Eric Blake wrote: > Thanks! That gnulib fix was May 12, for a regression on May 7, and I > released m4 1.4.20 on May 10, after an earlier snapshot was available > for testing prior to the point where gnulib was broken. Unfortunate > timing on my part
If you don't need any feature from gnulib's 'master' branch, you can minimize the risk of getting a broken gnulib by using a stable branch, such as currently the 'stable-202501' branch. See https://www.gnu.org/software/gnulib/manual/html_node/Stable-Branches.html > since I don't have easy access to MSVC compilation > to have noticed the broken build prior to the release. Is there a CI > engine that regularly does MSVC builds that I could have checked > before doing the release? Yes, we have such a CI, at https://github.com/gnu-m4/ci-check/actions , but the MSVC build there runs in an environment that does not have a libintl installed. And this CI runs only once per week, on Mondays. If there is a regression in gnulib on a Wednesday and you pick the newest gnulib on the following Saturday, there has not been a CI run in-between. > But the fact that Gary's version did not pick the best possible > gettext.m4 file It did. I just checked the m4-1.4.20 tarball. Its gettext infrastructure is old, but fine. > You can > also see places where both Bruno and I were fretting about other > oddities exposed by Gary's bootstrap not upgrading .m4 files in the > same way that gnulib's does, as I remember having a hard time getting > autoreconf to work: > > https://lists.gnu.org/archive/html/bug-m4/2025-04/msg00043.html This one has nothing to do with Gary's bootstrap script either, but with a bad interaction between autoreconf and autopoint. It is fixed in GNU gettext 0.24.1 and newer. Bruno
