PATCH_LIBTOOL_CHANGING_CMDS_IFS
, 

2006-10-13

Guido U. Draheim <guidod@gmx.de>

GPLWithACException
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation. As a special exception, the respective Autoconf Macro's copyright owner gives unlimited permission to copy, distribute and modify the configure scripts that are the output of Autoconf when processing the Macro. You need not follow the terms of the GNU General Public License when using or distributing such scripts

cryp.to ac-archive's Miscellaneous (released)

Do a `grep "^[a-z]*_cmds=" libtool` - these are "subroutines" encoded by libtool.m4 into the generated libtool script. Since libtool assumes that there is no subroutine-facility in the shell invoked, these are not actually subroutines, but actually a "list of commands". This looks correct, but the command separator is not ";" - it is "~", the tilde character.

Now, grep again, look for `grep 'IFS="~"' libtool` and see that libtool scripting uses a for-loop on the command-list, i.e for cmd in $some_cmds. This works correctly when the IFS was modified, where IFS stands for "input field separator" which is whitespace characters by default.

The problem: I have some real-world filesystems where there are directories using "~" inside of them, to be more to the point, it is a change control management software that uses source repositories of the form "path/master/project~version/src" and libtool has the tendency to resolve any symlinks so that it will paste such path into the $_cmds script when it gets evaluated a number of times.

This script is a workaround: I do not know why the ";" was not chosen as the IFS, perhaps it has some weird interactions in some shells since it is also the default record separator being one time bigger in context than the argument separator. I have made good success however with using "?" as the IFS, since there is no path-name that uses a question mark, and there is no _cmds ever around that uses "?" for some thing.

Oh yes, there are some usages of "*" to match shell-wise at the output file of some tool, so that might have triggered the choice to not use "?" in the first place - but in real life it never occured that a _cmds script was created that has gone to use "?". And so, this ac-macro exchanges the s/~/?/g in configured _cmds variables and replaces all occurences of s/IFS="~"/IFS="?"/ - and it all works smooth now.

AC_DEFUN([PATCH_LIBTOOL_CHANGING_CMDS_IFS],
[# patch libtool to change $_cmds IFS from ~ to ? character
if grep "^[[_$as_cr_letters]]_cmds=.*[[?]]" libtool >/dev/null; then
  AC_MSG_WARN(dnl
  [patching libtool skipped - _cmds already contain question marks])
else
 AC_MSG_RESULT([patching libtool to change cmds IFS from ~ to ?])
    test -f libtool.old || (mv libtool libtool.old && cp libtool.old libtool)
    sed -e "/^[[_$as_cr_letters]]*_cmds=/s/~/?/g" -e 's/IFS="~"/IFS="?"/g' \
        -e "s/IFS='~'/IFS='?'/g"    libtool > libtool.new
    (test -s libtool.new || rm libtool.new) 2>/dev/null
    test -f libtool.new && mv libtool.new libtool # not 2>/dev/null !!
    test -f libtool     || mv libtool.old libtool
fi
])