test: git-remote coverage (3/4) — divergence & recovery matrix by Soph · Pull Request #1638 · entireio/cli

Soph · 2026-07-04T21:26:57Z

https://entire.io/gh/entireio/cli/trails/759

Part 3 of the stack on #1637. Systematizes the historically most re-broken area (#953, #1251, #1252, #1260: destructive local-v1-ref moves) into an explicit matrix, and covers the git-refs equivalents that only had unit coverage.

Tests

D1 (git-branch): table-driven matrix — local v1 {ahead, behind, diverged, disconnected} × {pre-push (real hook), explain fetch-on-miss, doctor}: local-only commits always survive, behind→FF, diverged→both sides preserved with linear tip (double-replay guard), repeated pre-push leaves the remote byte-identical. Cells already covered by safely_advance_local_ref_test.go / explain_test.go are not duplicated; the resume trigger shares the SafelyAdvanceLocalRef engine exercised via explain.
D4 (replay fidelity): no-op commit replay must not clobber the remote tip's tree (743c43f had no test); root-commit replay; >1000-commit replay pinned — the historical cap-removal fix (4cf01ed) was never merged (it's stranded on origin/no-limit), so TestCollectCommitsSince_CommitCapPinsCurrentBehavior pins the hard failure with a flip-when-fixed note.
D5 (TOCTOU): reconcile uses the fetched remote hash, not a stale pre-fetch observation (1e8628a had no test) — asserted as a behavioral invariant against a bare remote that advanced mid-flight.
D2/D3 (git-refs): remote-ahead per-checkpoint ref → real-hook push does fetch+replay+non-force retry with both sides surviving; rejected refs stay queued with the remote untouched (via a bare-side pre-receive rejection — a content-level cherry-pick conflict isn't deterministically reproducible at this layer); concurrent pushers from two clones.
D6 (multi-worktree): two worktrees enqueue into the shared common-dir push queue; a push from worktree A also lands worktree B's refs, queue compacted. New AddWorktree test helper.
D7 (two-remote pin): push to remote A drains the queue so remote B never receives the refs — pins current behavior, tracked as git-refs: push queue is drained by the first successful push — checkpoints never reach a second remote #1635.

Verification

mise run fmt && mise run lint clean per commit
mise run test:integration green (444 tests at tip)
go test ./cmd/entire/cli/strategy/ green
mise run test:e2e:canary green under both E2E_CHECKPOINT_STORE values

🤖 Generated with Claude Code

https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq

Note

Low Risk
Test-only changes and integration helpers; no runtime checkpoint sync or push logic is modified.

Overview
Adds integration and strategy tests that map local v1 / per-checkpoint ref divergence to real CLI paths (pre-push, checkpoint explain, entire doctor) and git-refs push-queue behavior, without changing production code.

D1 (git-branch): Table-driven cases for local v1 ahead, behind, diverged, disconnected through the pre-push hook (checkpoint survival, diverged replay keeps both sides with a linear tip, idempotent double pre-push). Separate tests cover fetch-on-miss via checkpoint explain (behind/diverged) and doctor --force on disconnected orphans.

D2/D3 (git-refs): End-to-end pre-push when a per-checkpoint ref is amended on the remote vs locally (fetch+replay, both file sets kept), server rejection (ref stays queued, remote unchanged, retry succeeds), and two clones racing on the same ref.

D6/D7 (queue): Shared common-dir queue across linked worktrees (one pre-push drains both, stale entries pruned). D7 intentionally asserts today’s two-remote gap: first push drains the queue so a second remote never gets the refs (#1635).

Test harness: commitMsgPath() for linked worktrees; new AddWorktree helper.

Strategy (replay_fidelity_test.go): Guards for no-op commits not clobbering remote trees on replay, root-commit chains, reconcile using the post-fetch remote tip; pins the >1000-commit collectCommitsSince cap until the unmerged cap-removal lands.

^{Reviewed by Cursor Bugbot for commit d93c1e3. Configure here.}

…D4, D5) Systematize the local-v1 divergence & recovery matrix (the most re-broken area, #953/#1251/#1252/#1260) plus the specifically-untested replay edges. D1 (integration, git-branch): drive the local-v1 state {ahead, behind, diverged, disconnected} through the real triggers end-to-end: - pre-push (real hook): local-only survives; ahead fast-forwards; diverged replays both sides on a linear tip; disconnected cherry-picks; a repeated push is idempotent (double-replay guard, #1260). - explain fetch-on-miss for {behind, diverged}, extending the existing {missing, ahead} coverage: a remote-only checkpoint is fetched/reconciled and local-only checkpoints survive. - doctor --force reconciles a disconnected v1, preserving both sides. D4 (strategy unit): replay fidelity edges. A no-op commit must not clobber the remote tip's accumulated tree (regression 743c43f); a root-commit chain replays cleanly. Plus a KNOWN BUG pin: >1000-commit replay is still capped (collectCommitsSince) because the cap-removal (4cf01ed) lives on the unmerged branch origin/no-limit; flip the assertion when it lands. D5 (strategy): TOCTOU guard for 1e8628a -- fetchAndRebaseRefCommon replays onto the freshly fetched remote tip, not a stale earlier hash. The literal ls-remote seam the fix patched no longer exists (fix is an ancestor), so this pins the behavioural invariant. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: b19f4c2ff3bd

