Stage 7: add chained token support to input parser.

[m4/ericb.git] / NEWS

GNU M4 NEWS - User visible changes.
Copyright (C) 1992, 1993, 1994, 2004, 2005, 2006, 2007 Free Software
Foundation, Inc.

Version 1.4.11 - ?? ??? 2007, by ????  (git version 1.4.10a-*)

* Security fixes for the -F option, for bugs present since -F was
  introduced in 1.3: Avoid core dump with 'm4 -F file -t undefined', and
  avoid arbitrary code execution with certain file names.
* Fix regression introduced in 1.4.9b in the `divert' builtin when more
  than 512 kibibytes are saved in diversions on platforms like NetBSD where
  fopen(name,"a+") seeks to the end of the file.
* The output of the `maketemp' and `mkstemp' builtins is now quoted if a
  file was created.  This is a minor security fix, because it was possible
  (although rather unlikely) that an unquoted string could match an
  existing macro name, such that use of the `mkstemp' output would trigger
  inadvertent macro expansion and operate on the wrong file name.
* Enhance the `defn' builtin to support concatenation of multiple text
  arguments, as required by POSIX.  However, at this time, it is not
  possible to concatenate a builtin macro with anything else; a warning is
  now issued if this is attempted, although a future version of M4 may lift
  this restriction to match other implementations.
* Several improvements in `index', `regexp', and `patsubst' builtins to
  speed up typical Autoconf usage.
* Memory usage is greatly reduced in recursive macros.
* A number of portability improvements inherited from gnulib.

Version 1.4.10 - 09 Jul 2007, by Eric Blake  (CVS version 1.4.9c)

* Upgrade from GPL version 2 to GPL version 3 or later.
* A number of portability improvements inherited from gnulib.
* Avoid undefined behavior introduced in 1.4.9b in the `format' builtin
  when handling %c.  However, this area of code has never been documented,
  and currently does not match the POSIX behavior of printf(1), so it may
  have further changes in the next version.

Version 1.4.9b - 29 May 2007, by Eric Blake  (CVS version 1.4.9a)

* Fix regression introduced in 1.4.9 in the `eval' builtin when performing
  division.
* Fix regression introduced in 1.4.8 in the `-F' option that made it
  impossible to freeze more than 512 kibibytes of diverted text.
* The synclines option `-s' no longer generates sync lines in the middle of
  multiline comments or quoted strings.
* Work around a number of corner-case POSIX compliance bugs in various
  broken stdio libraries.  In particular, the `syscmd' builtin behaves
  more predictably when stdin is seekable.
* The `format' builtin now understands formats such as %a, %A, and %'hhd,
  and works around a number of platform printf bugs.  Furthermore, the
  sequence format(%*.*d,-1,-1,1) no longer outputs random data.  However,
  some non-compliant platforms such as mingw still have known bugs in
  strtod that may cause testsuite failures.
* The testsuite is improved to also run gnulib portability tests for the
  features that M4 imports from gnulib.

Version 1.4.9 - 23 Mar 2007, by Eric Blake  (CVS version 1.4.8c)

* Minor documentation and portability cleanups.

Version 1.4.8b - 24 Feb 2007, by Eric Blake  (CVS version 1.4.8a)

* Fix a regression introduced in 1.4.8 that made m4 unable to process
  files larger than 2GiB on some platforms.
* Fix a regression introduced in 1.4.8 that made m4 dump core when
  invoked as 'm4 -- file'.
* The `eval' builtin now follows C precedence rules.  Additionally, the
  short-circuit operators correctly short-circuit division by zero.  The
  previously undocumented alias of '=' meaning '==' in eval now triggers a
  deprecation warning, so that a future version of M4 can implement a form
  of variable assignment as an extension.
* The `include' builtin now affects exit status on failure, as required by
  POSIX.  Use `sinclude' if you need a successful exit status.
* The `-E'/`--fatal-warnings' command-line option now has two levels.  When
  specified only once, warnings affect exit status, but execution
  continues, so that you can see all warnings instead of fixing them one
  at a time.  To acheive 1.4.8 behavior, where the first warning
  immediately exits, specify -E twice on the command line.
* A new `--warn-macro-sequence' command-line option allows detection of
  sequences in `define' and `pushdef' definitions that match an optional
  regular expression.  The default regular expression is
  `\$\({[^}]*}\|[0-9][0-9]+\)', corresponding to the sequences that might
  not behave correctly when upgrading to the eventual M4 2.0.  By default,
  M4 2.0 will follow the POSIX requirement that a macro definition
  containing `$11' must expand to the first argument concatenated with 1,
  rather than the eleventh argument; and will take advantage of the POSIX
  wording that allows implementations to treat `${11}' as the eleventh
  argument instead of literal text.  Be aware that Autoconf 2.61 will not
  work with this option enabled with the default regular expression; but
  Autoconf 2.62 will be compatible with this option.
* Improved portability to platforms such as BSD/OS and AIX.

