[alexander-yevsyukov]

Recovers a commit that #804 merged without.

#804 took a Codex review fix as 6a39f556, but config's auto-merge fired
about two minutes before that commit was pushed: the merge commit 6356d50e
has branch-side parent 65cd8609, whose own parent f1423f5e sits one commit
below the fix. The PR reads "merged" with green checks, and master still ships
the exact bug the review reported. This is the same failure mode that cost
#708 a fix once before.

6a39f556 is cherry-picked here rather than retyped, so its message and
authorship survive.

What it fixes

pull resolves a branch = . submodule declaration — gitmodules(5)'s sentinel
for "the submodule's branch carries the same name as the superproject's current
branch" — to the superproject's branch name. It did so with
git rev-parse --abbrev-ref HEAD, which shortens a ref only as far as a
non-ambiguous name. A superproject on branch foo that also carries a tag
foo therefore resolved to heads/foo: the fetch of heads/foo succeeds, the
following git checkout fails, and the submodule is left stale at the pin while
the warning names origin/heads/foo.

git branch --show-current prints the name exactly, and prints nothing when
HEAD is detached — so it fixes the ambiguity and keeps signalling detachment.
git symbolic-ref -q --short HEAD is not an alternative: it applies the
same shortening and returns heads/foo too.

The [ "$branch" = "HEAD" ] half of the detachment guard goes with the old
command, since --show-current never prints that. Review confirmed this is
unreachable-code removal, not a narrowing: a branch literally named HEAD
cannot be created (fatal: 'HEAD' is not a valid branch name), so [ -z "$branch" ] is a complete guard. --show-current needs git 2.22 (2019).

Nothing in Spine declares branch = . today — all 20 local checkouts with a
tracked branch use the literal master — so this is latent robustness, not a
live outage. master's init-submodules does not resolve . at all (it only
initializes), so there is no sibling copy of this bug to keep in step with here.

Verification

Throwaway superprojects under a temp dir, each with its own config submodule
carrying either the pre-fix or the fixed pull, so the real ./config/pull
runs end to end. Submodule pin A; remote tips M2/F2/T2 per branch.

case	resolved	submodule after
pre-fix, branch foo + tag foo	origin/heads/foo → error: pathspec 'heads/foo' did not match any file(s), float warning	stale at pin A
fixed, branch foo + tag foo	origin/foo	F2, on branch foo
fixed, plain branch topic	origin/topic	T2, on branch topic
fixed, detached superproject	— warns, continue	untouched at pin A
fixed, literal branch = master, run twice	origin/master	M2, on master, idempotent

bash -n passes. shellcheck pull reports only the pre-existing SC1091 info on
source migrate, identical to the origin/master baseline.

Notes for review

• 6a39f556 was pushed after the review that prompted it and so was never
  itself reviewed. This PR is its first pass; review-docs flagged that the new
  comment wrapped git rev-parse --abbrev-ref HEAD and git symbolic-ref --short across two lines each, breaking copy-paste and hiding the command
  from a grep of the script that warns about it. Fixed in a separate commit so
  the cherry-pick stays faithful to the original.
• .agents/tasks/pull-resolve-dot-branch.md is kept, not deleted. Its
  status is in-review, not done, and .agents/tasks/README.md ties
  deletion to the work having landed on master — half of it did not. Deleting
  it now would erase the only tracking record of work still in flight. It should
  be deleted when this PR merges.
• Out of scope, worth a follow-up: pull's pre-existing comment (added by
  #804) says init-submodules "initializes before it resolves ." and
  "floats only a submodule it has just cloned". Master's init-submodules does
  neither — it does not float and never resolves .. The comment anticipates
  #802's unmerged behaviour.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-19T14:59:50.432644Z	e62460d	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[Copilot]

🟢 Approved

The focused fix addresses ambiguous branch names without changing literal-branch or detached-HEAD behavior, and no blocking issues remain.

No open issues 🎉

What changed in this PR

Fixes branch-name resolution in pull when a superproject branch and tag share the same name.

Changes:

• Uses git branch --show-current to preserve the exact branch name.
• Removes the obsolete HEAD check while preserving detached-HEAD handling.
• Updates the task record with verification and recovery history.

File	Description
pull	Resolves exact branch names and retains the detached-state guard.
.agents/​tasks/​pull-resolve-dot-branch.md	Records verification and tracks landing the previously omitted fix.

_{Review effort: Balanced}