables.c; move extern function declarations to variables.h - new function, all_local_variables - variable_in_context is now static, used only by all_local_variables variables.[ch],execute_cmd.c - push_context now takes the function name as an argument for future use - push_context takes an indication of whether or not the function is executing in a subshell and saves the positional parameters only if not in a subshell - new functions for managing a stack of variable contexts and scopes: new_var_context, dispose_var_context, push_var_context, pop_var_context, push_scope, pop_scope builtins/declare.def - call all_local_variables instead of map_over (...) in declare_internal - don't call make_local_variable if we're looking at functions ((flags_on & att_function) != 0), since it's wasted - make sure VAR is set to NULL if check for variable_context fails and we didn't just create or fetch a local variable in declare_internal - in non-function branch of declare_internal, only call find_variable if VAR is NULL -- if it's not null, we just created or fetched a local variable and don't need to do it again 1/29 ---- variables.[ch] - the temporary environments (temporary_env, builtin_env, function_env) are now HASH_TABLEs instead of argv-style arrays of strings (this is an intermediate step on the way to the new lcc-inspired symbol table scope structure) - new internal attribute for variables: att_propagate. This means to propagate the value out of the temporary environment up the (for now implicit) chain of variable scopes when the containing temporary environment is deleted variables.c - assign_in_env now adds to the HASH_TABLE temporary_env instead of making environment-style strings in an array of strings - changed the way the temporary environments are merged into the shell variable table to account for the new HASH_TABLE temp environments - changed the way the export environment is created due to the new structure of the temporary environments - new function, bind_variable_internal (name, value, table), binds NAME to have VALUE in TABLE without searching the temporary environments - removed: shell_var_from_env_string, bind_name_in_env_array - variable_in_context now checks the att_local attribute and makes sure the variable is not invisible - local_and_exported now makes sure the variable is not invisible execute_cmd.c - we no longer need to copy the temporary environment to function_env or builtin_env, we can simply use variable assignments {findcmd,subst,variables}.c, builtins/{declare,setattr}.def - since variables from the temporary environments are no longer turned into SHELL_VARs on the fly, don't dispose the SHELL_VAR returned by find_variable or find_variable_internal - need to savestring() the value returned by find_variable if it has the tempvar attribute before calling bind_variable on it, because bind_variable will search and bind into the temporary environments and will free the old value before binding the new. For temporary environments, these two pointers will be the same, and bind_tempenv_variable will end up using freed memory builtins/{declare,setattr}.def - set the att_propagate attribute when exporting or making readonly variables from the temp environment (i.e., `var=value declare -x var' or `var=value export var' sets the propagate attribute on the entry for `var' in the temporary environment HASH_TABLE) lib/readline/isearch.c - ^W when reading isearch string yanks the current word out of the current line into the search string, skipping the portion already matched - ^Y when reading isearch string yanks the rest of the current line into the search string, skipping the portion already matched 1/30 ---- {print_cmd,variables}.c - moved indirection_level_string() from variables.c to print_cmd.c {externs,variables}.h - moved extern declaration of indirection_level_string to externs.h {general,variables}.c - moved assignment() from variables.c to general.c {general,variables}.h - moved extern declaration of assignment() to general.h {externs,input}.h - moved extern declaration of decode_prompt_string to externs.h print_cmd.c - include flags.h, don't include stdc.h variables.c - moved some functions around to group functions better - changed new_shell_variable to explicitly initialize each member of the created struct variable instead of calling bzero() - make_new_variable now just calls new_shell_variable instead of duplicating what it does - removed some code in bind_function that duplicated what new_variable does on the newly-created SHELL_VAR - since there are no local function variables (functions are always made at the global scope), kill_all_local_variables() doesn't need to consider functions 1/31 ---- variables.c - sort the array of special variables - short-circuit the search in stupidly_hack_special_variables if the passed name can't be found in the rest of the array (that is, if name[0] < special_vars[i].name[0]) lib/readline/history.c - unstifle_history() was returning values exactly opposite of the documentation lib/readline/doc/{hsuser.texinfo,history.3} - clarified the unstifle_history() documentation a little 2/4 --- variables.c - in bind_variable, don't call bind_tempenv_variable after a find_tempenv_variable succeeds -- just change the value inline. There's no reason to look it up twice - change makunbound to only call stupidly_hack_special_variables if we're not unsetting a function variables.[ch] - new function, unbind_function, like makunbound but doesn't mess with previous contexts or calling stupidly_hack_special_variables builtins/set.def - change unset_builtin to call either unbind_func or unbind_variable builtins/getopts.def - call unbind_variable(name) instead of makunbound(name, shell_variables) 2/5 --- lib/glob/sm_loop.c - use malloc instead of xmalloc in BRACKMATCH and handle failures error.c - add extern declaration of executing_line_number with prototype, since execute_cmd.h can't be included without including other files lib/readline/parens.c - include lib/malloc/stats.c - include - add extern declaration of malloc_free_blocks() with prototype pathexp.c - added some forward declarations with prototypes for static functions lib/readline/rlprivate.h - removed declarations of rl_untranslate_keyseq, rl_discard_argument, rl_stop_output, rl_alphabetic since they appear in readline.h 2/6 --- {arrayfunc,execute_cmd,pcomplete,shell}.c - change calls to makunbound(name, shell_variables) to unbind_variable (name) 2/7 --- builtins/getopt.c - don't defer incrementing of OPTIND when an invalid option is encountered until the next call to sh_getopt() -- what if OPTIND is reset before that next call? This means that OPTIND is always incremented to the next option to be handled when an option is returned, whether it's valid or not. This is what POSIX-2002 says to do. syntax.h - new #define, CSUBSTOP mksyntax.c - add "-=?+" with value CSUBSTOP to the syntax table. These are the valid expansion operators OP in ${param[:]OPword} subst.c - use table lookup for CSUBSTOP in VALID_PARAM_EXPAND_CHAR - new flags for the string extraction functions: EX_NOALLOC. This indicates that the functions are being used only to skip over strings and the result won't be used, so the substring shouldn't be allocated, copied, and freed - new flag for string_extract: EX_VARNAME. This serves the same purpose as the old `varname' parameter. parameter_brace_expand() changed appropriately - extract_delimited_string and extract_dollar_brace_string now take an additional `flags' argument, which may include EX_NOALLOC - changed callers of extract_delimited_string and extract_dollar_brace_string appropriately - string_extract now understands EX_NOALLOC; callers changed - some smaller code cleanups - converted char_is_quoted(), unclosed_pair(), and skip_to_delim() to understand multibyte characters 2/11 ---- variables.