Version 1.4.8 - 20 November 2006, by Eric Blake  (CVS version 1.4.7a)

* The `divert' macro and `-H'/`--hashsize' command line option no longer
  cause a core dump when handed extra large values.  Also, `divert' now
  uses memory proportional to the number of diversions in use, rather than
  to the maximum diversion number encountered, so that large diversion
  numbers are less likely to exhaust system memory; and is no longer
  limited by the maximum number of file descriptors.
* The `--help' and `--version' command line options now consistently
  override all earlier options.  For example, `m4 --debugfile=trace
  --help' now no longer accidentally creates an empty file `trace'.
* The `-L'/`--nesting-limit' command line option can now be set to 0
  to remove the default limit of 1024.  However, it is still possible that
  heavily nested input can cause abrupt program termination due to stack
  overflow.
* Problems encountered when writing to standard error, such as with the
  `errprint' macro, now always cause a non-zero exit status.
* Warnings and errors issued during macro expansion are now consistently
  reported at the line where the macro name was detected, rather than
  where the close parenthesis resides.  Text wrapped by `m4wrap' now
  remembers the location that was in effect when m4wrap was invoked,
  rather than changing to line 0 and the empty string for a file.  The
  macros `__line__' and `__file__' now work correctly even as the last
  token in an included file.
* The `builtin' and `indir' macros now transparently handle builtin
  tokens generated by `defn'.
* When diversions created by the `divert' macro collect enough text that
  M4 must use temporary files, the environment variable $TMPDIR is now
  consulted, and a better effort is made to clean up those files in the
  event of a fatal signal.
* The `mkstemp' builtin is added with the same GNU semantics as `maketemp',
  based on the recommendation of POSIX to deprecate the POSIX semantics of
  `maketemp' as inherently insecure.  In GNU mode (no -G supplied on the
  command line), `maketemp' silently retains the secure GNU semantics, but
  a future release of M4 will change this to emit a warning.  In
  traditional mode (m4 -G), `maketemp' now uses the POSIX-mandated insecure
  semantics, and issues a warning that you should convert your script to
  use `mkstemp' instead.  Additionally, `mkstemp' and `maketemp' are now
  well-defined even if the template argument does not end in six `X'
  characters.
* The manual has been improved, including a new section on a composite
  macro `foreach'.
* The `changecom' and `changequote' macros now treat an empty second
  argument the same as if it were missing, rather than using the empty
  string and making it impossible to end a comment or quote.
* The `translit' macro now operates in linear instead of quadratic time,
  and is now eight-bit clean.
* The `-D', `-U', `-s', and `-t' command line options now take effect
  after any files encountered earlier on the command line, rather than up
  front, as is done in traditional implementations and required by POSIX.

Version 1.4.7 - 25 September 2006, by Eric Blake  (CVS version 1.4.6a)

* Fix regression from 1.4.5 in handling a file that ends in a macro
  expansion without arguments instead of a newline.
* The define and pushdef macros now warn when the first argument is not
  a string, rather than silently doing nothing.
* Standard input can now be read more than once, as in 'm4 - file -', and
  is not closed until all wrapped text is handled.  This makes a
  difference when stdin is not a regular file, and also fixes bugs when
  using the syscmd or esyscmd macros from wrapped text.
* When standard input is a seekable file, the m4exit, syscmd, and esyscmd
  macros now restore the current position to the next unread byte rather
  than discarding an arbitrary amount of buffered data.
* SysV command-line compatibility is no longer a goal of GNU M4; the
  focus will be instead on POSIX compatibility.  This release continues to
  support previous usage, but adds warnings in areas which will allow a
  future version of GNU M4 to use its own extensions without being tied to
  the SysV command line interface.
* The no-op compatibility command line options -B, -N, -S, -T, and
  --diversions may be withdrawn or assigned new meanings in future
  releases, so they now issue a warning if used.
* A new command line option -i replaces the compatibility -e as the
  short spelling of --interactive, for consistency with other GNU tools; a
  warning is issued if the old spelling is used, and it may be assigned new
  meaning in future releases.
* A new command line option --debugfile replaces the options -o and
  --error-output as the preferred spelling.  The old options were
  misleading in their names and inconsistent with other GNU tools; they are
  still silently accepted, but no longer documented in --help, and may be
  assigned new meanings in future releases.

