test: git-remote coverage (4/4) — OPF pre-push, degraded remotes, shallow/partial clones, e2e#1639
Open
Soph wants to merge 7 commits into
Open
test: git-remote coverage (4/4) — OPF pre-push, degraded remotes, shallow/partial clones, e2e#1639Soph wants to merge 7 commits into
Soph wants to merge 7 commits into
Conversation
Cover the OPF (OpenAI Privacy Filter) pre-push rewrite at the integration and strategy layers, exercising what the existing in-process unit tests structurally cannot: OPF enabled via committed settings, resolved and executed inside a spawned pre-push hook, redacting real committed checkpoints and pushing the Entire-OPF-Applied trailer to a real bare remote. - E1: happy path via the real git-invoked hook — remote v1 tip carries the trailer, the fake opf binary is invoked, checkpoint + branch land. - E2: diverged v1 -> V1DivergedError aborts the user push (non-zero exit, actionable message); the branch tip on the remote does not advance. - E4: bootstrap cap exceeded -> typed abort before OPF runs; nothing pushed and the opf binary is never executed. - E5: non-TTY push (execx.NonInteractive, no controlling terminal) completes without hanging and still ships the trailer (regression 626a034). - E3: the CAS V1RefMovedError path is deterministically impossible to hit from a spawned hook, so it lands at the strategy layer via a fake OPF runtime that advances the local v1 ref mid-rewrite; a re-run succeeds. Integration tests point the OPF `command` at a fake opf script on disk (the in-process runtime seam is unavailable across the process boundary). OPF is git-branch only, so all E-group tests stay on the default backend. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: 950ada63c568
- F1 (git-branch): a bare-remote pre-receive hook rejecting the checkpoint v1 branch with a GH013-style message drives the protected-ref path: the hook prints its loud one-shot banner and returns, the user's own push still lands, and a repeat push does not retry-loop the rejected branch onto the remote (regression #1033). - F2 (git-refs slice): pushCheckpointRefWithRecovery wraps its push, fetch+replay, and retry in the same shared checkpointPushBudget as the v1 path, so a stuck transport bounds total wall clock to ~one budget rather than stacking a full budget per attempt. Mirrors the existing git-branch TestDoPushRef_SharedBudget_BoundsTotalWallClock (regressions #1282, 2e2c1b7). - F4 (git-refs): the git-refs pre-push path honors the checkpoint policy — a local policy this CLI cannot satisfy skips the per-checkpoint ref push (leaving refs queued) but does NOT abort the user push (regression 7bbdad0 follow-up). - F5 (git-branch): a session that checkpoints while HEAD is detached, pushed via `git push origin HEAD:<branch>` through the real hook, syncs and is readable from a fresh clone. Pins current (working) behavior. F6 (entire:// provider-host routing) is deferred: it needs an injectable provider host table / fake provider mapping that does not yet exist. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: ca5c18be1b8a
Cover the two clone shapes a user can arrive with, across both checkpoint backends: - C5: a `git clone --depth=1` user clone runs the full session -> commit -> real-hook push -> read flow, and the checkpoint machinery never makes the repo shallower than git left it (no self-inflicted .git/shallow; regressions #1443/#1276). Reading the checkpoint tolerates the shallow boundary instead of corrupting it. - C6: a `git clone --filter=blob:none` user clone reading a remote-only checkpoint lazily fetches the omitted metadata blobs, and the CLI stamps no new promisor/partialclonefilter config onto [remote "origin"] (regression #934). The .git/config guard enforces this at cleanup; an inline snapshot gives a clearer failure. Refactors CloneFrom into cloneFromWithArgs so shallow/partial clone helpers can pass --depth/--filter over the smart file:// transport (required for those flags to be honored from a local bare). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: 4da47566b4a2
- G2: an agent session runs inside a linked `git worktree`, the user commits and pushes the worktree's branch with a plain `git push` (the real pre-push hook syncs checkpoints — no explicit PushCheckpointRefs), and a fresh clone `entire resume`s the session. Exercises the worktree shadow-branch namespace and the shared common-dir push queue end to end. - G3: `entire doctor` on a repo whose origin points at an unreachable (hermetic, missing local path) target completes without hanging or crashing and still reports its local session-health scan — pinning the degrade-to-local behavior. Both are vogon-only (deterministic, free, part of the CI canary matrix). G2 reuses the resume-from-clone prompt phrasing verbatim so vogon's regex prompt parser needs no changes. Adds entire.DoctorCtx, a context-bounded doctor runner that returns output+error instead of failing the test, so G3 can surface a hang as a deadline. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: 8f2fa44f0ab7
Groups A/C/D/E/F(1,2,4,5)/G delivered on this branch; B-group blocked on decision D-1, F6 deferred, two pinned behaviors await product decisions. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: 4f990d0112c0
Contributor
There was a problem hiding this comment.
Pull request overview
This PR completes the git-remote test-plan coverage by adding integration and E2E tests around OPF pre-push rewriting, degraded remote behaviors, and shallow/partial clone scenarios, plus a small E2E helper and a doc status update.
Changes:
- Add integration tests for OPF pre-push rewrite (real hook + fake OPF binary seam), degraded remote cases, and shallow/partial user clones.
- Add E2E scenarios for worktree-originated sessions resuming from a fresh clone, and
entire doctorbehavior with an unreachable remote. - Extend integration test harness cloning to support shallow/partial clones; add a git-refs push-budget regression test and document test-plan status.