[ch] - moved to a symbol organization inspired by lcc. The basic structure is no longer a HASH_TABLE, but a VAR_CONTEXT, which includes a hash table as one of its members. VAR_CONTEXTs are linked together to do variable scoping. One nice thing about this is that the entire symbol table doesn't need to be searched at function scope exit to remove local variables. Fixes problems with only one instance of builtin_env and function_env, even though it really is a stack - shell_variables is now a VAR_CONTEXT *, with a global_variables variable that points to the bottom of the stack for fast access - function-scope local variables (assignments specified on the command line before a function call) and function-local variables (declared with the `local' builtin) have been unified in the same variable context, replacing function_env - assignment statements preceding the `.' and `eval' builtins are now a separate variable scope VAR_CONTEXT, replacing builtin_env - temporary_env (a HASH_TABLE) is now the only separate environment - changes to export environment creation, variable binding, variable lookup, local variable propagation all changed to work with the new symbol table/scope structure - a SHELL_VAR no longer has a `prev_context' member; it's not needed execute_cmd.c - changes to push_context calls to include any temporary variables in temporary_env; pop_context takes care of propagating any temporary variables if necessary - calls to push_scope if `eval' or `.' is called with a list of preceding variable assignments, and pop_scope called at end of builtin's execution. pop_scope takes care of merging temporary variables into the shell environment when appropriate builtins/{setattr,declare}.def - changes to account for variable assignments preceding `local', `export', `readonly', `declare', etc. to work with the new variable scoping implementation shell.c - since shell_variables is now a VAR_CONTEXT, call delete_all_contexts() when the shell is reinitializing instead of delete_all_variables() builtins/common.c - new function, get_job_by_name(), used by execute_simple_command() for the `auto_resume' stuff and get_job_spec() builtins/common.h - new set of #defined constants for flags argument to get_job_by_name() 2/12 ---- command.h - new redirection operator: r_reading_string for `here strings' parse.y - new token, LESS_LESS_LESS, for new redirection `here string' operator: [N]<<< word - recognize LESS_LESS_LESS and create the appropriate redirection {dispose_cmd,copy_cmd,make_cmd,print_cmd}.c - recognize r_reading_string and do the right thing (dispose_redirects, copy_redirect, print_redirection, and make_redirection, respectively) redir.c - here_document_to_fd now takes the redirection operator as its second argument - new function, write_here_string, expands a here string and writes it to the here document file descriptor - here_document_to_fd calls write_here_string for r_reading_string operator - handle r_reading_string in do_redirection_internal() and stdin_redirection() 2/18 ---- doc/{bash.1,bashref.texi} - documented here strings {configure,Makefile}.in - bumped version number up to bash-2.05b and the release status to alpha1 expr.c - make expr_streval understand that variables with the `invisible' attribute are really unset, and accessing such a variable when `set -u' is set should be an error variables.h - new accessor macros: var_isset(var) and var_isnull(var), test whether var->value is NULL {eval,subst,variables}.c, builtins/{declare,setattr}.def - be more consistent about using value_cell(var) instead of directly referencing var->value - use var_isset and var_isnull where appropriate builtins/help.def - augmented a couple of help strings with pointers to `info' and `man -k' 2/14 ---- variables.h - new macros to use when setting variable values directly instead of through bind_variable and its siblings {arrayfunc,variables}.c - use var_setarray and other lvalue macros instead of assigning to var->value directly builtins/setattr.def - change show_var_attributes to show function definitions separately from function attributes. This allows the output of `declare -f' (with other flags), `export -f', and `readonly -f' to be reused as shell input, instead of the old declare -f[flags] func() { foo } which has syntax errors. When in posix mode, `export -fp' and `readonly -fp' still don't print function definitions 2/16 ---- parse.y - comment out calls to discard_parser_constructs; no need to call empty functions 2/18 ---- lib/sh/memset.c - replacement function for memset(3) lib/sh/Makefile.in, Makefile.in - additions for memset.c configure.in,config.h.in - check for memset, define HAVE_MEMSET if found, add memset.o to LIBOBJS if not lib/malloc/malloc.c - removed zmemset(), replaced with calls to memset(3) {subst,execute_cmd,lib/sh/netopen}.c - replaced calls to bzero with calls to memset subst.c - word_split() now takes a second argument: the value of $IFS, so it doesn't have to look up IFS every time - word_list_split() now calls getifs() and passes the result to each call to word_split() as its second arg - do a quick scan for CTLNUL in remove_quoted_nulls before allocating new string, copying old string to it, copying over original string and freeing new string eval.c - don't bother calling dispose_used_env_vars if temporary_env is NULL execute_cmd.c - fix fix_assignment_words to only look up the builtin corresponding to the first word if one of the words in the list is marked as W_ASSIGNMENT hashlib.c - renamed hash_string to hash_bucket, which better reflects what it does - extracted the portion of hash_bucket that computes the hash out into a new hash_string() - made new body of hash_bucket into a macro HASH_BUCKET; function just calls the macro - calls to hash_bucket in this file now call HASH_BUCKET macro - in add_hash_item, just add a new item at the front of the appropriate bucket list instead of at the end hashcmd.h - reduced FILENAME_HASH_BUCKETS to 53 from 107 2/19 ---- hashlib.[ch] - find_hash_item, remove_hash_item, add_hash_item all take a new third `flags' argument - add_hash_item doesn't call find_hash_item if HASH_NOSRCH passed in flags arg - find_hash_item will create a new hash table entry if HASH_CREATE is passed in flags arg - new function, hash_walk, takes a pointer to a function and a table and calls the function for each item in the table. If the function returns < 0, the walk is terminated - fixed flush_hash_table to set table->nentries to 0 after freeing all entries - BUCKET_CONTENTS now has a new `khash' member, what key hashes to; set by HASH_BUCKET macro (which calls hash_string), assigned in find_hash_item (HASH_CREATE) and add_hash_item - find_hash_item and remove_hash_item check `khash' against the hash of the string argument before calling strcmp {alias,hashlib,hashcmd,pcomplib,variables}.c - changed all calls to {find,remove,add}_hash_item builtins/hash.def - return immediately from print_hashed_commands if there are no entries in the hash table (this eliminates need for `any_printed' variable) - change print_hashed_commands to use hash_walk alias.c - short-circuit all_aliases and map_over_aliases if HASH_ENTRIES(aliases) == 0 - simplify map_over_aliases by just allocating enough room in the returned list for all entries in the aliases hash table, instead of doing the check and xrealloc - add_alias now calls add_hash_item with HASH_NOSRCH argument pcomplete.h - sh_csprint_func_t is no more; use hash_wfunc instead pcomplib.c - short-circuit print_all_compspecs if HASH_ENTRIES(prog_completes) is 0 - print_all_compspecs now takes a `hash_wfunc *' argument - print_all_compspecs now just calls hash_walk builtins/complete.def - new function, print_compitem, takes a BUCKET_CONTENTS *, extracts the right info, and calls print_one_completion variables.c - short-circuit map_over_funcs if HASH_ENTRIES(shell_functions) == 0 - short-circuit flatten if the passed table has no