Refine complex_graph regex_matches partial suppressions
#1719
Conversation
This builds on GitoxideLabs#1635 by: - Updating the comment about GitoxideLabs#1622, including by adding links to Git mailing list posts by Aarni Koskela, who discovered the bug that turns out to be the cause of this, and Patrick Steinhardt, who analyzed the bug and wrote a fix (currently in testing); and GitoxideLabs#1622 (comment), which summarizes that and reports on its connection to GitoxideLabs#1622. - Narrowing the partial suppression of the failing test code (which consists of conditionally using `parse_spec_no_baseline` instead of `parse_spec` in some assertions) so that it is only done if Git is at one of the versions that is known to be affected. If any future Git versions are affected, such as by the currently cooking patch not being merged as soon as I expect, then this could potentially fail on CI again. But that is something we would probably want to find out about.
EliahKagan
force-pushed
the
run-ci/complex-graph-no-baseline
branch
from
72eacb8 to
f4b4bf0
Compare
|
The pure-rust-build job is stalled on the This occasionally happens locally but I don't think I've seen it on CI before. Maybe using the |
|
Thanks for the hint - I cancelled and restarted and would think that it will go through now. From my experience it's very rare indeed, and I'd hope it stays that way. |
The bug in Git that causes GitoxideLabs#1622 has been fixed since 2.48.0 and does not affect any versions prior to 2.47, but the fix is not backported to subsequent 2.47.* point releases. In particular, 2.47.2 has been released, with backported security fixes, but it does not have backported fixes for this non-security bug. This builds on GitoxideLabs#1635 and GitoxideLabs#1719 by updating the range of Git versions where we skip the affected baseline checks on CI (for platforms this affects on our CI) from `(2, 47, 0)..(2, 47, 2)` to `(2, 47, 0)..(2, 48, 0)`, i.e., with the exclusive upper bound changed from 2.47.2 to the correct value of 2.48.0. This also revises the comments accordingly. For further details, see: GitoxideLabs#1622 (comment) (This change is item (1) there.)
This builds on #1635 along the lines of the suggestion in #1622 (comment), by:
Updating the comment about #1622, including by adding links to Git mailing list posts by Aarni Koskela, who discovered the bug that turns out to be the cause of this, and Patrick Steinhardt, who analyzed the bug and wrote a fix (currently in testing); and #1622 (comment), which summarizes that and reports on its connection to #1622.
Narrowing the partial suppression of the failing test code (which consists of conditionally using
parse_spec_no_baselineinstead ofparse_specin some assertions) so that it is only done if Git is at one of the versions that is known to be affected.If any future Git versions are affected, such as by the currently cooking patch not being merged as soon as I expect, then this could potentially fail on CI again. But that is something we would probably want to find out about.