| Summary: | =dev-python/pythran-0.10.0 fails tests | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Marek Szuba (RETIRED) <marecki> |
| Component: | Current packages | Assignee: | Python Gentoo Team <python> |
| Status: | CONFIRMED --- | ||
| Severity: | normal | CC: | mgorny |
| Priority: | Normal | Keywords: | TESTFAILURE |
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| URL: | https://dev.gentoo.org/~marecki/bugs/pythran-0.10.0-build.log.xz | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
| Attachments: | emerge --info | ||
|
Description
Marek Szuba (RETIRED)
2021-10-09 09:08:45 UTC
Created attachment 744063 [details]
emerge --info
build.log provided as a link because even when compressed and stripped to test output only it is too large for Bugzilla. I've just finished trying to emerge this ebuild on amd64, same test failures. No idea if it's the same problem as in Bug #802264, as the latter hasn't got a build log attached. The problem is that pythran test suite (or to be precise the constructor of its TestEnv class, see pythran/tests/__init__.py) adds -Werror and -Wno-absolute-value to compiler CXX flags. The latter is not valid for C++, and the former causes the presence of the latter to be treated as an error. Removing the two from PYTHRAN_CXX_FLAGS in the aforementioned file makes the test suite pass, on amd64 anyway. Could you attach the amd64 build log? The riscv log suggests it's actually failing because of:
/var/tmp/portage/dev-python/pythran-0.10.0/work/pythran-0.10.0/pythran/xsimd/types/xsimd_types_include.hpp:85:24: error: #warning "No SIMD instructions set detected, please enable SIMD instructions or activate the fallback mode. (e.g. for x86 -march=native or for ARM -mfpu=neon)" [-Werror=cpp]
85 | #warning "No SIMD instructions set detected, please enable SIMD instructions or activate the fallback mode. (e.g. for x86 -march=native or for ARM -mfpu=neon)"
| ^~~~~~~
(the failing amd64 log, to be more precise) Will have to reproduce the amd64 build log but at least some of the errors in it are the same as on riscv - i.e. g++ invocation fails with exit code 1. Conversely, having applied the aforementioned change on riscv has fixed at least some (I expect the full test run to conclude some time tomorrow, it's VERY slow on my system) of the previously observed test failures. As for the xsimd error you have quoted, pythran ebuilds should probably set -DXSIMD_ENABLE_FALLBACK -DXSIMD_SKIP_ON_WERROR - the former to enable fallback code on systems with no SSE3, AVX or Neon, the latter to suppress the compiler warning (escalated to an error by -Werror) appearing on such systems regardless of whether fallback mode has been enabled or not. On the other hand, perhaps we should only set the former and ditch -Werror from the test suite - according to the documentation Pythran does not use xsimd unless explicitly requested by setting -DHAVE_SIMD on its command line, and if I see correctly none of the tests actually use this feature (the only mention of SIMD in pythran/tests/ is in the context of OpenMP). |
