On Mon, Aug 25, 2025 at 04:30:54PM +0200, Patrice Dumas wrote: > On Mon, Aug 25, 2025 at 11:55:44AM +0200, Bruno Haible wrote: > > Hi Patrice, > > > > Last week, you investigated and fixed 84 "ERRORS". > > > > Today's CI still shows a regression, compared to two weeks ago: > > On OpenBSD, the swig/perl tests fail: > > Indeed, and also on Cygwin. I actually already had a look, but I didn't > have time to report. > > I think that this is the same issue that was discussed some months ago > for the XS modules used internally by texi2any, the loader does not find > the libraries since they are not in the locations it looks for and there > is no rpath or similar either: > https://lists.gnu.org/archive/html/bug-texinfo/2024-11/msg00098.html > > At that time, Gavin found a solution, by modifying our XS loading code > such that the libraries the XS extension depends on are pre-loaded: > https://lists.gnu.org/archive/html/bug-texinfo/2024-11/msg00106.html > > This solution cannot be used as is for the SWIG module, as the > generated Perl module uses the usual Dynaloader Perl code to load the > extension binary object, and not some code we could modify. The > corresponding code is very simple and short, though (at the very > beginning of tta/swig/perl/Texinfo.pm), so we could use a regexp or > something like that to replace by a call to our loader instead of > Dynaloader. But I am not sure that it is the way to go, it would > probably be difficult to maintain.
The simplest solution would be to disable the SWIG modules by default, as they are not used for anything useful for users, as far as I am aware. Or maybe the CI could be amended to provide the --without-swig flag to 'configure' on OpenBSD, Cygwin and any others where this problem occurs.