entries - bind_variable_internal takes a new fourth argument: `hflags', to pass to hash table functions - make_new_variable now passes HASH_NOSRCH flag to add_hash_item - set_if_not now calls bind_variable_internal and passes HASH_NOSRCH as flags argument - bind_function now calls add_hash_item with HASH_NOSRCH argument - fixed make_local_variable: old_var == 0 && was_tmpvar can never be true - if we didn't find an old variable in make_local_variable, call bind_variable_internal with HASH_NOSRCH argument - fix push_temp_var to reset variable context to 0 if binding into global_variables->table parse.y - fix to parse_compound_assignment to avoid core dumps on empty compound array assignments subst.c - getifs() is now global so read_builtin can call it subst.h - extern declaration for getifs() 2/20 ---- hashlib.c - changed hash_string to use a better hash function - changed HASH_BUCKET to use masking rather than modulus to hash a string to a bucket -- HASH TABLES MUST NOW BE SIZED BY POWERS OF TWO hashlib.h - DEFAULT_HASH_BUCKETS is now 64 hashcmd.h - FILENAME_HASH_BUCKETS is now 64 pcomplib.c - COMPLETE_HASH_BUCKETS is now 32 variables.c - TEMPENV_HASH_BUCKETS is now 4 alias.c - new define, ALIAS_HASH_BUCKETS, set to 16, used to size alias table hashlib.c - removed initialize_hash_table; folded code into make_hash_table - fixed copy_bucket_array to copy the `khash' member of an item - renamed functions to be more systematic and easier for me: make_hash_table -> hash_create hash_table_nentries -> hash_size copy_hash_table -> hash_copy find_hash_item -> hash_search remove_hash_item -> hash_remove add_hash_item -> hash_insert flush_hash_table -> hash_flush dispose_hash_table -> hash_dispose print_table_stats -> hash_pstats get_hash_bucket -> hash_items - changed hash_search to short-circuit if table->nentries == 0 and HASH_CREATE has not been passed in the flags argument {alias,variables,hashcmd,pcomplib}.c - renamed calls to all renamed functions from hashlib.c builtins/kill.def - don't drop a leading `-' in a pid argument - call kill_pid with an explicit third argument of 1 if the pid argument to kill is < -1, rather than rely on the behavior of kill(2) 2/21 ---- subst.c - quoted_strchr is no longer declared `inline' - skip_double_quoted is no longer declared `inline' - string_extract_double_quoted is no longer declared `inline' lib/readline/input.c - rl_gather_tyi is now an `int' valued function; returns the number of characters read (0 or 1) or -1 on error - if rl_gather_tyi() returns -1 to rl_read_key(), set rl_done to 1 and return a newline; something is wrong with the input fd 2/25 ---- variables.[ch] - IFS is now a special variable - new special var function, sv_ifs(), called when IFS is set or unset - call setifs() when IFS is first set in initialize_shell_variables - call setifs() from make_local_variable and assign_in_env if appropriate - if assign_in_env() is called with a var assignment like `VAR=', make the value in the new SHELL_VAR created be "" like do_assignment_internal does, since certain parts of the shell use a NULL value as evidence that the variable is unset (though attributes may have been assigned) - if push_temp_var pushes something up to the global_variables table, make sure that the context is set to 0 - new function dispose_temporary_env, called by both dispose_used_env_vars and merge_temporary_env with different `free func' function pointers; calls sv_ifs after disposing the temporary environment - push_exported_var now calls bind_variable_internal instead of bind_variable - pop_scope and pop_context now call sv_ifs subst.[ch] - new global variables used to keep track of IFS state, to avoid having to call find_variable("IFS") all the time: ifs_var the SHELL_VAR for IFS ifs_value ifs_var ? value_cell (ifs_var) : " \t\n" ifs_cmap bitmap of characters in ifs_value ifs_firstc first character in ifs_value - new function setifs(), sets the aforementioned ifs variables each time IFS is set or unset, and at nested scope exit - instead of calling getifs() from inside subst.c, use ifs_value - getifs() now just returns ifs_value - use ifs_firstc in string_list_dollar_star() - only call member() in issep() if separators is more than one char - don't cache a bitmap every time expand_word_internal() is called; use ifs_cmap instead - new macro, isifs(c), checks whether C is in ifs_cmap builtins/read.def - use issep() and isifs() macros instead of looking at $IFS directly syntax.h - make sure macros that access sh_syntaxtab cast the argument to `unsigned char' before array access - new macros: issyntype(c, type) and notsyntype(c, type), check sh_syntaxtab[c] for a particular flag value `type' 2/26 ---- hashlib.h - the `data' member of a `BUCKET_CONTENTS' is now a PTR_T {hashlib,alias,variables,hashcmd,pcomplib}.c - removed some casts when assigning to and using `data' member of a `BUCKET_CONTENTS' subst.c - in split_at_delims, call make_word_list instead of allocating and initializing a WORD_LIST * directly make_cmd.[ch] - add_string_to_list is now just a macro that calls make_word_list - make_simple_command now calls make_word_list instead of allocating a WORD_LIST * directly 2/27 ---- copy_cmd.c - copy_word now calls make_bare_word to allocate the copy - copy_word_list now calls make_word_list to allocate the copy shell.h - include `ocache.h' for simple object caching - call cmd_init() to initialize the WORD_DESC and WORD_LIST object caches {make,dispose}_cmd.c - allocate WORD_DESC * and WORD_LIST * vars from their respective ocaches, and return them to the cache when disposing jobs.c - renamed old `waiting_for_job' variable to `queue_sigchld', which better reflects its intent: sigchld_handler does not call waitchld if `queue_sigchld' is non-zero, it simply increments the count of waiting children - cleanup_dead_jobs now just sets and clears queue_sigchld instead of blocking and unblocking SIGCHLD; it calls waitchld at the end if `sigchld' is non-zero, but that's not really necessary - in setjstatus, only call xrealloc if `statsize' is less than the number of processes passed -- no reason to do it if they're the same 2/28 ---- sig.[ch] - reinitialize_signals is no more; initialize_signals takes an argument saying whether or not we are reinitializing builtins/exec.def - reinitialize_signals() -> initialize_signals(1) test.c - fix filecomp() to work right when one file has a non-positive timestamp and the other file does not exist doc/{bash.1,bashref.texi} - document what happens for test's -nt and -ot operators when one file operand exists and the other does not jobs.c - if we haven't messed with SIGTTOU, just manipulate queue_sigchld in notify_of_job_status instead of calling sigprocmask() - list_one_job now calls pretty_print_job directly instead of going through print_job - pretty_print_job now must be called with SIGCHLD blocked or held instead of blocking SIGCHLD itself - changed start_job so that it doesn't call UNBLOCK_CHILD and then immediately call BLOCK_CHILD again (explicitly or via last_pid()), call find_last_pid instead of last_pid and then UNBLOCK_CHILD - changed wait_for_job the same way - find_last_pid now takes a second argument: block; uses BLOCK_CHILD if `block' is 1, not otherwise. Changed existing calls: find_last_pid(j) -> find_last_pid(j, 0) last_pid(j) -> find_last_pid(j, 1) `last_pid()' is now gone - rewrote wait_for_background_pids(); it was a little strange copy_cmd.c - copy_if_command: don't copy null false_case commands - copy_simple_command: don't copy a null redirection list subst.c - in get_word_from_string and list_string, just check for " \t\n" directly rather than calling strcmp - in get_word_from_string and strip_trailing_ifs_whitespace, use isifs() instead of issep(), since they're never called with separators != $IFS - change issep() to call isifs if separators is longer than one character, since it's never called with anything but "", " ", or $IFS 3/1 --- sig.h - enclose the BLOCK_SIGNAL macro in a do {...