diff options
| author | user <user@clank> | 2026-08-27 14:43:02 +0200 |
|---|---|---|
| committer | user <user@clank> | 2026-08-27 14:43:02 +0200 |
| commit | 36dd24325b3a070389d4d45c98dc5822ea9d2c27 (patch) | |
| tree | 507040ad2ccc6ecd2fc1b3b13029d708ffa1d15f /tests | |
| parent | release: one 'make release' for the three target binaries (diff) | |
| download | slosh-36dd24325b3a070389d4d45c98dc5822ea9d2c27.tar.gz slosh-36dd24325b3a070389d4d45c98dc5822ea9d2c27.tar.xz slosh-36dd24325b3a070389d4d45c98dc5822ea9d2c27.zip | |
macos-remote-build: tolerate notarization ticket propagation lag
The post-notarization Gatekeeper check ran spctl exactly once, immediately.
A bare CLI binary cannot be stapled (stapler only takes .pkg/.dmg/.app), so
there is nothing on disk to carry the ticket and Gatekeeper must fetch it
online from Apple's CloudKit -- which lags the notary's own 'Accepted' by
seconds to a couple of minutes. The single check raced that propagation and
failed the whole release with 'Unnotarized Developer ID' for a build Apple
had already accepted (status: Accepted), taking the good linux tarballs and
the SHA256SUMS down with it.
Poll spctl instead: pass the instant the ticket appears, and if it still has
not after ~2 min, report it as pending and let the release stand -- the notary
verdict is authoritative and codesign already vouched for the signature. Any
other rejection still fails hard, so a real signing problem is still caught.
The accepted vs pending states are told apart by 'Notarized' vs 'Unnotarized'
so there is no false positive.
Diffstat (limited to 'tests')
0 files changed, 0 insertions, 0 deletions
