Instead of testing one commit per round, pick N evenly-spaced commits
and test them all at once.
With n=7 and 1000 commits you're looking at ~4 rounds instead of 10.
Git has the primitives for this (git rev-list --bisect-all, etc.).
Depends on having decent build caching in place, otherwise you're
just burning resources. Guillaume noted this was considered before
for KernelCI and wasn't worth it without caching.
Existing tools: git-pisect (Perl), risect (Rust).
Instead of testing one commit per round, pick N evenly-spaced commits
and test them all at once.
With n=7 and 1000 commits you're looking at ~4 rounds instead of 10.
Git has the primitives for this (git rev-list --bisect-all, etc.).
Depends on having decent build caching in place, otherwise you're
just burning resources. Guillaume noted this was considered before
for KernelCI and wasn't worth it without caching.
Existing tools: git-pisect (Perl), risect (Rust).