testing numprocesses to fix pytest-xdist tests
treefmt / nix fmt (pull_request) Successful in 4s
build_systems / build-bob (pull_request) Failing after 10m48s
build_systems / build-rhapsody-in-green (pull_request) Failing after 30m11s
build_systems / build-jeeves (pull_request) Failing after 1h12m27s
build_systems / build-portal-1 (pull_request) Canceled after 1h46m30s
build_systems / build-brain (pull_request) Canceled after 1h46m45s
pytest / pytest (pull_request) Failing after 3h0m1s
test ebook search / test-ebook-search (pull_request) Failing after 3h0m0s

This commit is contained in:
2026-09-11 14:57:01 -04:00
parent 0ad20c09e3
commit 5c54354890
+12 -3
View File
@@ -4,9 +4,10 @@
# nixpkgs binary cache cannot be used and upstream test suites run locally.
# The jeeves builder uses /tmp/nix-builds so filesystem tests run on tmpfs
# instead of ZFS with normalization=formD and utf8only=on; those tests remain
# enabled. The remaining exclusions cover a UDP readiness race and
# architecture-dependent floating-point differences whose risk we accept for
# our workloads. Keep these exceptions visible until their causes are fixed.
# enabled. The remaining workarounds cover UDP readiness and nested-worker
# scheduling races, plus architecture-dependent floating-point differences
# whose risk we accept for our workloads. Keep these exceptions visible until
# their causes are fixed.
_final: prev: {
gnutls = prev.gnutls.overrideAttrs (old: {
# This test uses a fixed four-second sleep instead of checking UDP
@@ -18,6 +19,14 @@ _final: prev: {
pythonPackagesExtensions = prev.pythonPackagesExtensions ++ [
(_pythonFinal: pythonPrev: {
pytest-xdist = pythonPrev.pytest-xdist.overridePythonAttrs (old: {
# Its acceptance tests start nested worker pools. Limit the outer pool
# so deliberately crashing inner workers do not race under heavy load.
preCheck = builtins.replaceStrings [ "--numprocesses=$NIX_BUILD_CORES" ] [ "--numprocesses=8" ] (
old.preCheck or ""
);
});
scipy = pythonPrev.scipy.overridePythonAttrs (old: {
# x86-64-v3 FFT implementations produce rounding differences outside
# these tests' strict tolerances. We accept the numerical-precision