Triage and clean up an unreviewed automated commit stream on main

[julianknutsen]

Triage and clean up an unreviewed automated commit stream on main

Summary

Between 9a1c88b63 (the reviewed dolt-2.1.4 bump, PR #4326) and the current origin/main
tip 8e18581dc, an automated agent committed a large stream of changes directly onto the
shared branch without going through PR review. A commit-by-commit triage found that the
cited issue references could not be located in the tracker, so each commit's stated
justification could not be independently verified.

This PR does not discard the stream. The triage found that roughly half of it is
genuine, diff-provable bug-fixing, and that several commits are load-bearing on an
already-published Dolt schema — reverting them would brick deployed clones. So this PR:

1. Reverts the no-op / characterization commits that don't fix a demonstrable defect (4 commits).
2. Reverts the unreviewed changes in the gated, not-running serverV2 subsystem (the
   proxied / external-dolt-server layer) back to BASE — itemized in the serverV2 audit
   comment for the subsystem owner to review.
3. Keeps the real fixes, and — since the cited references can't serve as the source of
   truth — backs every kept real defect with a proven born-failing regression test.
4. Flags the remaining judgement calls for maintainer decision rather than silently
   keeping or reverting them.

Commits (6): 4 reverts of no-op commits + 1 commit adding proven regression tests + 1
serverV2 revert. Per-commit justification for the reverts and the flagged decisions is in the
Removal Rationale comment; the serverV2 fixes being removed are in the serverV2 audit
comment.

serverV2 revert (8515879dc)

The proxied / external-dolt-server subsystem is gated off and not running yet. This reverts
the unreviewed changes to its schema-independent layers — internal/storage/dbproxy/*,
internal/storage/uow/*, and the cmd/bd proxied glue — back to BASE (26 files, +107/-870).
A strict serverV2→BASE revert is impossible without breaking main: internal/storage/domain/db
is welded to published migration 0051 (BASE inserts without an id and relies on the
DEFAULT(UUID()) that 0051 removed), so the data layer + the issueops widening it needs are
kept at HEAD. The removed serverV2 fixes — most of them real, proven bugs in the gated
subsystem — are itemized for the owner to audit / re-apply.

The source-of-truth finding

All 66 cited issue IDs (bd-6dnrw.* ~47 items, bd-578h9.* ~19, plus bd-2rd37, bd-pkim8,
bd-hj85c, bd-4mpy7, …) are not present in the tracker. Verified: closed issues included
(bd list over all statuses); created-then-deleted ruled out (dolt_history_issues has zero
rows for any cited family across full history); not a parsing artifact (dotted epics like
bd-am3.1 exist); not on any sibling federated DB or remote beads-* branch; the IDs appear
only in commit messages.

Consequence: with no tracker entry to corroborate a commit's rationale, we treat the diff
and a born-failing test — never the cited reference — as the source of truth.

What this PR reverts (4 commits)

Non-schema, non-entangled, no demonstrated defect (reverted newest-first):

Reverted	Why
93b186dcf	Seam B list-parity characterization test for already-shipped behavior
1425e7338	Re-enables a CI lane the stream itself disabled + ungates a dark-launched feature; no shipped bug
81489b8c1	Autocommit-sweep over-firing "fixer" for a condition introduced earlier in the same stream
d2b6cc525	Speculative dirty-working-set autocommit safety net, no demonstrated defect

Reclassified revert → keep (3 commits, entanglement discovered while reverting)

Originally slated to revert but entangled with commits we keep, so they cannot be cleanly
reverted and none is harmful — left in place:

Commit	Blocked by (kept)
1825cf357 (tree-walk dotted orphans)	341c7a5a4 dolt-2.1.6 named-CTE workaround builds on it
9187152cb (doctor rekey-repair)	6a1e7af9b gate-hardening modifies its dep_keys.go
f97fd4a8b (merge-refusal messaging)	d80ed8146 no-remote fix shares the cmd/bd/dolt.go region

What this PR keeps (and why)

• Schema / migration stacks — keep & forward-fix only. Migration 0051 (dropped
  DEFAULT(UUID())) and the aux-row-id rekey have already executed on every published clone;
  0051.down.sql is a no-op and re-adding the default is banned by migration-hygiene CI. Reverting
  their source does not un-run them — it bricks the write path (NOT NULL on every insert). Any
  future change ships as a new forward migration.
• Sync settle/recompute subsystem — keep & re-land under review. Clones now rely on its
  auto-resolve / FK-repair on pull; the is_blocked-stale-after-merge defect it fixes is real at BASE.
• Middle-tier hardening — surgical. Diff-provable fixes and hardening of the bd dolt pull unmergeable after multi-machine upgrade — migration 0043 forks the dependencies primary key (conditional DDL + DEFAULT (UUID())) #4259 gate are kept
  (re-attributed through this review); only clear no-op commits are reverted.