Version 1.4.6 - 25 August 2006, by Eric Blake  (CVS version 1.4.5a)

* Fix buffer overruns in regexp and patsubst macros when handed a trailing
  backslash in the replacement text, or when handling \n substitutions
  beyond the number of \(\) groups.
* Fix memory leak in regexp, patsubst, and changeword macros.
* The format macro now understands %F, %g, and %G.
* When loading frozen files, m4 now exits with status 63 if version
  mismatch is detected.
* Fix bugs that occurred when invoked with stdout or stderr closed,
  and detect write failures to stdout or to the target of the
  debugfile macro.  In particular, the syscmd and esyscmd macros can
  no longer interfere with the debug stream or diversions.
* The m4exit macro now converts values outside the range 0-255 to 1.
* It is now an error if a command-line input file ends in the middle of a
  comment, matching the behavior of mid-string and mid-argument
  collection.
* The dnl macro now warns if end of file is encountered instead of a
  newline.
* The error message when end of file is encountered now uses the file and
  line where the dangling construct started, rather than `NONE:0:'.
* The debugmode and __file__ macros, and the -s/--synclines option, now
  show what directory a file was found in when the -I/--include option or
  M4PATH variable had an effect.
* The changequote and changecom macros now work with 8-bit characters, and
  quotes and comments that begin with `(' are properly recognized
  following a word.
* The new macro __program__ is added, which allows the input file to issue
  an error message that resembles messages from m4.  Warning and error
  messages have been reformatted to comply with GNU Coding Standards.
* The errprint, m4wrap, and shift macros are now recognized only with
  arguments.
* The index, substr, translit, regexp, and patsubst macros now produce
  output when given only one argument, but still warn about a missing
  second argument.
* The patsubst macro now reliably finds zero-length matches at the end
  of a string.

Version 1.4.5 - 15 July 2006, by Eric Blake  (CVS version 1.4.4c)

* Fix sysval on BeOS, OS/2, and other systems that store exit status
  in the low-order byte.  Additionally, on Unix platforms, if syscmd was
  terminated by a signal, sysval now displays the signal number shifted
  left by eight bits, to match traditional m4 implementations.
* The maketemp macro is no longer subject to platform limitations (such as
  26 or 32 max files from a given template).
* Frozen files now require that the first directive be V (version), to
  better diagnose version mismatch.  Additionally, if the F directive
  (builtin function) names an unknown builtin that existed in the m4 that
  froze the file but not in the current m4 (for example, changeword), the
  warning is deferred until an attempt is made to actually use the
  builtin.  This allows downgrading from beta m4-1.4o to stable m4-1.4.5
  without breaking autoconf.
* The format and indir macros are now recognized only with arguments.
* The eval macro no longer crashes on x86 architectures when dividing the
  minimum integer by -1.
* On systems with ecvt and fcvt, format no longer truncates trailing
  zeroes on integers printed with %.0f.  On systems without these
  functions, format is no longer subject to a buffer overflow that
  permitted arbitrary code execution.
* On native Windows builds, the macro __windows__ is provided instead of
  __unix__.  Likewise, on OS/2 builds, the macro __os2__ is provided.  This
  allows input files to determine when syscmd might behave differently.
* Fix bug in 1.4.3 patch to use \n line-endings that did not work for
  cygwin.
* When given the empty string or 0, undivert is now documented as a no-op
  rather than closing stdout, warning about a non-existent file, or trying
  to read a directory as a file.
* Many documentation improvements.  Also, the manual is now distributed
  under FDL 1.2, rather than a stricter verbatim-only license.
* Raise the -L (--nesting-limit) command line option limit from 250 to
  1024.
* The decr, incr, divert, m4exit, and substr macros treat an empty number
  as 0, issue a warning, and expand as normal; rather than issuing an error
  and expanding to the empty string.
* The eval macro now treats an empty radix argument as 10, handles radix 1,
  and treats the width argument as number of digits excluding the sign,
  for compatibility with other m4 implementations.
* The ifdef, divert, m4exit, substr, and translit macros now correctly
  ignore extra arguments.
* The popdef and undefine macros now correctly accept multiple arguments.
* Although changeword is on its last leg, if enabled, it now reverts to the
  default (faster) regexp when passed the empty string.
* The regexp and substr macros now warn and ignore a trailing backslash in
  the replacement, and warn on \n for n larger than the number of
  sub-expressions in the regexp.