} while (0) loop, at it should have been all along lib/readline/doc/rltech.texinfo - document that readline defaults to stdin/stdout if rl_instream/ rl_outstream are NULL lib/readline/terminal.c - if an application is using a custom redisplay function, rl_resize_terminal just calls rl_forced_update_display to tell (*rl_redisplay_func) to update the display, otherwise call _rl_redisplay_after_sigwinch lib/readline/readline.c - change readline_internal_setup() so the change to vi insertion mode happens even if readline_echoing_p is 0 - don't print the prompt to rl_outstream in readline_internal_setup if we're not echoing and the caller has defined a custom redisplay function -- let the redisplay function deal with it configure.in - new option: --enable-mem-scramble, controls memory scrambling on free() (on by default; only affects use of bash malloc) config.h.in - new option MEMSCRAMBLE, controlled by --enable-mem-scramble 3/5 --- parse.y - added ksh-like behavior of [...] to read_token_word: if a `[' is seen in an assignment context and the previous characters in the token form a valid identifier, parse the [...] with parse_matched_pair to allow spaces (and newlines) in the subscript bashline.c - new function bash_servicename_completion_function, for completing service names from /etc/services bashline.h - new extern declaration for bash_servicename_completion_function builtins/complete.def - allow new `-s/-A service' option to complete and compgen builtins pcomplete.h - new CA_SERVICE define, new ITEMLIST variable it_services pcomplete.c - add callback to bash_servicename_completion_function to generate list of matching service names for completion doc/bash.1,lib/readline/doc/rluser.texinfo - documented new `-s/-A service' option to complete and compgen 3/6 --- builtins/read.def - change hard-coded `0' to new variable `fd' (initially 0) in preparation for adding `-u fd' option bashline.c - bash_directory_completion_hook calls expand_prompt_string instead of expand_string (it does the right thing). This keeps expansion errors from causing a longjmp, which shouldn't happen because of completion - command_subst_completion_function was augmented very slightly to do filename completion on a non-command-word in a command substitution - command_subst_completion_function now skips over the lcd that rl_completion_matches puts in matches[0] if there is more than one possible completion 3/7 --- builtins/read.def - only add the unwind_protect to free `rlbuf' if `edit' is non-zero, since we won't be using readline otherwise lib/sh/zread.c - renamed zread1 -> zreadintr redir.c - small change to redirection_error() to make a slightly better guess about the invalid file descriptor if the redirection op is r_duplicating_input or r_duplicating_output include/stdc.h - new macro, SH_VA_START, to encapsulate the difference between stdarg va_start and varargs va_start {error,pcomplete,print_cmd}.c,builtins/common.c,lib/sh/snprintf.c - use SH_VA_START 3/8 --- builtins/read.def - support for the ksh-like `-u fd' option general.c - new function sh_validfd(fd), returns 1 if fd is a valid open file descriptor general.h - extern decl for sh_validfd bashline.c - don't call posix_readline_initialize() from initialize_readline(); sv_strict_posix() should already have taken care of it 3/11 ---- {error,pcomplete,print_cmd}.c, builtins/common.c - removed non-varargs versions of functions builtins/printf.def - if the string argument to %q has non-printing characters, call ansic_quote to quote it rather than sh_backslash_quote variables.h - new attribute: att_trace (and corresponding trace_p() macro). Functions with this attribute will inherit the DEBUG trap. Currently ignored for variables builtins/declare.def - new `-t' option to declare/typeset toggle the `att_trace' attribute builtins/setattr.def - check for att_trace and output `-t' flag in show_var_attributes execute_cmd.c - if a function is being traced (it has the `-t' attribute set), don't turn off the DEBUG trap when it executes doc/{bash.1,bashref.texi} - document the new `-t' option to declare/typeset 3/12 ---- execute_cmd.c - don't execute the debug trap in the `cm_simple:' case of execute_command_internal; run it in execute_simple_command so we get the line number information right when executing in a shell function - run a DEBUG trap before executing ((...)) arithmetic commands, like ksh93 - run a DEBUG trap before executing [[...]] conditional commands, like ksh93 eval.c - add a static forward declaration for alrm_catcher() general.c - add static forward declarations for bash_special_tilde_expansions, unquoted_tilde_word, initialize_group_array variables.h - add extern declarations for sh_get_env_value, map_over_funcs, local_exported_variables variables.c - add static forward declarations for dispose_temporary_env, make_func_export_array bashhist.c - add static forward declaration for check_history_control configure.in - add a call to AC_CHECK_DECLS for strcpy config.h.in - add placeholder for HAVE_DECL_STRCPY define, set by configure general.h - don't declare strcpy if HAVE_DECL_STRCPY is defined with a non-zero value sig.h - add prototype to typedef of SigHandler lib/readline/histlib.h - removed extern declaration of strcpy() - include string.h/strings.h directly in histlib.h instead of source files lib/readline/{histexpand,histfile,history,histsearch}.c - don't include string.h/strings.h now that histlib.h includes it lib/tilde/tilde.c - removed extern declaration of strcpy(), rely on string.h/strings.h command.h - four new redirection types: r_move_input, r_move_output, r_move_input_word, r_move_output_word, for [N]<&word- and [N]>&word- from ksh93 print_cmd.c - changes to print r_move_input[_word] and r_move_output[_word] copy_cmd.c - changes to copy r_move_input[_word] and r_move_output[_word] dispose_cmd.c - changes to dispose r_move_input_word and r_move_output_word make_cmd.c - changes to make r_move_input[_word] and r_move_output[_word] from r_duplicating_{input,output}_word, which is how the new redirs are passed by the parser redir.c - changes to make r_move_input[_word] and r_move_output[_word] do the right thing when executed builtins/read.def - print an error message and return failure immediately if zread/zreadc return < 0 doc/{bash.1,bashref.texi} - documented new [n]<&word- and [n]>&word- redirections 3/13 ---- lib/readline/isearch.c - enabled code to allow chars bound to rl_rubout to delete characters from the incremental search string shell.c - add `-l' invocation option to parse_shell_options; equivalent to `--login' - fixed set_login_shell to check first char of base pathname of argv0 for `-', like other shells - move the check for make_login_shell after the call to parse_shell_options because the `-l' option might set it doc/{bash.1,bashref.texi} - documented new `-l' invocation option array.c - new function, array_shift, shifts an array left by a specified number of elements - array_walk is now compiled in by default - array_to_assignment_string now takes a second argument: int quoted. If non-zero, the result is single-quoted before being returned - quoted_array_assignment_string has been removed array.