Eric Blake wrote: > Simon Josefsson <simon <at> josefsson.org> writes: > >> I get errors from autoreconf when bootstrapping Libidn on a Ubuntu 8.04 >> LTS system, and removing the recent warn-on-use.m4 file fixes it. Is >> there something in the file that requires a recent GNU M4? Ubuntu 8.04 >> has GNU M4 1.1.10, and according to gnulib's DEPENDENCIES 1.4.5 or later >> should be OK. > > m4 1.4.6 or better is preferred (and in fact, enforced by newer autoconf), but > yes, 1.4.5 is okay with older autoconf. > > The problem is not old m4, but old autoconf. Which version are you using? > > Meanwhile, I can work around the problem by making sure warn-on-use.m4 defines > enough helper macros if they are not already defined by autoconf. I just need > to know which helper macros weren't available. We already guarantee > m4_foreach_w in gnulib-common.m4, so I'm guessing that it might be AS_VAR_IF > or > AS_VAR_SET. > > <time passes> > > Sure enough, we have other uses of AS_VAR_SET in gnulib, but I appear to be > the > first user of AS_VAR_IF (well, warnings.m4 uses gl_AS_VAR_IF); that macro was > only introduced in autoconf 2.63b, even though gnulib still claims to work > with > as old as 2.59. > > So, the right patch would be to move gl_AS_VAR_IF out of warnings.m4, and make > gnulib-common.m4 define a fallback AS_VAR_IF. I'll take care of that. > > Meanwhile, based on the recent security hole in older automake, should we bite > the bullet and require automake 1.10.3 or better (which in turn bumps the > minimum autoconf requirement from 2.59 up to 2.60)?
That sounds worthwhile to me. However, I'd go a little further and require 1.10.x, 3 <= x or 1.11.y, 1 <= y -------------- or maybe just exclude the buggy 1.11 release, like require 1.10.3, but explicitly exclude 1.11. But that doesn't address the problem with "1.11a", which is what I tend to use, but which may come before or after the 1.11.1 release. IMHO, this is a good excuse to make automake switch to using "git describe"-based --version output.