fix(init): back-fill missing hop.json on idempotent re-init

[jadb]

Summary

• git hop init was a no-op when a bare-worktree-shaped repo had the structure but was missing hop.json (legacy bare repos cloned outside git hop, or repos whose hop.json was lost). The command printed "Repository already initialized" and exited, leaving the directory un-registered as a hub.
• handleAlreadyInitializedWithFlags now back-fills a missing hop.json from runtime git state before printing the idempotent summary, so re-running git hop init becomes a true recovery path.
• Inputs gathered: origin URL via git remote get-url origin (best-effort, with <parent>/<dir> fallback matching registerAsIs); default branch via git symbolic-ref HEAD (falls back to main); branches by parsing git worktree list --porcelain (bare/detached entries skipped; short branch name used as the map key).
• For the worktree-child case the hub root is derived from <cwd>/.git's gitdir: pointer rather than hop.FindProjectRoot, because the latter aborts when its DetectRepoStructure walk hits an intermediate hops/ directory that returns NotGit.

Companion PR

• fix(status): diagnose bare-worktree repo missing hop.json #26 (fix(status): diagnose bare-worktree repo missing hop.json) — pairs with this change: status now points users at git hop init, which then heals the repo.

Test plan

• go build ./... clean.
• go test ./internal/... ./cmd/... — 865 pass, 0 fail (8 new test cases across cmd/init_backfill_test.go, all written test-first: parseWorktreeListPorcelain parser shape, backfillHubConfigIfMissing happy-path / no-op / no-remote fallback / no-default-branch fallback / error propagation, and resolveBackfillRoot for BareWorktreeRoot / WorktreeRoot / WorktreeChild / declined-on-StandardRepo).
• go vet ./cmd/ clean.
• Manual end-to-end on a legacy bare-worktree repo (10 worktrees, mixed branch paths including a path-with-suffix refactor/kit-12fcc-baseline-T-0119-20 whose branch name has no suffix): removed hop.json, ran git hop init from the bare-repo root → prints Created missing hop.json at <path>. and writes a valid config; subsequent git hop status shows the full hub table. Repeated from inside hops/main/ (worktree-child) → also resolves the correct hub root via the gitdir pointer and writes hop.json there.

Notes

• envPatterns defaults to ["dev","staging","qa"] so the back-filled hub matches CreateHub's defaults.
• Write failures emit a warning: line to stderr but do not abort init — hooks installation and the "Repository already initialized" summary still run.
• Existing TestIdempotentRun_* tests previously passed g=nil; updated to pass mocks.NewMockGit() since the function now touches git state on first invocation.

[Copilot]

Pull request overview

This PR makes git hop init a recovery path for legacy/partially-initialized bare-worktree repos by back-filling a missing hop.json from runtime git state before printing the usual idempotent “already initialized” summary.

Changes:

• Add backfill logic to reconstruct and write hop.json using git remote get-url origin, git symbolic-ref HEAD, and git worktree list --porcelain.
• Invoke the backfill step from handleAlreadyInitializedWithFlags (warning-only on failure).
• Add unit tests for the porcelain parser, backfill behavior, and hub-root resolution; update existing init flag tests to use a non-nil mock git interface.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
cmd/init.go	Calls the new backfill step during idempotent init runs to heal missing hop.json.
cmd/init_flags_test.go	Updates idempotent-init tests to pass a mock git interface (now required).
cmd/init_backfill.go	New helper implementation for parsing worktree porcelain, deriving hub root, and writing hop.json.
cmd/init_backfill_test.go	New tests covering parser behavior, backfill behavior, and hub-root resolution.

Comments suppressed due to low confidence (2)

cmd/init_backfill_test.go:247

• The comment above TestResolveBackfillRoot says the WorktreeChild case "walks up via FindProjectRoot", but resolveBackfillRoot no longer calls FindProjectRoot—it parses the worktree's .git gitdir pointer via hubFromWorktreeChild. Please update this comment to reflect the current behavior to avoid misleading future changes.

// resolveBackfillRoot picks the path to back-fill given the cwd and the
// repo structure detected there. BareWorktreeRoot → cwd; WorktreeChild
// → walk up via FindProjectRoot. Anything else → ("", false).
func TestResolveBackfillRoot(t *testing.T) {

cmd/init_backfill_test.go:283

• The WorktreeChild test data uses a gitdir pointer of "/worktrees/", but elsewhere in the repo the worktree-child shape is typically "/.git/worktrees/" (e.g. cmd/init_idempotent_test.go). Add coverage for the ".git/worktrees" form (and ensure resolveBackfillRoot returns the hub root, not the .git directory) so the backfill logic matches real repos.

		// A worktree child under <tmp>/repo/hops/main.
		hopsMain := filepath.Join(repo, "hops", "main")
		if err := fs.MkdirAll(hopsMain, 0755); err != nil {
			t.Fatal(err)
		}
		if err := afero.WriteFile(fs, filepath.Join(hopsMain, ".git"), []byte("gitdir: "+filepath.Join(repo, "worktrees", "main")+"\n"), 0644); err != nil {
			t.Fatal(err)
		}

		got, ok := resolveBackfillRoot(fs, hopsMain, config.WorktreeChild)
		if !ok || got != repo {
			t.Errorf("got (%q, %v), want (%q, true)", got, ok, repo)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[jadb]

Fixed in bab9f56. hubFromWorktreeChild now strips an optional trailing ".git" segment after locating the "worktrees" parent, so both bare-hub ("/worktrees/") and regular-hub ("/.git/worktrees/") shapes resolve to the hub root. Added a test for the regular-hub shape plus a malformed-gitdir negative case.

Skipped the relative-gitdir-path suggestion: git emits absolute paths by default, and the codebase has no precedent for handling --relative here. Happy to add it if you can point to a repro.

[jadb]

Fixed in bab9f56. Removed the dead var _ = hop.ParseRepoFromURL line and the now-unused hop import.

The "also appears at line 244, 272" claim looks hallucinated — there was only the one occurrence in this file.