[ch] - renamed most of the array functions so that all have an array_ prefix and are more systematically named - array_slice now preserves the indicies from the original array - change array_to_assign to use a static buffer for expanding the array indices, instead of malloc/free {arrayfunc,subst,variables}.c, builtins/read.def - changed calls to various array functions to use new names lib/sh/stringvec.c, externs.h - renamed all of the functions to have a strvec_ prefix and to have a more sensible name scheme - strvec_search's arguments are now supplied in reverse order, so the char **array is first, like the other functions - new function, strvec_resize, xrealloc for strvecs {alias,array,bracecomp,braces,bashline,execute_cmd,findcmd,general,pathexp, pcomplete,variables}.c lib/sh/stringlist.c builtins/{bind,complete,exec,getopts,pushd,set}.def - change calls to all functions from lib/sh/stringvec.c - use strvec_resize where appropriate externs.h - only declare dup2() if HAVE_DUP2 is undefined or DUP2_BROKEN is defined lib/readline/{macro,readline,util}.c, lib/readline/rlprivate.h - _rl_defining_kbd_macro is gone, use RL_ISSTATE(RL_STATE_MACRODEF) lib/readline/readline.h - new struct readline_state, encapsulates most of readline's internal state in case you need reentrancy or nested calls to readline() - extern declarations for rl_save_state, rl_restore_state lib/readline/readline.c - add (undocumented) int rl_save_state (struct readline_state *), int rl_restore_state (struct readline_state *) 3/14 ---- array.[ch] - new function, array_rshift, shifts an array right by a specified number of elements, optionally inserting a new element 0 examples/bashdb/bashdb - new single-file version of bash debugger, originally modified from version in bash-2.04 by Gary Vaughan (the old debugger still appears in examples/obashdb). This version has a more gdb-like command set examples/bashdb/bashdb.el - new emacs bashdb debugger mode from Masatake YAMATO execute_cmd.c - don't make $LINENO relative to function start unless the shell is currently interactive -- this is what ksh93 does and what I believe to be the intent of POSIX.2 (this required changing some of the test checks because the output has changed) - run the debug trap for each command in an arithmetic for expression, like ksh93 does lib/readline/vi_mode.c - redid rl_vi_subst (binding func for `s' and `S') in terms of rl_vi_change_to: `S' == `cc' and `s' == `c '. This makes undo work right 3/18 ---- hashlib.c - fixed hash_walk to return if the item function returns < 0, instead of breaking out of the current hash chain array.c - fixed array_walk to return if the item function returns < 0, like hash_walk lib/sh/stringlist.c, externs.h - new function: strlist_walk, takes a stringlist and a pointer to an item func. Like other _walk funcs, if item func returns < 0 the walk is cancelled - new function: strlist_flush, frees items in the contained list with strvec_flush - renamed functions to have a strlist_ prefix and be more systematic pcomplib.c,pcomplete.h - removed redundant `progcomp_initialized' variable - renamed functions to have `progcomp_' or `compspec_' prefixes like the hash library {bashline,pcomplete}.c,builtins/complete.def - fixed calls to stringlist functions to use new names - fixed calls to functions in pcomplib.c to use new names pcomplete.c - made the debugging code #ifdef DEBUG -- it should be mature enough builtins/hash.def,parse.y - use REVERSE_LIST(x, t) instead of (t)reverse_list(x) list.c,{externs,general}.h - renamed the list functions to have a list_ prefix, changed callers externs.h,{execute_cmd,stringlib,subst}.c,builtins/common.c,lib/sh/stringvec.c - word_list_to_argv -> strvec_from_word_list - argv_to_word_list -> strvec_to_word_list - moved functions to lib/sh/stringvec.c lib/sh/stringvec.c - changed name of second argument to strvec_from_word_list from `copy' to `alloc' so the use of `copy' between strvec_from_word_list and strvec_to_word_list isn't as confusing - changed name and sense of second argument to strvec_to_word_list from `copy' to `alloc' for the same reason -- now both functions agree on semantics of second argument lib/sh/stringlist.c - ditto for strlist_from_word_list and strlist_to_word_list subst.c - changed callers of strvec_to_word_list 3/19 ---- builtins/hash.def - added `-l' option to list table or individual targets in reusable format - added `-d' option to remove one or more names from the table of hashed commands (provides `unhash' or `unalias -t' functionality) doc/{bash.1,bashref.texi} - documented new `-l' and `-d' options to `hash' hashcmd.[ch] - renamed functions to have a `phash_' prefix and follow new naming convention - phash_remove now returns an int: 1 if command not in hash table, 0 if filename removed OK {findcmd,variables}.c, builtins/{hash,type}.def - changed callers to use new names from hashcmd.c builtins/common.[ch] - new function, sh_notfound(s), prints standard `not found' message - new function, sh_invalidid(s), prints standard `invalid identifier' message - new function, sh_restricted(s), prints standard `restricted' message for restricted shells - new function, sh_invalidnum(s), prints standard `invalid number' message - renamed bad_option to sh_invalidopt, changed to print `invalid option' instead of `unknown option' - new function, sh_invalidoptname, prints standard `invalid option name' for long options - new function, sh_badjob (s), prints standard `no such job' message - new function, sh_invalidsig (s), prints standard `invalid signal specification' message - new function, sh_nojobs (s), prints standard `no job control' message - new function, sh_needarg (s), prints standard `option requires an argument' message - new function, sh_neednumarg (s), prints standard `numeric argument required' message - new function, sh_badpid(s), prints standard `not a pid...' message - new function, sh_erange (s, desc) prints standard `out of range' message, optionally using `desc' to say what the argument is builtins/{alias,command,declare,exec,hash,type}.def - call sh_notfound() instead of calling builtin_error directly builtins/{declare,getopts,read,set,setattr}.def - call sh_invalidid() instead of calling builtin_error directly builtins/{cd,command,enable,exec,hash,source}.def - call sh_restricted() instead of calling builtin_error directly builtins/{printf,read,ulimit}.def, builtins/common.c - call sh_invalidnum instead of calling builtin_error directly builtins/{complete,declare,pushd,set}.def, builtins/bashgetopt.c - call sh_invalidopt instead of bad_option or builtin_error directly builtins/{complete,set,shopt}.def - call sh_invalidoptname instead of builtin_error directly builtins/{fg_bg,jobs,kill,wait}.def - call sh_badjob instead of calling builtin_error directly builtins/common.c, builtins/{kill,signal}.def - call sh_invalidsig instead of calling builtin_error directly builtins/{fg_bg,suspend,wait}.def - call sh_nojobs instead of calling builtin_error directly builtins/{common,bashgetopt}.c, builtins/{hash,kill}.def - call sh_neednumarg and sh_needarg where required builtins/{kill,wait}.def - call sh_badpid where required builtins/{break,fc,history,pushd,shift,ulimit,umask}.def - call sh_erange where appropriate builtins/printf.def - new static function, printf_erange, prints standard out-of-range warning message builtins/set.def - changed so that calls to sh_invalidopt always include the leading `+' or `-' builtins/shopt.def - changed SHOPT_ERROR macro to shopt_error function builtins/bind.def - regularized error messages to `bind: object: error string' like other error messages builtins.h - the `short_doc' member of a `struct builtin' is now of type `const char *' - the strings in `long_doc' array of a struct builtin are now const builtins/mkbuiltins.c - changes for new `const' members of struct builtin 3/20 ---- lib/readline/histfile.c - use pointers instead of indexing into buffer when reading the contents of the history file in read_history_range and history_truncate_file 3/21 ---- lib/readline/histfile.c - new file, with code to mmap the history file for reading and writing (depends on HAVE_MMAP, currently nothing checks for that) 3/25 ---- error.