Proven regression tests for every kept real defect

For all 45 kept real-defect commits, a regression test now lives in this PR and is proven
born-failing by mutation — reintroduce the specific defect into current code and the test goes
RED; restore the fix and it goes GREEN. 42 were verified against the tests the commits
already shipped; 3 were newly authored here because the shipped tests were characterization-only
or absent:

• internal/lockfile TestIsLocked — detect a wrapped ErrLocked (guards the errors.Is fix; orig 1a83cb08c)
• internal/storage/domain/db TestGetDescendantsFilteredByStatus — level-filter across edge + dotted branches (guards the dolt-2.1.6 named-CTE workaround; orig 341c7a5a4)
• cmd/bd TestGOMODCACHENotUnderTestHome — GOMODCACHE pinned outside the temp test HOME (orig bc909834c)

Full proof ledger (45 defects)

short	cluster	regression test	pkg	test name(s)	conf
33e71d21e	doctor	proven (existing)	doctor	TestShouldFlagTrackedFile_RuntimeArtifacts	high
928f9ff2d	doctor	proven (existing)	doctor	TestCheckMigrationContentSkew_UnexpectedErrorWarns	high
03cdc6c86	dolt-runtime	proven (existing)	dolt, issueops	TestRecomputeIsBlockedAfterMerge_PreservesUpdatedAt	high
0664ee7be	dolt-runtime	proven (existing)	dolt	TestInitSchemaOnDBWithRetryAndGate_GateErrorClassification	high
1f8331d16	dolt-runtime	proven (existing)	dolt, versioncontrolops	TestFKCascadeViolationRepairOnSQLPull; TestFKCascadeViolationRepairOnC	high
2d5f11e5a	dolt-runtime	proven (existing)	bd	TestSharedServerEmbeddedMismatchDoesNotRewriteMetadata	high
341c7a5a4	dolt-runtime	AUTHORED (new)	db	TestGetDescendantsFilteredByStatus	high
49a803050	dolt-runtime	proven (existing)	bd	TestBulkDepAddCycleGateRollsBack; TestNewCycleThroughEdges	high
6a1e7af9b	dolt-runtime	proven (existing)	dberrors	TestAncestorPKMismatch_CasingConsistency; TestAncestorPKMismatch_NonMa	high
8095e4f4e	dolt-runtime	proven (existing)	bd	TestIsConfirmedNoRemote	high
9264f0ec2	dolt-runtime	proven (existing)	issueops	TestCycleThroughEdgesInGraph	high
a768035af	dolt-runtime	proven (existing)	doltserver	TestDetectCorruptManifest_ReportsWithoutModifying	high
ccadfce9f	dolt-runtime	proven (existing)	dolt, versioncontrolops	TestTryAutoResolveMergeConflicts_SchemaMigrationsTheirHashWins; TestTr	high
d80ed8146	dolt-runtime	proven (existing)	bd	TestIsConfirmedNoRemote	high
1fe4993cf	import-export	proven (existing)	embeddeddolt	TestCreateIssuesRejectStaleUpserts	high
33bf6d0c0	import-export	proven (existing)	embeddeddolt	TestCreateIssuesRejectStaleUpserts/equal_timestamp_keeps_local_row; Te	high
1a83cb08c	proxied-server-uow	AUTHORED (new)	lockfile	TestIsLocked	high
27bbecbd1	proxied-server-uow	proven (existing)	db	TestDomainDB/TestParityWispInclusion; TestDomainDB/TestParityReadyBloc	high
29df600aa	proxied-server-uow	proven (existing)	uow	TestDoltServerTxRunnerAfterCommitErrorsInsteadOfPanicking	high
439fb4eb1	proxied-server-uow	proven (existing)	bd	TestEmbeddedInit/remote_behind_schema_gates_with_guidance; TestPrintBo	high
4b0509c11	proxied-server-uow	proven (existing)	uow	TestIsRetryableWarmupError	high
794ff0790	proxied-server-uow	proven (existing)	uow	TestDoltServerTxCommitFailureRollsBackBeforeRelease; TestDoltServerTxC	high
8866e5b2b	proxied-server-uow	proven (existing)	proxy	TestReadAndDial_DeadPidfileWriterRejected; TestSpawnAndHandoff_ReapsOr	high
914ac3c12	proxied-server-uow	proven (existing)	db	TestDomainDB/TestParityReadyBlockedDescendants	high
99c70904f	proxied-server-uow	proven (existing)	uow	TestNewDoltServerUOWProvider_RemoteMigrateGate_BlocksReopen	high
9f531717d	proxied-server-uow	proven (existing)	db, issueops	TestScanIssue_StringTimestamps; TestIssueSelectColumns_MatchClassic	high
b5cdb70ab	proxied-server-uow	proven (existing)	bd	TestCommandSupportsProxiedServer	high
f1c72dbde	proxied-server-uow	proven (existing)	doltutil	TestPersistedRemotes	high
f483ac2b4	proxied-server-uow	proven (existing)	bd	TestPrintBootstrapRemoteBehindGuidance; TestPrintBootstrapRemoteBehind	high
f880a985b	proxied-server-uow	proven (existing)	bd	TestNewProxiedServerUOWProvider_CorruptSidecarErrorsInsteadOfManagedFa	high
12a47948c	schema-migration	proven (existing)	schema, rowid	TestRekeyAuxRowTableKeepsHeldTargetsStable; TestRekeyAuxRowTableIdempo	high
1be8989fe	schema-migration	proven (existing)	schema	TestCheckRemoteMigrateGateWithRemoteCheck	high
8ffcf698e	schema-migration	proven (existing)	schema	TestCheckRemoteMigrateGate	high
969e27ead	schema-migration	proven (existing)	embeddeddolt	TestEmbeddedOpenForReadOnlyCommand_LenientGate	high
e4f040741	schema-migration	proven (existing)	schema	TestMigrationWorkNeededAddsContentHashColumnOnUpToDateDB; TestMigratio	high
3f2b4e08d	storage-sql	proven (existing)	embeddeddolt	TestEmbeddedOpenReadOnly_SkipsGateAndMigrations	high
8a92dacf4	storage-sql	proven (existing)	embeddeddolt	TestEmbeddedOpenReadOnly_SkipsGateAndMigrations	high
f3d31e444	storage-sql	proven (existing)	issueops	TestSearchIssuesWithCountsHonorsSkipWisps; TestSearchIssuesWithCountsP	high
001c6c258	sync-federation	proven (existing)	embeddeddolt, versioncontrolops	TestEmbeddedMergeAndSettleFKCascadeRepair; TestEmbeddedMergeAndSettleR	high
59ba57f02	sync-federation	proven (existing)	dolt, versioncontrolops	TestFKCascadeRepairWithAutoResolvableConflicts	high
7171f8f8e	sync-federation	proven (existing)	embeddeddolt, versioncontrolops	TestEmbeddedMergeAndSettleReportsOperatorConflicts	high
792260c0a	sync-federation	proven (existing)	tracker	TestEnginePullDoesNotTreatPreviousPullAsLocalConflict	high
99ed06b6f	sync-federation	proven (existing)	dolt, issueops	TestRecomputeIsBlockedAfterMerge_BlockerClosedRemotely; TestRecomputeI	high
c708e00c3	sync-federation	proven (existing)	dolt	TestPullFromSettlesFKCascadeViolations; TestPullFromReportsOperatorCon	high
bc909834c	test-only	AUTHORED (new)	bd	TestGOMODCACHENotUnderTestHome	high

