alma9-specific problem with ppc64le QEMU + pthreads

[mfansler]

We're observing segfaults in a library that uses pthreads, documented in conda-forge/r-base-feedstock#366. Posting here in case anyone else is seeing this in other feedstocks.

This only seems to arise when running QEMU ppc64le on alma9 images and occurs in a cleanup hook where the threads are presumably terminated. I believe we first started seeing this in mid-January 2025 and it impacts both new builds and previously successful builds.

Switching to cos7 or alma8 images resolves the issue. All documented failures in R ecosystem have r-cli as a common denominator, however, the upstream package is very mature, which raises the likelihood that the issue could happen elsewhere. Please report if this is observed in other feedstocks.

[mfansler]

In the direction of this being package-specific, there are some related Issues upstream (though not about QEMU on alma9):

• Crash when R process ends r-lib/cli#334
• Valgrind error in package that depends on {cli} r-lib/cli#311
• double clic_unload r-lib/cli#319

[mfansler]

We end up resolving this using package-specific flags (conda-forge/r-cli-feedstock#42). I'm going to close, but if anything does popup elsewhere we can reopen.