starred/orca
1
0
Fork 0
mirror of https://github.com/stablyai/orca.git synced 2026-10-22 08:03:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs(worktree): spell out why the retirement bound sits above the pool

Browse Source 
Names the trap directly: the neighbouring 50/200 bounds cap histories, so
lowering this one to match them would silently start reissuing names whose
agent state is still on disk. Also states that oldest-first eviction is a
deliberate least-bad choice rather than a neutral one.
This commit is contained in:
Brennan Benson
2026-08-13 18:01:10 -07:00
parent 9234cb4055
commit f583faebc7
1 changed files with 9 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/main/persistence.ts
+9 -5
View File
@@ -2335,13 +2335,17 @@ const MAX_CLAUDE_LIVE_PTY_SESSION_IDS = 200
// Why: bound removed-SSH-target history so remove/re-add churn can't grow the file unbounded.
const MAX_REMOVED_SSH_TARGET_TOMBSTONES = 50
// Why: bound per-repo retirement so long-lived repos can't grow the file unbounded. Deliberately
// far above the 552-name pool — a cap below it would start reissuing names and reintroduce the bug
// this registry exists to prevent, so only `-2`/`-3` tier accumulation can ever reach it.
// Bounds per-repo retirement so a long-lived repo can't grow the state file unbounded.
//
// DO NOT lower this to match the two bounds above. Those cap histories, where evicting the oldest
// entry is harmless. This set is a correctness guarantee, so any cap below the 552-name pool hands
// back a name whose agent conversation is still on disk — the exact bug this registry exists to
// prevent. Only `-2`/`-3` tier accumulation can reach 2000, at roughly 21KB per repo.
const MAX_RETIRED_WORKTREE_NAMES_PER_REPO = 2000
/** Oldest-first eviction: entries are appended in retirement order, and the oldest paths are the
* ones whose agent state is likeliest to be gone already. */
/** Oldest-first once the bound bites — a deliberate least-bad degradation rather than a neutral
* policy. Something must be dropped, and the oldest directories are the likeliest to have been
* reclaimed since, so their names are the safest of a bad set to reissue. */
function boundRetiredWorktreeNames(names: readonly string[]): string[] {
return names.length > MAX_RETIRED_WORKTREE_NAMES_PER_REPO
? names.slice(names.length - MAX_RETIRED_WORKTREE_NAMES_PER_REPO)