Version 1.4.4b - 17 June 2006, by Eric Blake  (CVS version 1.4.4a)

* Fix a recursive push_string crashing bug, which affected changequote of
  three or more characters on some compilers.
* Use automake to fix build portability issues.
* Fix a recursive m4wrap crashing bug.
* Fix a 1 in 2**32 hash crashing bug.
* Tracing a macro by name is now persistent, even if the macro is
  subsequently undefined or redefined.  The traceon and traceoff macros no
  longer warn about undefined symbols.  This solves a crash when using
  indir on an undefined macro traced with the -t option, as well as an
  incorrect result of ifdef.  Furthermore, tracing is no longer transferred
  with builtins, solving the bug of "m4 -tm4_eval" failing to give trace
  output on the input "define(`m4_eval',defn(`eval'))m4_eval(1)".
* Fix a crash when a macro is undefined while collecting its arguments, by
  always using the definition that was in effect before argument
  collection.  This behavior matches the C pre-processor, and means that
  the sequence "define(`f',`1')f(define(`f',`2'))f" is now documented to
  result in "12", rather than the previously undocumented "22".
* Update the regex engine to fix several bugs.
* Fix a potential crash on machines where char is signed.

Version 1.4.4 - October 2005, by Gary V. Vaughan

* ./configure --infodir=/usr/share/info now works correctly.
* When any file named on the command line is missing exit with status 1.

Version 1.4.3 - March 2005, by Gary V. Vaughan

* DESTDIR installs now work correctly.
* Don't segfault with uncompilable regexps to changeword().
* Always use \n line-endings for frozen files (fixes a Windows bug).
* Portability fix for systems lacking mkstemp(3).
* Approximately 20% speed up in the common case of usage with autoconf.
* Supported on QNX 6.3.

Version 1.4.2 - August 2004, by Paul Eggert

* No user visible changes; portability bug fixes only.

Version 1.4.1 - June 2004, by Paul Eggert

* maketemp now creates an empty file with the given name, instead of merely
  returning the name of a nonexistent file.  This closes a security hole.

Version 1.4 - October 1994, by Franc,ois Pinard

(No user visible changes)
\f
Version 1.3 - September 1994, by Franc,ois Pinard

* Diversions are created as needed.  Option `-N' is still accepted, but
otherwise ignored.  Users should use only negative diversion numbers,
instead of high positive numbers, for diverting to nowhere.

* Diversions should also work faster.  No temporary files will be needed
at all if all diversions taken altogether do not use more than 512K.

* Frozen state files may be produced with the `--freeze-state' (-F)
option and later brought back through the `--reload-state' (-R) option.
\f
Version 1.2 - July 1994, by Franc,ois Pinard

* In patsubst(STRING, REGEXP, REPLACEMENT), \& in REPLACEMENT has been
changed to represent this part of STRING matched by the whole REGEXP,
instead of the whole STRING as before.  \0 does the same, but emits a
diagnostic saying it will disappear in some subsequent release.

* eval(EXPR) emits a diagnostic if EXPR has suffixed crumb.  The same for
other numeric conversions in incr(), decr(), divert(), etc.

* `--fatal-warnings' (-E) stops execution at first warning.

* `--nesting-limit=LEVEL' (-L LEVEL) sets a limit to macro nesting.
It is initially fixed at 250.

* `--word-regexp=REGEXP' (-W REGEXP) modifies macro name syntax, like
does the new `changeword(REGEXP)' macro.  This feature is experimental,
tell me your opinions about it.  You do need --enable-changeword at
configure time to get these things.  Do *not* depend on them yet.

* Trace output format is scannable by GNU Emacs' next-error function.

* Stack overflow is detected and diagnosed on some capable systems.

