On Sun, Jun 07, 2026 at 08:59:11PM +0100, Gavin Smith wrote: > The Solaris 10 failure is a problem as it shows that the SWIG code may > not be reliable if it is enabled. Will you be able to investigate the > failure?
I could not reproduce on cfarm, even after Dagobert added swig, because iconv is not found in tta although it is found by the main configure. There could be some issues with Perl compile flags again. But with the new logs in opencsw, it is clear to me that the difference is because old Perl sorts differently than the C code (through SWIG), which is in line with recent Perl (probably difference in unicode table), we already skip many tests of sorting in the test suite for this. I skipped the comparison using the same criteria. I also checked manually that the new test that has the same code fails on cfarm. As a side note, we do not test automatically the old Perl sorting difference anywhere anymore, now that XS is used in opencsw (and the github CI only has too recent Perls). It is possible to manually test on cfarm, but it is not very practical. The github CI is now all good, with swig when possible and C texi2any for platforms that use XS. Netbsd fails early during system setup, so its status is unknown. -- Pat
