Repository navigation
With diff.ignoreSubmodules=all in the user's config, Status() reports clean and Diff()/Patch() report nothing despite submodule commits and untracked files #190
Description
Activity
Triage
- Category: Bug
- Priority: High. A host-level
diff.ignoreSubmodules=allmakesStatus()reportIsClean == truewhile unpushed submodule commits and untracked files exist. A caller that asks "is this safe to discard?" gets the wrong answer with no warning, which can lead to lost work. This is the same class of host-config leak already closed forstatus.showUntrackedFiles(Status() can silently omit untracked files depending on the host's status.showUntrackedFiles config #112). - Area / suggested assignment:
GitStatusBuilder,GitDiffBuilder,GitPatchBuilder. The fix is to pin--ignore-submodules=noneby default, in the same way the other host-config overrides are pinned. - Duplicates / in progress: No duplicate. The issue already distinguishes this from With
diff.submodule=logor=diffin the user's config, Patch() silently drops submodule changes or reports the submodule's inner files as superproject files #124 (closed,diff.submoduleformat) and Patch() reports a submodule with only uncommitted changes as a "Subproject commit …-dirty" hunk that Apply().ToIndex() stages with exit 0 while changing nothing #173 (dirty-submodule hunks). No open PR covers it. The open GitIntegration PRs are Report conflicted binary files and modify/delete conflicts as unmerged in Patch() #192, Write a partial rename patch as a plain change to the new path #193 and Clear inherited GIT_DIR, GIT_INDEX_FILE and GIT_DIFF_OPTS before running git #184, and they address Patch() reports a conflicted binary file as an ordinary modification and omits modify/delete conflicts entirely, disagreeing with Diff() #123, Reverse-applying one hunk of a staged rename (PatchFor + Apply().ToIndex().Reversed()) undoes the whole rename #126 and Inherited GIT_DIR / GIT_INDEX_FILE / GIT_DIFF_OPTS override-Cand-U: inside a git hook every verb works on the wrong repository, and Patch() can return zero-context hunks #139.
Generated by Claude Code
What's wrong
GitStatusBuilder(~L84-99),GitDiffBuilder(~L160) andGitPatchBuilder(~L167-188) never pin--ignore-submodules, so a host-leveldiff.ignoreSubmodulessetting changes what they report. No--ignore-submodulesappears anywhere underGitIntegration/.Failure scenario (reproduced on git 2.43)
Submodule
shas a local commit plus an untracked file:status --porcelain=v2→1 .M SC.U ... s-c diff.ignoreSubmodules=all→ nothing, soIsClean == truediff --name-status:M s→ empty;diff --submodule=shortpatch → empty--ignore-submodules=nonerestores the output in all three.Why it matters
A caller asking "is this working copy safe to discard?" gets "clean" on a machine with that setting while unpushed submodule work exists. It is the same class of host-config hazard the repo already closed for
status.showUntrackedFiles(CLAUDE.md design point 14, #112).Suggested fix
Emit
--ignore-submodules=noneby default in Status, Diff and Patch (optionally with a builder option to choose another level).Acceptance criteria
diff.ignoreSubmodules=allset in the repo config and still see the dirty submodule.Not #124 (closed,
diff.submoduleformat) or #173 (dirty-submodule patch hunks).