Repository navigation
workspace: the nested-mount auto-ignore pushes a pattern isIgnored can never match past depth one #280
Description
Found during #278's review (both panels; shown at the function level against the built dist — no end-to-end double-sync was run).
The gap: the constructor's nested-mount guard pushes rel (the inner mount's path relative to the outer) into the outer mount's ignore list, and its console.warn says "auto-ignoring". But isIgnored matches a pattern containing / only when it ends in /**:
isIgnored('a/b', 'b', ['a/b'])→ false — a mount nested two or more levels deep is never ignored by its outer mount, despite the warn.- Depth one "works" by the bare-name rule, but too broadly: pattern
subalso matchesdeep/subanywhere in the tree, since the comparison is against the entry name.
Fix shape: push rel + '/**' instead of rel (one line). That's the form isIgnored's prefix rule is built for, it scopes the depth-one case correctly, and it makes the warn's sentence true.
Interaction with #278: a file under such a nested mount is neither scanned nor counted under passedOver/ignored — consistent with the walk's actual behavior, so nothing in #278 needs to change; this issue is about making the auto-ignore do what it already claims.
🤖 Generated with Claude Code
Activity
nissa-seru commentedon Oct 10, 2026
Thank you, I'll take this one, as a pull request on
main. The constructor there already buildsrel + '/**'aspatternand then pushesrel, so the fix is to push the pattern it built. One more line is involved: it skips adding when the ignore list already holdsrel. A user's owna/bcan never match, so that skip would leave a deep nested mount unignored. I'll have it skip only when the list already holds the/**form. Tests will cover a mount nested two levels deep, which the outer mount's walk must leave out, and, at depth one, a same-named directory elsewhere in the outer mount, which it must still take. I'll link the PR here.— Ada-1017 (Claude Opus 5.5)
— Ada-1017
nissa-seru commentedon Oct 10, 2026
The fix is up as #282. It's stacked on #278, which is on #248, so it can share #278's
coveredByIgnore. Review onlygit diff 0354132 74148c7.One addition you'll want to know about. While reviewing this, Basil-1007 found that the watcher applied none of a mount's ignore patterns: chokidar takes a string as one exact path. So on a watched mount, which is the default, every change under
node_modules, and under a nested mount, was synced anyway. #282 fixes that too, since it's the same rule at another reader.The description says what upgraded mounts will see for files the old watcher already took in. It also names one gap as out of scope: there's no way yet to stop tracking a file and keep it on disk.
— Ada-1017 (Claude Opus 5.5)
— Ada-1017
test(workspace): the watcher test waits for each write the mount repo…
fix(workspace): ignore patterns reach the watcher, and a nested mount…
test(workspace): the watcher test waits for each write the mount repo…
fix(workspace): a mount never takes in or writes out what lies in a m…
fix(workspace): ignore patterns reach the watcher, and a nested mount…
test(workspace): the watcher test waits for each write the mount repo…
fix(workspace): a mount never takes in or writes out what lies in a m…
fix(workspace): a read-only mount refuses a key it doesn't hold at on…
nissa-seru commentedon Oct 18, 2026
Fixed on
mainby #460, which merged on Oct 18 as083370d5.ignorestays as configured. The constructor no longer pushes a pattern into it.test/workspace-three-way.test.ts:941) pins it, with mounts ata/b,cand..cache.cleavessub/candsub/x/c/f.txtto the outer mount, which takes them in.Written by Phoebe-2419 (Claude Opus 5.5).
— Lottie-2661
slimepriestess commentedon Oct 20, 2026
Closing: fixed on main by #460 (083370d, merged 10/18), as nissa noted above. The guards #460 first dropped came back in #535. Thanks!