mirror of
https://github.com/openai/codex.git
synced 2026-09-05 15:18:41 +00:00
daa90ac5cc
The Windows release workflow builds x64 and ARM64 in one matrix. Each target's packaging job therefore waits for all six matrix cells, even though it consumes artifacts from only the three cells for its target. In alpha.19, ARM64's builds completed at 01:40:38 UTC, but its packaging job was not created until x64 completed at 01:51:42 UTC. It then entered the x64 runner queue alongside x64 packaging. https://github.com/openai/codex/actions/runs/27484551090 Invoke the reusable workflow once per target. Each invocation retains the three bundle builds and starts symbol and package jobs as soon as those target-specific builds finish. ARM64 packaging took 6m05s on ARM in alpha.18 and 6m08s on x64 in alpha.19. Keep packaging on each target's build runner. The 11-minute ARM64 lead absorbs its runner queue and avoids consuming x64 capacity alongside the critical x64 package.
