Created attachment 922351 [details] dev-haskell:crypton-0.32:20250324-003445.log When rebuilding as part of an "emerge --emptytree @world" run (because I changed my CFLAGS from -march=core2 to -march=alderlake, which I am led to believe is correct for my CPU), =dev-haskell/crypton-0.32 fails its test suite. It appears to fail with a general-protection fault in the second test under "Curve448" under "KATs". The GPF doesn't appear in the build log, but only in dmesg: [4428351.279295] traps: test-crypton[9499] general protection fault ip:7fbaf82ae512 sp:7ffe8000b900 error:0 in libHScrypton-0.32-Cag5rbRt4SGErKbAB16Pie-ghc9.2.8.so[7fbaf809a000+236000]
Created attachment 922352 [details] emerge-info.txt
Could you try a build without AESNI (-support-aesni) and rerun the tests ? Upstream reports a recurrent number of breakage for specific architecures. Since it might be sensitive cryptographic code, reporting there too could be a good thing. It should be noted that the version running in the 'gentoo' repository is 0.32 where as upstream is 1.0.2. Closest minor version is 0.34. This is highly suspicious to me. The 'haskell' overlay has a more recent version '1.0.1'. If dev-haskell/crypton::gentoo is not directly pulled by another ebuild, could you try using dev-haskell/crypton::haskell ? See https://github.com/gentoo-haskell/gentoo-haskell/blob/master/dev-haskell/crypton/crypton-1.0.1.ebuild See https://github.com/gentoo-haskell/gentoo-haskell/blob/master/README.rst See https://wiki.gentoo.org/wiki/Haskell#Getting_started
I tried building with USE="-support-aesni", and the test phase failed in the same way. This is getting pulled in as a dep of Pandoc, so *installing* a version from ::haskell would require significant work to have a consistent depgraph between that, dev-util/shellcheck, and the one or two other implemented-in-Haskell packages I have installed via @world. (An initial "emerge -p --complete-graph --autounmask=y" with the latest version of all of those @world packages reports dependency conflicts)---which I *could* likely work through eventually.) But I tried "ebuild /var/db/repos/haskell/dev-haskell/crypton/crypton-1.0.1.ebuild clean install", and the test suite of version 1.0.1 completed without error.