[ch] - new function, err_badarraysub(s), calls report_error with standard `bad array subscript' message - new function, err_unboundvar(s), calls report_error with standard `unbound variable' message - new function, err_readonly(s), calls report_error with standard `readonly variable' message {arrayfunc,subst}.c - call err_badarraysub where appropriate {expr,subst}.c - call err_unboundvar where appropriate {arrayfunc,variables}.c - call err_readonly where appropriate shell.c - changed text of bad option error messages to be the same as that printed for builtin errors builtins/common.c - changed sh_invalidopt to print the invalid option before the rest of the error message (required some tests to be modified) - new function, sh_readonly, calls builtin_error with standard `readonly variable' message variables.c,builtins/declare.def - call sh_readonly where appropriate lib/sh/stringvec.c - added strvec_remove (sv, s), removes S from SV and shuffles rest of elements down 1 lib/sh/stringlist.c - added strlist_remove(sl, s), just calls strvec_remove on the component list externs.h - new extern declarations for strvec_remove and strlist_remove - fixed extern declaration for strvec_search; the arguments were reversed (unimportant, it's not compiled into the shell) subst.c - change param_expand to call quote_escapes on values retrieved when expanding the positional parameters - change parameter_brace_expand_word to quote escapes on values retrieved when expanding the positional parameters - fix parameter_brace_substring to quote escape characters on unquoted substrings extracted from variable values (needed to separate case VT_VARIABLE from VT_ARRAYMEMBER for this, since, because get_var_and_type calls array_value for VT_ARRAYMEMBER, we need to skip over quoted characters in an already-appropriately-quoted string to find the substring we want) - fix parameter_brace_substring to quote escape characters in the value returned by pos_params when expanding subsets of the positional parameters and not within double quotes (in which case pos_params() quotes the string for us) - fix parameter_brace_substring to quote escape characters in the value returned by array_subrange when expanding subsets of an array and not within double quotes (in which case array_subrange() quotes the string for us) - new function, quoted_strlen(s), does strlen(s) while skipping over characters quoted with CTLESC (#ifdef INCLUDE_UNUSED, since it's not used yet) - changed pos_params() so it always returns a list whose members are quoted strings if (quoted&(Q_HERE_DOCUMENT|Q_DOUBLE_QUOTES)) != 0 arrayfunc.c - fix array_value to consistently call quote_escapes, even when a non-array variable is being subscripted with element 0, in which case we return the variable value lib/sh/strtrans.c - make the for_echo parameter to ansicstr a `flags' parameter that has its old `for echo' meaning if flags&1 is non-zero (which is consistent with the old code) - Added code to the `flags' parameter to ansicstr so that if flags&2 is non-zero, CTLESC and CTLNUL are escaped with CTLESC in the expanded string - change ansiexpand() to call ansicstr with a `flags' parameter of 2 3/26 ---- lib/readline/histfile.c - when reading and writing the history file, use malloc instead of xmalloc and handle failures gracefully, so the application doesn't abort if the history file or history list is too big 3/27 ---- arrayfunc.c - changed array_value_internal to take an additional `int *' parameter, in which is returned the type of array indexing performed (array[@] vs. array or array[index]) - changed array_value and get_array_value to take a corresponding extra parameter and pass it to array_value_internal - changed array_value_internal to no longer return newly-allocated memory or quote CTLESC and CTLNUL in the returned string if `simple' array indexing (subscript not `@' or `*') is being performed. This makes it more like a variable lookup arrayfunc.h - changed prototypes for array_value and get_array_value expr.c - added new parameter to call to get_array_value in expr_streval - don't need to free memory returned by get_array_value any more subst.c - quote_escapes now works with multibyte characters - dequote_string now works with multibyte characters - dequote_escapes is now needed, so it's compiled in, and it now works with multibyte characters - remove_quoted_escapes now just calls dequote_escapes and copies the result over the argument string - remove_quoted_nulls now returns its char * argument, parallels remove_quoted_escapes - parameter_brace_expand_word now passes the new argument to array_value and quotes CTLESC and CTLNUL in the result if it's a `simple' array expansion by calling quote_escapes - get_var_and_type now returns VT_ARRAYMEMBER for references like ${array} where `array' is an array variable (just like ${array[0]}). Documented (in comment) that a VT_VARIABLE return value means that quote_escapes has been called at some point - changed callers of get_var_and_type to no longer free value if VT_ARRAYMEMBER is returned as type - changed parameter_brace_substring and parameter_brace_patsub to call dequote_escapes on the value from get_var_and_type if the type is VT_VARIABLE, since the substring and pattern substitution code doesn't understand CTLESC quoting - parameter_brace_substring no longer needs to call quoted_substring for the VT_ARRAYMEMBER case - changed parameter_brace_patsub to call quote_escapes on the result of pat_subst for the VT_VARIABLE and VT_ARRAYMEMBER cases, and to quote the returned string in the VT_ARRAYVAR and VT_POSPARAMS cases if the `MATCH_QUOTED' flag isn't set (if it is, the pattern substitution functions perform any necessary quoting) - quoted_substring is no longer used; it's now #ifdef INCLUDE_UNUSED lib/malloc/mstats.h - new member in _malstats: u_bits32_t bytesreq, the total number of bytes requested by the caller via calls to malloc() and realloc() lib/malloc/stats.c - print bytesreq member in _print_malloc_stats - don't print statistics for buckets for which nmal == 0 (no mallocs) lib/malloc/malloc.c - modified internal_malloc, internal_realloc to keep running total of number of bytes requested by calling application shell.c - sh_exit is now compiled in; exit_shell calls sh_exit error.c - changed fatal_error, report_error, parser_error to call sh_exit 3/28 ---- subst.[ch] - changed Q_NOQUOTE to Q_PATQUOTE; it makes the intent more clear subst.c - moved code from parameter_brace_expand into a new function that dispatches for pattern substitution: parameter_brace_remove_pattern - changed structure of parameter_brace_remove_pattern to be like parameter_brace_patsub and its ilk: call get_var_and_type to isolate the variable name, move the pattern isolation code out of the various *_remove_pattern functions into parameter_brace_remove_pattern and pass the results to the various functions, use a switch on the return value from get_var_and_type to decide which function to call, regularized the arguments to the separate pattern removal functions - parameter_brace_remove_pattern now properly quotes escape chars in the returned value - changed get_var_and_type to call dequote_escapes on the `value' parameter for case VT_VARIABLE and return the result in *valp, so the calling functions don't have to do it themselves; changed callers appropriately - fixed getpattern() where it broke posix compliance: if you enclose a pattern removal spec in double quotes, the outer double quotes have no effect on the pattern (POSIX.1-200x 2.6.2). This uncovered a bug in the test suite (!) pathexp.c - fixed a problem with quote_string_for_globbing where it would change consecutive CTLESC chars all to \ instead of changing every other quoted char 3/31 ---- lib/malloc/{malloc,stats}.c - moved declaration of _mstats to malloc.c so stats.o doesn't get linked into the shell if the stats functions aren't called 4/2 --- lib/glob/smatch.c - introduce `XCHAR' define, which is the type of arguments passed to strcoll/strcmp/strlen and their wide-character equivalents, added appropriate casts - static arrays in single-byte version of rangecmp() are `char', not `unsigned char', so compilers don't complain