You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Selftest 340 flakes once in 4096 runs: an unreadable pg_config's stamp key keeps 3 hex characters #1478
That is 4096 values. Selftest 340's check "two unreadable pg_configs do not alias onto one stamp" builds its two broken pg_configs in a random mktemp directory, so they collide about once in 4096 runs. #1476's suites (PG 18) job hit one (run 38855440878):
>> FAIL two unreadable pg_configs do not alias onto one stamp: got [SAME] want [different]
The same head passed twice locally.
Reproduced deterministically with lib.sh's own function. In /tmp/pgc-collide-870, pg_config.broken and pg_config.broken2 both get /tree/.pgc_source_stamp.0.nolib235: SAME. At 8 characters they differ (23593d63 against 2356c898). The readable branch already keeps 8 characters.
Fix
Keep 8 characters in the nolib branch. Add a selftest arm that uses those two colliding paths, so the property is tested every run, not once in 4096.
Activity
test: an unreadable pg_config's stamp key keeps 8 hex characters, not…
test: selftest 340's new collision arm gets its ledger row (#1478)
test: an unreadable pg_config's stamp key keeps 8 hex characters, not…
test: selftest 340's new collision arm gets its ledger row (#1478)
test: an unreadable pg_config's stamp key keeps 8 hex characters, not…
test: selftest 340's new collision arm gets its ledger row (#1478)
Merge pull request #1479 from commandprompt/fix/1478-stamp-key-width