Reviewed changes
Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| e2e/tests/worktree_remote_test.go | Adds E2E coverage for worktree session → push → clone → resume, and doctor against an unreachable origin. |
| e2e/entire/entire.go | Adds DoctorCtx helper to run doctor under a caller-supplied context timeout. |
| docs/testing/git-remote-test-plan.md | Marks the plan as implemented on this branch and records remaining open items/pins. |
| cmd/entire/cli/strategy/push_common_budget_unix_test.go | Adds shared-budget wall-clock bound test for git-refs checkpoint push recovery path. |
| cmd/entire/cli/strategy/manual_commit_opf_rewrite_test.go | Adds deterministic ref-move seam to cover V1RefMovedError + successful retry. |
| cmd/entire/cli/integration_test/testenv.go | Extends clone helper to accept extra clone args (used for shallow/partial clone test setup). |
| cmd/entire/cli/integration_test/shallow_partial_clone_test.go | Adds C5/C6 integration coverage for --depth=1 and --filter=blob:none clones (both backends). |
| cmd/entire/cli/integration_test/opf_prepush_test.go | Adds E-group integration tests validating OPF pre-push rewrite via real hook subprocess. |
| cmd/entire/cli/integration_test/degraded_remotes_test.go | Adds F-group integration tests for protected v1 branch, blocked policy behavior, and detached-HEAD push/resume. |
Comment on lines
1880
to
1884
| cloneArgs := []string{"clone"} | ||
| cloneArgs = append(cloneArgs, extraArgs...) | ||
| if currentBranch != "" { | ||
| cloneArgs = append(cloneArgs, "--branch", currentBranch) | ||
| } |
Collaborator
Author
There was a problem hiding this comment.
Fixed in 2182fe0 — -c key=val pairs are hoisted before the clone subcommand. Note git clone -c is actually valid here (--config applies before the remote history is fetched, which is why these tests passed), but the hoist is still the right call: it avoids persisting the setting into the clone's .git/config and matches the git -c ... fetch pattern used elsewhere in the suite.
Global -c applies to the transport for the invocation only; `git clone -c` would also persist the setting into the clone's .git/config. Matches the `git -c protocol.file.allow=always fetch` pattern used elsewhere. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq Entire-Checkpoint: 9d3594469d76
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
https://entire.io/gh/entireio/cli/trails/760
Part 4 (final) of the stack on #1638. First integration-level coverage for the OPF pre-push rewrite, degraded-remote/protocol edges, shallow and partial user clones, and the two remaining e2e scenarios.
Tests
Entire-OPF-Applied: truewith the OPF binary invocation proven via a fakeopfscript seam (the in-process runtime seam can't cross the hook process boundary; content-level redaction stays unit-covered); diverged v1 →V1DivergedErroraborts the user push; CASV1RefMovedError+ successful re-run (strategy layer — the timing seam isn't deterministically reachable from a spawned hook); bootstrap cap aborts before OPF runs; non-TTY push completes without /dev/tty (regression 626a034).checkpointPushBudgethas no env seam); blocked checkpoint policy under git-refs skips the ref push, keeps refs queued, doesn't abort the user push (F4); detached-HEAD session +HEAD:branchpush + read-from-clone all work — no bug found (F5). F6 (entire://provider routing) deferred: needs an injectable provider host table.--depth=1clone — full session/commit/push/read without self-inflicted shallow grafts (regressions git_operations: stop shallow-fetching the metadata tip (root-cause fix for shallow-metadata false-disconnect) #1443/Prevent pack file race condition during checkpoint sync #1276);--filter=blob:noneclone — lazy blob fetch on read with no promisor config stamped onto[remote "origin"](regression Fetch checkpoint refs by URL to avoid polluting origin config #934).Also marks the test plan's implementation status in
docs/testing/git-remote-test-plan.md. Remaining open items: B-group (blocked on #1634), F6, and the pinned decisions (#1635, the unmerged >1000-commit replay cap fix).Verification
mise run fmt && mise run lintclean per commitmise run test:integrationgreen (457 tests at tip, 0 skips)go test ./cmd/entire/cli/strategy/greenmise run test:e2e:canarygreen under bothE2E_CHECKPOINT_STOREvalues (61 vogon + 4 external-agent)🤖 Generated with Claude Code
https://claude.ai/code/session_012yi3hHGAGepwfrfjPETjGq
Note
Low Risk
Test-only changes and documentation; no modifications to production push, OPF, or remote logic.
Overview
Adds the final slice of the git-remote test plan: integration and e2e tests only (no production behavior changes), plus doc status updates.
OPF pre-push (git-branch, real hook): New
opf_prepush_test.godrives OpenAI Privacy Filter through committed settings and a fakeopfscript in the spawned hook subprocess—happy path (Entire-OPF-Appliedon remote v1), diverged v1 aborting the user push, bootstrap cap before OPF runs, and non-TTY completion. E3 (V1RefMovedError+ retry) lives inmanual_commit_opf_rewrite_test.goviafakeOPFMovesRefOncebecause the race isn’t deterministic across a hook subprocess.Degraded remotes (F-group):
degraded_remotes_test.gocovers GH013-style protected v1 (banner, no retry, user push OK), git-refs blocked checkpoint policy (skip ref push, keep queue), and detached-HEAD checkpoint push/read from clone.Shallow/partial clones (both backends):
shallow_partial_clone_test.goplusTestEnv.cloneFromWithArgsfor--depth=1and--filter=blob:noneflows—checkpoint sync without worsening shallow state; lazy blob fetch without extra promisor config onorigin.Strategy: F2 shared push budget for
pushCheckpointRefWithRecovery(git-refs analogue of the v1 budget test).E2E: Worktree session → plain push → clone → resume (G2);
DoctorCtx+ unreachableoriginwithout hang (G3).docs/testing/git-remote-test-plan.mdis marked implemented with remaining open items (B-group, F6, pinned decisions).Reviewed by Cursor Bugbot for commit 2f04953. Configure here.