about calls to strcoll lib/glob/sm_loop.c - casts for `XCHAR' and `XCHAR *' arguments to libc functions - use prototype declaration for BRACKMATCH if `PROTOTYPES' is defined to avoid problems with type promotion (unsigned char -> int) lib/glob/collsyms.h - `name' member of struct _COLLSYM is now of type `XCHAR *', since some compilers don't like `unsigned char *' initializers from constant strings [bash-2.05b-alpha1 released] 4/3 --- builtins/{evalstring.c,common.h} - new flag for parse_and_execute, SEVAL_NOFREE, means to not free the argument string when finished lib/readline/text.c - fixed a trivial typo in _rl_insert_char when reading multibyte char sequences - replace calls to ding() with rl_ding() include/chartypes.h - remove SIGN_EXTEND_CHAR and TOASCII macros; they're unused make_cmd.c - include dispose_cmd.h for extern function declarations lib/glob/glob.c - include `shmbutil.h' and `xmalloc.h' for extern function declarations lib/glob/smatch.c - include `xmalloc.h' for extern function declarations shell.c - fix maybe_make_restricted to use its argument instead of global `shell_name' version.c - update copyright message to include this year lib/readline/display.c - fixes from Jiro SEKIBA to fix autowrapping when using multibyte characters lib/glob/sm_loop.c - fixed a problem in BRACKMATCH where not enough memory was allocated to hold a multibyte character when parsing POSIX.2 char class names support/config.{guess,sub} - updated via patch from Paul Eggert with latest GNU additions variables.c - var_lookup should use its `vcontext' argument instead of unconditionally using `shell_variables' 4/4 --- builtins/bind.def,doc/{bash.1,bashref.texi} - changed the usage summary and help text to make it clear that any readline command that may appear in ~/.inputrc may be supplied as one of the non-option arguments to `bind' builtins/mkbuiltins.c - added support for `-H' option, which means to write long documentation for each builtin to a separate file in the `helpfiles' directory builtins/Makefile.in - new target `helpdoc', just creates long doc files in helpfiles directory lib/sh/zcatfd.c - new file, with zcatfd(int fd, int ofd, char *fn); dumps data from FD to OFD Makefile.in,lib/sh/Makefile.in - added zcatfd.c, zcatfd.o member of libsh.a builtins/evalstring.c - changed cat_file to call zcatfd(fd, 1, fn) builtins/{shopt,colon}.def - removed the $DOCNAME directive for `shopt', `true', and `false'; just use the names - changed $DOCNAME for `:' to just be `colon' instead of `colon_builtin' builtins/reserved.def - added help entries for ((, [[, `for ((' builtins/let.def - add id++, id--, ++id, --id, ** to help text 4/8 --- builtins/bashgetopt.[ch] - changed to allow options beginning with `+', enabled by a leading `+' in the option string - new variable, list_opttype, set to `-' or `+' builtins/{common.c,{builtin,eval,exit,fg_bg,let,printf,pushd,return,source,wait}.def - changes to allow a `--' option for every builtin that accepts operands but not options, as per posix.1-2001 builtins/{declare,setattr}.def - use internal_getopt for parsing options, now that it supports `+' builtins/set.def - use internal_getopt for initial option parse, now that it supports a leading `+' {configure,Makefile}.in, builtins/{Makefile.in,help.def,mkbuiltins.c} - support for a new configure option, ``--enable-separate-helpfiles'', moves the `long' help text to separate help files, installed by default into ${datadir}/bash, one file per builtin. Off by default -- it saves 47K, but it's only 47K, and it's in the text segment flags.c - build internal_getopt() option string argument from flags array at runtime in shell.c shell.c - new variable to control writing malloc stats at exit: malloc_trace_at_exit, 0 by default lib/malloc/malloc.c - heavily updated: o partial page allocated on first call to malloc to make subsequent sbrks page-aligned no longer wasted o begin and end range guards are now the same value: the chunk requested o coalescing code was changed to attempt to coalesce first two adjacent blocks on the free list; enabled by default o blocks of size 32 are now candidates for larger block splitting, since 32 is the most popular size o blocks of size 32 are now candidates for smaller block coalescing o the IN_BUCKET check was changed to just make sure that the size isn't too big for the bucket, since the `busy block' checking code may increase the bucket by one or more, meaning that the old check would fail and cause a panic when a chunk allocated in such a way was freed o bin sizes are now precomputed and looked up in an array rather than being computed at runtime o moved the _mstats declaration here to avoid the stats code being linked in even when no stats functions were called (only matters if MALLOC_DEBUG is defined) o malloc now keeps track of the address of the top of the heap and will return large chunks to the system with calls to sbrk with a negative argument when freeing the top chunk. Two thresholds: LESSCORE_FRC means to unconditionally return memory to the system; LESSCORE_MIN means to return memory if there's at least one block already on the free list lib/malloc/mstats.h - stats struct now keeps track of number of block coalesces by bin, and the number of times memory was returned to the system by bin lib/malloc/stats.c - trace_malloc_stats now takes a second argument: the name of the file to write to. The first `%p' in the template file name is replaced by the pid 4/9 --- lib/malloc/imalloc.h - added some macros derived from dlmalloc and glibc malloc to inline memcpy and memset if the requested size is <= 32 bytes lib/malloc/malloc.c - use MALLOC_MEMSET instead of memset in internal_{malloc,free} include/ocache.h - use OC_MEMSET (variant of MALLOC_MEMSET) in ocache_free configure.in, config.h.in - check for getservent(), define HAVE_GETSERVENT if found bashline.c - punt immediately from bash_servicename_completion_function if HAVE_GETSERVENT is not defined (cygwin seems to not define it) - include "input.h" for extern save_token_state() and restore_token_state() declarations - change bash_execute_unix_command to call parse_and_execute with SEVAL_NOHIST flag so the command doesn't get saved on the history list - change bash_execute_unix_command to save and restore the current command line count and the token state (last_read_token, etc.). Everything else is saved by either parse_and_execute directly or the call it makes to push_stream(). The shell_input_line stuff doesn't need to be saved and restored; it's not computed until readline() returns 4/10 ---- lib/glob/glob.[ch] - glob_filename and glob_vector now take an additional `flags' arg - define GX_MARKDIRS as possible flag value for glob_filename and glob_vector lib/sh/snprintf.c - fixed some bugs with handling of `g' and `G' formats - make sure numtoa returns the fractional part correctly when passed 0 - implemented thousands grouping for `'' flag character lib/sh/rename.c - a few changes to make it more bulletproof 4/11 ---- lib/glob/glob.c - added the couple of dozen lines of code to glob_dir_to_array to finish implementing GX_MARKDIRS builtins/set.def - changed unset builtin so that it no longer considers unsetting an unset variable or function to be an error lib/readline/display.c - fix to rl_redisplay for a problem which caused display to be messed up when the last line of a multi-line prompt (possibly containing invisible characters) was longer than the screen width 4/15 ---- aclocal.m4 - use AC_DEFINE_UNQUOTED in BASH_SYS_DEFAULT_MAIL_DIR instead of enumerating all of the possible values and using AC_DEFINE 4/16 ---- Makefile.in, {builtins,support}/Makefile.in - new variables, CFLAGS_FOR_BUILD and CPPFLAGS_FOR_BUILD, substituted by `configure' - changed CCFLAGS_FOR_BUILD to BASE_CCFLAGS, removing $(CPPFLAGS); CCFLAGS and CCFLAGS_FOR_BUILD now include $(BASE_CCFLAGS) with (possibly) different values