* Various bugs have been corrected, m4 should be more portable.  See the
ChangeLog for details.
\f
Version 1.1 - November 1993, by Franc,ois Pinard

Changes which might affect existing GNU m4 scripts:

* Option `-V' has been removed, use `--version' instead.  `--version'
writes on standard output instead of standard error, and inhibits any
script execution.

* `--no-gnu-extensions' has been renamed `--traditional'.

* In `eval', `^' used to indicate exponentiation, use `**' instead.

* The automatic undiversion which takes place at end of all input is
forced into the main output stream.

Changes which are unlikely to affect existing scripts:

* `--help' prints an usage summary on standard output.  Script execution
is then inhibited.

* `--prefix-builtins' (-P) prefixes all builtin macros by `m4_'.

* Most builtin macros for which arguments are mandatory, called without
any arguments, are no more recognized as builtin macros: they are
consequently copied verbatim to the output stream.

* `define' and `pushdef' are usable with only one argument, they give
this argument an empty definition.

* `eval' new operators for binary representation handling: `^' for
exclusive-or, `~' for the bitwise negation, `<<' and `>>' for shifts.

* `eval' recognizes the notation 0bDIGITS for binary numbers and the
notation 0rRADIX:DIGITS for numbers in any radix from 1 to 36.
\f
Version 1.0.3 - December 1992, by Franc,ois Pinard

Changes for the user:

* `dnl' outputs a diagnostic if immediately followed by `('.  Usually,
`dnl' is followed by newline or whitespace.

* `ifelse' accepts without complaining the common idiom of having only
one argument.  This is useful for introducing long comments.

* `eval' always expresses values as signed, whatever the radix.

* M4OPTS environment variable is no longer obeyed.

* `--no-warnings' option is renamed `--silent'.

* Debug lines use a new format more compatible with GNU standards.

* Various bugs have been corrected.  See the ChangeLog for details.

Changes for the installer:

* GNU m4 now uses an Autoconf-generated configure script, and should be
more easily portable in many ways.  (Cray is not supported yet).

* `make check' has been made more portable, expect no errors.

Changes for the programmer:

* Sources have been fully reindented to comply with GNU standards, and
cleaned up in many ways.

* Sources have been protoized.  Non-ANSI compilers are automatically
detected, then sources are unprotoized on the fly before compilation.

* GNU m4 uses newer versions of obstack, regex, getopt, etc.
\f
Version 1.0 - October 1991, by Rene' Seindal

* Uses GNU configure, taken from the gdb distribution.

* Uses GNU getopt(), with long option names.

* The -Q/+quiet option is added, which suppresses warnings about missing
or superflous arguments to built-in macros.

* Added default options via the M4OPTS environment variable.

* Several minor bugs have been fixed.
\f
Version 0.99 - July 1991, by Rene' Seindal

* The builtins `incr' and `decr' are now implemented without use of
`eval'.

* The builtin `indir' is added, to allow for indirect macro calls
(allows use of "illegal" macro names).

* The debugging and tracing facilities has been enhanced considerably.
See the manual for details.

* The -tMACRO option is added, marks MACRO for tracing as soon as it
is defined.

* Builtins are traced after renaming iff they were before.

* Named files can now be undiverted.

* The -Nnum option can be used to increase the number of divertions
available.

* Calling changecom without arguments now disables all comment handling.

* A bug in `dnl' is fixed.

* A bug in the multi-character quoting code is fixed.

* Several typos in the manual has been corrected.  More probably persist.
\f
Version 0.75 - November 1990, by Rene' Seindal

* Implemented search path for include files (-I option and M4PATH
environment variable).

* Implemented builtin `format' for printf-like formatting.

* Implemented builtin `regexp' for searching for regular expressions.

* Implemented builtin `patsubst' for substitution with regular
expressions.

* Implemented builtin `esyscmd', which expands to a shell commands output.

* Implemented `__file__' and `__line__' for use in error messages.

* Implemented character ranges in `translit'.

* Implemented control over debugging output.

* Implemented multi-character quotes.

* Implemented multi-character comment delimiters.

* Changed predefined macro `gnu' to `__gnu__'.

* Changed predefined macro `unix' to `__unix__', when the -G option is
not used.  With -G, `unix' is still defined.

* Added program name to error messages.

* Fixed two missing null bytes bugs.
\f
Version 0.50 - January 1990, by Rene' Seindal

* Initial beta release.
\f
Local Variables:
mode: outline
fill-column: 75
End: