Hello autoconf patchers, I notice that recently, the git master branch of Autoconf has regressed in a few cases.
1) The configure script has not been recreated a couple of times. I'm not yet sure whether the occasional autotools-rerun I've been seeing after a pull, are only due to this fact or also due to git not preserving timestamp order for checked-in regenerated files. If also the latter, maybe some git hooks can help here? 2) Recreating it helps find bugs by looking at the diff: the alignment of `configure --help' output shows a bug: AC_ARG_VAR help for EMACS and EMACS_LOAD_PATH are not aligned. Likewise, --prefix, --exec-prefix, and --docdir alignment is wrong. git-bisect tells me that this is the first offending commit: | commit 4e9078ba0b9a919ee35a1ff8b6bd8c764362f33e | Author: Eric Blake <[EMAIL PROTECTED]> | Date: Thu Oct 4 12:29:45 2007 -0600 | | Provide better short-circuiting operation. AFAICS this has not been fixed by Eric's latest commit, a83ac461c619e7c92fc82d9bd791bedb3dd41078. The patch is way to large (and too complicated) for me to investigate quickly, and I'm not yet fluent with Eric's demonstration of dissecting single patches, sorry. 3) For me, tests 45 (AS_HELP_STRING) and 315 failed when using a recent m4 1.4.x (both similar): | +/home/ralf/local/bin/m4:script.as:11: Warning: unrecognized specifier in `%0s' This seems to be fixed now, thanks! 4) A few regressions have already been found and fixed, at least one more is still pending (see Paolo's report). 5) ChangeLog contained some lines matching `^ ' (leading spaces instead of TABs), fixed by a recent commit. Add git hook or recommend clcommit? I'm wondering whether it's the goal to release as soon as the license issues are resolved, how far that may be away, whether more experimental features should be put on a git branch first. Comments appreciated. Cheers, Ralf