for CPPFLAGS and CFLAGS - GCC_LINT_CFLAGS now includes $(BASE_CCFLAGS) and $(CPPFLAGS) instead of CCFLAGS_FOR_BUILD - new variable, LDFLAGS_FOR_BUILD, right now equivalent to LDFLAGS - remove $(CPPFLAGS) from recipes for buildversion, mksignames, and mksyntax configure.in - compute and substitute CFLAGS_FOR_BUILD, CPPFLAGS_FOR_BUILD, and LDFLAGS_FOR_BUILD - changed qnx to use LOCAL_LDFLAGS and LOCAL_LIBS instead of putting everything in LOCAL_LDFLAGS builtins/Makefile.in - remove $(PROFILE_FLAGS) from recipe for building `mkbuiltins' - use LDFLAGS_FOR_BUILD instead of LDFLAGS in recipe for building `mkbuiltins' Makefile.in - use $(CC_FOR_BUILD) and $(CCFLAGS_FOR_BUILD) to build auxiliary test programs (printenv, recho, zecho) support/Makefile.in - use CC_FOR_BUILD and CCFLAGS_FOR_BUILD in recipe for building `man2html' lib/tilde/Makefile.in - substitute PROFILE_FLAGS, use PROFILE_FLAGS in $(CCFLAGS) 4/25 ---- Makefile.in, configure.in - moved RELSTATUS to configure.in; configure substitutes it into the generated Makefile lib/sh/snprintf.c - fix wchars() to deal with systems where MB_CUR_MAX is not a constant expression 5/2 --- lib/sh/shquote.c - add `,' to list of chars that are backslash-quoted. It doesn't hurt normal usage and prevents filenames with commas from being inappropriately split by brace expansion after using complete-into-braces 5/6 --- lib/sh/xstrchr.c - we only need the check of MB_CUR_MAX and the slow code for a few encodings, and even then only for a subset of the charset arrayfunc.c - some speedups for skipsubscript and multibyte chars from Bruno Haible locale.c - changed set_lang to call setlocale(LC_ALL, ...) if LC_ALL doesn't already have a value, but doesn't change any shell variables include/shmbutil.h - major speedups from Bruno Haible, mostly concerned with reducing the number of strlen(3) calls subst.c - change callers of macros in shmbutil.h to add extra argument as necessary - skip_single_quoted and skip_double_quoted take another argument: the length of the string; mostly useful when using multibyte chars - many speedups from precomputing string lengths at function start - fixed a small bug in de_backslash in the midst of rewriting for better efficiency {braces,make_cmd,pathexp}.c - change callers of macros in shmbutil.h to add extra argument as necessary pathexp.c - fix a one-too-far problem with multibyte chars in unquoted_glob_pattern_p braces.c - brace_gobbler takes a new argument, the length of the passed string - expand_amble takes a new argument, the length of the passed string 5/7 --- subst.c - modified remove_quoted_nulls to eliminate the memory allocation and do the copy in place using the same strategy as de_backslash lib/readline/{rldefs.h,complete.c} - new define RL_QF_OTHER_QUOTE, so _rl_find_completion_word can note that it found a quoting character other than \'" that appears in rl_completer_quote_characters 5/9 --- jobs.c - save and restore old value of jobs_list_frozen when calling trap handlers from set_job_status_and_cleanup to avoid seg faults when running recursive trap handlers 5/10 ---- builtins/common.h - new #defines to use for value of changed_dollar_vars (provides information about the caller who wants to blow away the old dollar variables) builtins/common.c - changed set_dollar_vars_changed to set changed_dollar_vars to one of the ARGS_* values depending on the caller and environment builtins/source.def - source restores the positional parameters unless the `set' builtin was called to specify a new set while not executing a shell function 5/13 ---- POSIX - new file, was in CWRU/POSIX.NOTES doc/{Makefile.in,Makefile} - changed `posix' rule to modify ../POSIX doc/mkposix - write to `POSIX' by default lib/sh/strtrans.c - when ansicstr is parsing a format string for `echo -e' (or the equivalent xpg_echo option is enabled), obey the POSIX-2001/SUSv3 standard and accept 0-3 octal digits after a leading `0' doc/{bash.1,bashref.texi} - updated `echo' description to note that up to three octal digits are now accepted following `\0' 5/16 ---- doc/Makefile.in - remove the generated documentation on `make distclean' if the build directory and source directory are not the same Makefile.in - descend into `support' subdirectory on a `make clean' and `make distclean' - remove parser-built, y.tab[ch] on a `make distclean' if the build directory and source directory are not the same support/Makefile.in - support various `clean' targets and remove man2html.o and man2html {configure,Makefile}.in - move values for DEBUG and MALLOC_DEBUG into configure.in; on by default for development versions; off by default for releases (off for profiling, too) 5/21 ---- parse.y - modified the grammar to allow a simple_list followed by yacc_EOF to terminate a command. This fixes problems with things like a backslash-newline at the end of an `eval'd string - change handle_eof_input_unit() to reset the token state before calling prompt_again(), in case the prompt to be evaluated contains a command substitution 5/23 ---- lib/readline/vi_mode.c - fix `r' command (rl_vi_change_char) when HANDLE_MULTIBYTE is defined but MB_CUR_MAX == 1 5/24 ---- lib/malloc/watch.c - don't try to print `file' argument to _watch_warn if it's null lib/malloc/malloc.c - changed guard checking code in internal_{malloc,free,realloc} to access memory as (char *) and copy into a union instead of casting and dereferencing a pointer to u_bits32_t, since that results in unaligned accesses which will cause Sparcs to upchuck 5/30 ---- [bash-2.05b-beta1 released] lib/readline/text.c - fixed a problem with rl_transpose_chars on systems supporting multibyte characters with a locale that doesn't have any multibyte chars 6/4 --- expr.c - fix a/=0 and a%=0 to throw evaluation errors rather than core dumps lib/readline/display.c - fix core dump when line wrapping a multibyte character (line accidentally dropped from the original patch) lib/readline/mbutil.c - fix reversed return value from _rl_is_mbchar_matched; fixes problem with backward-char-search 6/10 ---- lib/sh/getenv.c - fix getenv to not free value returned by find_tempenv_variable - add setenv, putenv, unsetenv for completeness 6/12 ---- shell.c - change init_noninteractive to init expand_aliases to the value of posixly_correct - don't initialize expand_aliases to posixly_correct anywhere else. This allows the -O expand_aliases invocation option to work correctly general.c - fix move_to_high_fd to not try the dup2 unless the fd loop results in an fd > 3; just return the passed file descriptor otherwise - use HIGH_FD_MAX, defined in general.h, instead of hard-coded 256 as highest file descriptor to try subst.c - in process_substitute, call move_to_high_fd with `maxfd' parameter of -1 instead of 64, so move_to_high_fd will use its maximum 6/21 ---- lib/malloc/malloc.c - don't bother calling MALLOC_MEMSET if the requested size is 0 builtins/setattr.def - note in short doc that export and readonly can take assignment statements as arguments error.c - new function, error_prolog(), to capture common error message prefix code (except for parser errors) 6/25 ---- aclocal.m4 - add tests for standard-conforming declarations for putenv and unsetenv in system header files {configure,config.h}.in - call BASH_FUNC_STD_PUTENV and BASH_FUNC_STD_UNSETENV, define HAVE_STD_GETENV and HAVE_STD_UNSETENV, respectively, if they succeed lib/sh/getenv.c - change putenv and unsetenv to take differing prototypes in stdlib.h into account 6/27 ---- [bash-2.05b-beta2 released] 6/28 ---- builtins/common.c - fix get_job_spec so that %N works when N is the size of the jobs list (%8 means job 8, but the 7th member of the jobs array, so it's OK if N == job_slots because the function returns N-1) 7/1 --- shell.c - turn off line editing if $EMACS is set to `t' 7/10 ---- builtins/set.def - remove mention of `-i' from long help doc, since it has no effect