m4/TODO

GNU m4 TODO - Tasks that need implementing.		-*- outline -*-
Copyright (C) 2000, 2001, 2006, 2007, 2008 Free Software Foundation, Inc.

Tell the maintainers at <bug-m4@gnu.org> if you feel like volunteering
for any of these ideas or if you have others to add.

* KNOWN BUGS

  + The following patch needs to be ported from the branch:
    http://lists.gnu.org/archive/html/m4-patches/2007-03/msg00005.html
    May be a couple of issues in that thread.

  + The format builtin needs more power to be like printf(1):
    http://lists.gnu.org/archive/html/m4-discuss/2007-05/msg00015.html
    But be aware of compatibility issues in making too many changes.

  + The test case `other-tests/stackovf.test' does not work.

  + stack overflow is basically broken

    The routines to detect stack overflow throuh segv  are basically
    broken. The idea may be fine, but it ends up calling a sigv handler
    that uses gettext, printf, stdout... all of which are definitely NOT
    sig-safe.  Pity, because the hardcode routine is basically very careful
    to use write(2), to avoid this.

    I haven't checked that translation magic is signal-safe.
    - I doubt it;
    - stdio is not signal-safe in any kind of portable setting anyways.
    --
    Marc Espie
    espie@schutzenberger.liafa.jussieu.fr


* FEATURES OR PROBLEMS

  + m4 should keep an ``execution stack'' of macros, which applications could
    use in their error messages.

  + Implement discarding comment delimiters with the syntax table.

  + Implement qindir.  Like indir, except that the result of the macro call
    is not expanded.  Because the input stack might contain a file or a
    string, it is probably best achieved by making note that the TOS input
    should be copied rather than rescanned.

      $ echo "a'b" > f
      $ m4
      define(a,z)dnl
      include(f)dnl
      z'b
      indir(`include', f)dnl
      z'b
      qindir(`include', f)dnl
      a'b
      define(b,NONO)dnl
      patsubst(qindir(`include', f), `b', x)
      z'x

  + Use the TOS input quoting for qindir to fix this undesirable POSIX
    behavior:

      define(`x', -'-)
      define(y, defn(`x'))
      y
      --'
    --
    Stepan Kasal <kasal@ucw.cz>

  + If configured --with-gmp for multiple precision arithmetic there are
    some warnings, but it passes the tests.

  + Make m4 show include dependencies like gcc so Makefile targets are
    updated when their (included) input files are updated (Erick B).

  + Add support for wide character sets.


* OPTIMIZATION AND CLEAN UP

  + Have NULs go really undisturbed through GNU m4
	GNU m4 is lousy regarding NULs in streams (this would require
	maintaining the string lengths, and avoiding strlen, strcpy,
	etc.).  (Almost there, once argv_ref is ported).

  + The argument count limits are handled for all tokens passed around by
    the internals:  we should enable attaching these values to text macros
    too.

  + The context parameter is just a placeholder for formerly global state.
    We should be making the library reentrant so that multiple instances
    of m4 can be run in the same process at the same time.

  + The path management stuff (in path.c/m4private.h) is reinventing the
    wheel.  There are a bunch of fast path management and search functions
    in ltdl.c:  These need to be sanitized, exported through ltdl.h, and
    then wrapped by the m4module.h path api.  path.c can probably be removed
    entirely at that point.

* MODULE SPECIFIC ISSUES

  + Some way of linking a module statically is needed, for systems
    without support for dynamic loading.

  + Some sort of module interface versioning system needs to be implemented
    in the module loader and the freezer so that m4 can tell if it is being
    asked to load a frozen file that requires versions of modules with
    interface versions unsupported by the current release.

  + The module API should probably use functions, not data exports:
    http://lists.gnu.org/archive/html/m4-patches/2007-09/msg00012.html
    Actually, there are several good ideas for module in that thread.

  + Setting of the module search path within m4 scripts:

	append(__modulepath__, `/some/modules/live/here')

  + Module autoloader.  This would allow an m4 core with no builtins except
    for loadmodule().  A default startup script would mark the recognised
    set of builtins for autoload from the installed module directory on first
    use.  A new cli parameter would inhibit initialisation from this script,
    so that customised m4 interpreters could be built on the fly!

  + The module loader needs to differentiate between modules that are in
    memory and modules that are loaded (i.e. visible) from various context
    structures.

  + The perl module should only be built if a suitable perl interpreter
    is found on the build machine.  For that matter, it has been a
    while since the perl module has even been built, and perl has had
    new releases in the meantime.

* OTHER TOOLS

  + Copy coreutils' idea of using rsync, not wget, for grabbing .po
    files efficiently.

  + Use gnulib's build-aux/git-version-gen for reliable version
    numbers, now that CVS timestamps are no longer available.