So, a few things about this bug. It's almost certainly not a glibc bug, but rather a bug in the bundled library that gets (ab)used to handle POSIX extended regexes in non-UTF-8 locales. The substing failure doesn't show when either using PCRE and a C locale, or using any combination of flags in a UTF-8 locale.
Another point worth noting is that it seems completely daft to use a substring match to detect that you're on an architecture where substring matching is broken. My original patch that just forced PCRE on all arches is sane enough, in that PCRE works well everywhere and has a sane test suite to make sure that it the case. That said, the reason I hadn't forwarded my patch to Debian yet was because it's a hideous workaround and, obviously, the root cause needs to be hunted down, as this only helps r-base's install routine, but does nothing for applications that actually call into regex routines in a C locale. Anyhow, I think that having arch-specific testing and twiddling there probably just makes things even less obvious and harder to track and debug later, but that's just my opinion. Your call, I suppose. ... Adam PS: (FWIW, the original patch was mine, not Luca's, it seems to have gotten mis/re-attributed along the way...) -- To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org