…ery (D2, D3) Drive the git-refs recovery end-to-end through the real pre-push path (drain queue -> batch push -> per-ref fetch+replay), complementing the unit-level coverage in strategy/refs_push_test.go. D2: - remote ref amended by another clone (valid child commit) + a divergent local write of the same ref: the fast-forward-only batch push is rejected, recovery fetches the remote tip and replays the local change on top (non-force), so both sides' files survive and the remote commit stays an ancestor. - queue safety: when the remote refuses the ref update (bare pre-receive declining the checkpoint namespace), the ref stays queued and the remote is untouched (never force-overwritten); a later push retries and lands it. A pure content-level cherry-pick conflict is not reproducible at this layer (tree-delta apply is last-writer), so a server-side rejection stands in. D3: two clones race on the SAME per-checkpoint ref from a shared base -- the second pusher fetch+replay+retries (non-force), preserving both. The git-refs analogue of TestConcurrentPush_SecondPusherRebasesAndRetries (v1-only). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: faa45ac4455f

…ee, D7 two-remote pin) D6: the git-refs push queue lives in the shared git common dir, so sessions in two worktrees enqueue into ONE queue. A single pre-push from the main worktree pushes both worktrees' checkpoint refs, drains + compacts the queue, and prunes a stale entry (a queued ref with no local ref -> partitionLocalRefs). Adds an AddWorktree test helper (its own .entire/, shared common dir, offset session counter so session IDs don't collide in the shared session store). D7: PIN the known two-remote queue-clearing gap (test plan decision D-2). The queue records only the ref, not the remote, so a push to remote A drains the queue and remote B never receives those refs. Asserts current behavior with a clear "flip when per-remote tracking lands" note. No production change. Also resolves the COMMIT_EDITMSG scratch path via `git rev-parse --absolute-git-dir` so the shadow-commit test helpers work in a linked worktree (where .git is a file); identical to the prior path for normal checkouts. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: 057f6136f6f5

Copilot

Pull request overview

Adds a broader, end-to-end integration-test matrix around historically fragile checkpoint ref divergence/recovery behavior (v1 git-branch and git-refs per-checkpoint refs), plus strategy-level replay-fidelity regression guards. This strengthens confidence that ref reconciliation is correct and remains idempotent across pre-push, explain fetch-on-miss, and doctor flows—without changing production behavior.

Changes:

Add strategy tests that lock replay fidelity invariants (no-op commits, root chains, fetched-remote-hash TOCTOU) and pin the current >1000-commit traversal cap behavior.
Add integration tests covering (a) v1 divergence state × trigger matrix, (b) git-refs per-checkpoint ref divergence/rejection/concurrency recovery, and (c) push-queue semantics (multi-worktree shared queue + pinned two-remote gap).
Extend the integration test harness with linked-worktree support (AddWorktree) and correct commit message scratch resolution in linked worktrees.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
cmd/entire/cli/strategy/replay_fidelity_test.go	New strategy-level regression tests for replay correctness and TOCTOU invariants; pins current commit traversal cap behavior.
cmd/entire/cli/integration_test/testenv.go	Adds linked-worktree helpers and fixes COMMIT_EDITMSG resolution to work under `git worktree`.
cmd/entire/cli/integration_test/refs_queue_semantics_test.go	New integration coverage for git-refs push-queue behavior across multi-worktree and multi-remote scenarios (including pinned known bug).
cmd/entire/cli/integration_test/refs_divergence_test.go	New integration coverage for git-refs per-checkpoint ref divergence recovery, server-side rejection retry semantics, and concurrent pushers.
cmd/entire/cli/integration_test/divergence_matrix_test.go	New integration matrix for v1 local/remote divergence states across pre-push, explain fetch-on-miss, and doctor reconciliation triggers.

Soph · 2026-07-05T08:16:27Z

+		c := exec.CommandContext(env.T.Context(), "git", args...)
+		c.Dir = env.RepoDir
+		c.Env = append(testutil.GitIsolatedEnv(), "GIT_INDEX_FILE="+env.RepoDir+"/.git/entire-orphan-index")
+		if stdin != "" {


Fixed in 56408fa — GIT_INDEX_FILE path now built with filepath.Join.

…3-divergence

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: 4521fd749232

Soph and others added 3 commits July 4, 2026 21:53

Copilot AI review requested due to automatic review settings July 4, 2026 21:26

Soph requested a review from a team as a code owner July 4, 2026 21:26

Soph mentioned this pull request Jul 4, 2026

test: git-remote coverage (4/4) — OPF pre-push, degraded remotes, shallow/partial clones, e2e #1639

Open

Copilot started reviewing on behalf of Soph July 4, 2026 21:27 View session

Copilot AI reviewed Jul 4, 2026

View reviewed changes

Soph and others added 2 commits July 5, 2026 10:12

Merge branch 'soph/git-remote-tests-2-p0' into soph/git-remote-tests-…

80066a1

…3-divergence

test: use filepath.Join for GIT_INDEX_FILE path (Copilot review)

56408fa

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: 4521fd749232

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

test: git-remote coverage (3/4) — divergence & recovery matrix#1638

test: git-remote coverage (3/4) — divergence & recovery matrix#1638
Soph wants to merge 5 commits into
soph/git-remote-tests-2-p0from
soph/git-remote-tests-3-divergence

Soph commented Jul 4, 2026 •

edited by cursor Bot

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Soph Jul 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

Soph commented Jul 4, 2026 • edited by cursor Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Tests

Verification

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Soph Jul 5, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

2 participants

Soph commented Jul 4, 2026 •

edited by cursor Bot

Loading