Hola. Y'all yanked the forced DEPEND in autotools, ok, understand that. 'Cept anyform of autoreconf call triggers autoconf --trace which is a 2.59 thing How you handle it (sticking a note/check into the code or replacing the dep) is up to you, but that's one helluva gotcha :)
Actually... better addition. Unless people add WANT_AUTOCONF (or whatever the heck that var was) to the ebuild, the autotools_check_macro will always return false on a box that has autoconf defaulted to <autoconf-2.59 You could check the autoconf --trace return, and echo on failure also- tis a hack though, that forces rerun of everything
Sit tight until somebody tries it on a autoconf-2.13 old pos, and then fix it ? :) PS, until further notice, all versions of autoconf are anyhow installed, so no need to worry about if its present or not.
well, since _my_ box seems to default to 2.13 (or at the very least, the source I was regenerating was last regenerated via 2.13), it's an issue :P The nasty aspect of this is the fact that it doesn't handle 2.13 worth a damn; specifically, it won't regenerate anything if the default autoconf == autoconf-2.13 Granted I have a work around (WANT_AUTOCONF=2.5), but the silent failure/ignore == bad imo ;)
Specifics ?
Umm... Specifics? autoconf-2.13 lacks --trace arg , yet all of the underlying autotools checks of what to run rely on --trace. If autoconf chooses to use 2.13, none of the autoconf --trace tests will work, and none of the source will be rebuilt. Silent error. bug 105858 is where I noticed this; grab the patch there, and removed the WANT_AUTOCONF=2.5, and it should implode on running fex.
Yes, I know .. I just wanted to know what package used the autoconf-2.13 ... Anyhow, if it can use 2.59, I do not see a problem with how you solved it, as I will imagine that it should be the better idea to use newer if its needed for darwin support :/ I will look into it either way.
*nudge*, anything on this?
List of older stuff still using autoconf-2.13 (and autotools.eclass - or would like to) would still be nice. If it was one or two, we could rather fix them to use the tools correctly than the eclass, than adding junk that only one or two use anyhow.
Okay I fixed this with a bit of input from exg who was able to explain me where the problem lied, not sure why I never hit this before. Anyway, as tracing works perfectly fine with autoconf 2.5 _also if the underlying autoconf is 2.1x_, I've changed the eclass to force autoconf 2.5 for the tracing only, the rest will be run as the wrapper prefers. Considered fixed unless it's not (oh my god I'm talking like ciaranm, please spare me from this!).
