mirror of
https://github.com/atuinsh/atuin.git
synced 2026-10-27 11:15:05 +01:00
Ellie Huxtable
a5fd5e5ee6
The shared `codspeed-macro` pool has been queueing PR benchmark runs for 20+ minutes, and CodSpeed flags its comparisons with "different runtime environments detected" — the base and head runs land on different runner generations, which makes small regressions/improvements hard to trust (see the noise on #3782). Depot's macOS runners are whole dedicated machines, so walltime numbers should be stable and comparable run to run. CodSpeed's runner ships `aarch64-apple-darwin` builds, so walltime mode is supported there. Merging this first lets `main` record baseline benchmark data on the new runner; #3782 will then rebase and re-benchmark against a like-for-like base.