Background
Wave 2.9's standard switch recipe (commit 4f41c99) handles the case where Y is warm + X is canonical via:
- worktree_remove(Y) — release Y's branch
- evacuate(X) →
stash + checkout Y in main + worktree_add X + pop in worktree
= 5 git ops per repo.
The fast-path 3-checkout swap (originally suggested by user during 2.9 design — see ~/.claude/plans/wave-2-9-switch-canonical.md "Atomicity" section, Layer 2) replaces that with:
cd main && git stash push -u (if dirty)cd main && git checkout <default> — frees Xcd worktree-Y && git checkout X — worktree adopts X, Y freedcd main && git checkout Ymv .canopy/worktrees/Y/<repo> .canopy/worktrees/X/<repo> — naming convention demands the dir name match the featuregit worktree repair — fix gitlinks after mvcd new-X-worktree && git stash pop (if stashed)
= 7 ops per repo (more than the standard recipe!).
Why "fast" then?
The win is purely from skipping git worktree add's full-file-walk checkout. For huge repos that's 5-30 sec; for canopy's small test repos it's negligible.
Cost / risk
- More steps = more half-flip surface, especially without the rollback walker (issue #2)
mv + git worktree repair are new failure modes- The naming-rename step is non-trivial: it has to update the .git/worktrees/ dir too
- Conditional logic (when does fast-path apply? what if main is dirty? what if both X and Y have warm worktrees but the pre-check missed Y?) — adds branching to switch.py
Verdict (deferred)
Ship only when:
- Real-world latency complaint from a user with a huge-repo workspace
- Or after issue #2 (journal + rollback walker) lands so the extra steps don't increase blast radius
For now, the standard recipe is correct, predictable, and scales fine for canopy's typical workspaces.
Related
Background
Wave 2.9's standard switch recipe (commit 4f41c99) handles the case where Y is warm + X is canonical via:
stash + checkout Y in main + worktree_add X + pop in worktree= 5 git ops per repo.
The fast-path 3-checkout swap (originally suggested by user during 2.9 design — see
~/.claude/plans/wave-2-9-switch-canonical.md"Atomicity" section, Layer 2) replaces that with:cd main && git stash push -u(if dirty)cd main && git checkout <default>— frees Xcd worktree-Y && git checkout X— worktree adopts X, Y freedcd main && git checkout Ymv .canopy/worktrees/Y/<repo> .canopy/worktrees/X/<repo>— naming convention demands the dir name match the featuregit worktree repair— fix gitlinks after mvcd new-X-worktree && git stash pop(if stashed)= 7 ops per repo (more than the standard recipe!).
Why "fast" then?
The win is purely from skipping
git worktree add's full-file-walk checkout. For huge repos that's 5-30 sec; for canopy's small test repos it's negligible.Cost / risk
mv+git worktree repairare new failure modesVerdict (deferred)
Ship only when:
For now, the standard recipe is correct, predictable, and scales fine for canopy's typical workspaces.
Related
~/.claude/plans/wave-2-9-switch-canonical.md(Atomicity → Layer 2)