Maintainer decisions (see the Removal Rationale comment for full per-commit analysis)

• 33bf6d0c0 — equal-timestamp upsert keeps the local row; changes merge-idempotency behavior (has tests). Decided: keep.
• f1c72dbde — reads persisted remotes from repo_state.json; fixes a real gate weakness but rests on an unverified dolt-2.x format assumption. Decided: keep (verify the assumption as a follow-up).
• 7ebf4df6a — reverses reviewed feature feat(linear): ambient staleness signal — bd prime auto-pull + --pull-if-stale #3669 (ambient Linear staleness). Decided: keep (loop in the feature owner).
• 3fa29e325 — removes a DOLT_GC call added in reviewed PR fix(fetch): run GC after failed DOLT_FETCH to prevent tmp_pack_* accumulation #3326. Decided: keep (confirm with the PR owner).
• 9a9da38c4 — maintainer-policy docs. Decided: keep.

Git mechanics

Cleanup is done as git revert on this branch — no rewrite of published history. HEAD is
published and clones have replayed migrations; a history rewrite cannot un-run them and would
create a worse source-vs-DB split-brain.

Verification

• make build / go build -tags gms_pure_go ./... — green
• go vet over touched packages — clean
• The 3 authored regression tests — pass (and proven RED under their mutations)
• Full make test on the serverV2 revert — green (60 ok packages, 0 failures); domain/db,
  uow, dbproxy all ok.
• Regression-free proof: any sandbox-only test flakiness observed (a cmd/bd package-level
  test-isolation issue with os.Setenv/os.Chdir) reproduces identically on untouched
  origin/main and on the BASE before the stream, so this branch introduces no new failures.
  Real CI (Main, Regression Tests) is green at the current tip.

Optional follow-up (not required): harden the cmd/bd suite by converting its unguarded
os.Setenv/os.Chdir to hermetic t.Setenv/t.Chdir.

[julianknutsen]

Removal Rationale — commit-by-commit

Per-commit justification for the reverts on this branch, plus the maintainer-decision commits.
The cited issue references are not present in the tracker, so the diff and a born-failing test
are treated as the only source of truth.

---

Part 1 — The four reverts on this branch (no demonstrable defect)

1. 93b186dcf — test(cmd/bd): Seam B cross-mode list parity → reverted by e409ab35e
A characterization test, not a fix. It asserts the already-shipped behavior of the
embedded-vs-proxied bd list paths (offset rejected in direct mode but honored proxied,
truncation rules). It passes against the existing code, so it neither reproduces nor guards a
defect — there is no "before" failure. Removing it loses no bug coverage.

2. 1425e7338 — test(ci): restore test-proxied-server-cmd lane behind a test-only init ungate → reverted by dac4fb49e
Re-enables a CI lane that was disabled earlier in the same stream and adds a test-only "init
ungate" so the gated (not-running) proxied-server can be exercised. No born-failing reproduction —
the test assertion was rewritten to match new expected behavior, not added as a failing repro.
CI/test scaffolding for the gated subsystem, not a fix.

3. 81489b8c1 — fix(cli): keep the autocommit sweep off read-only and inspection commands → reverted by 97f999a14
Self-referential: it fixes a condition inside the autocommit sweep that was introduced earlier
in this same stream by d2b6cc525. The over-firing it guards against does not exist in the
baseline. Reverting it together with d2b6cc525 removes the whole speculative mechanism and nets
to zero.

4. d2b6cc525 — fix(autocommit): dirty-working-set safety net behind commandDidWrite → reverted by 548d7308e
Speculative/defensive, no demonstrated defect. No actual write path is identified that forgets
commandDidWrite; the change defends against a scenario never shown to occur. No reproduction, no
born-failing test. (Its companion 81489b8c1 then fixed a condition this very mechanism
introduced — see #3.)

---

Part 2 — Flagged maintainer decisions (all resolved: KEEP)

Honest framing: not all of these are "not needed." Two actually fix real weaknesses and carry
tests — they are kept.

33bf6d0c0 — fix(import): equal-timestamp upsert keeps the local row — KEEP
Carries born-from-this-commit regression tests (equal_timestamp_keeps_local_row,
equal_timestamp_empty_notes_does_not_wipe_local_notes, stale_incoming_does_not_reopen_closed_issue)
that assert real DB state and demonstrate a concrete clobber under the old >= comparison. It does
change equal-timestamp merge-idempotency behavior, so it is a real policy choice — kept as intended.

f1c72dbde — fix(dolt): read persisted remotes from repo_state.json — KEEP (verify assumption)
The pre-image (the legit #4259/#4268 work, 1be8989fe) genuinely has the described weakness:
ListCLIRemotes shells out to dolt remote -v and the migrate-gate swallows any subprocess error,
so a transient failure reads as "no remote" and silently bypasses the gate. This commit reads
repo_state.json directly instead. Its live fix (dolt/federation.go + doltutil/remotes.go) is
kept at HEAD, so the live gate keeps the stronger check. It rests on an unverified assumption about
the dolt-2.x repo_state.json format — recommended follow-up: verify it and add a fallback/test.
(Its gated uow consumer reverted with the serverV2 layer — fine, that path isn't running.)

7ebf4df6a — fix(linear): unwire staleness hook from core — KEEP (loop in the feature owner)
A design reversal of a reviewed feature (ambient Linear-staleness warnings in every command's
pre-run and bd prime auto-pull). The objection (unpinned network I/O + version skew at session
start) is legitimate; it does reverse merged work, so the feature owner should ratify or re-land a
network-free version.

3fa29e325 — fix(fetch): drop session-killing DOLT_GC from failed-fetch path — KEEP (confirm with PR owner)
Reverses a DOLT_GC call added in reviewed PR #3326. dolt_gc invalidates all other open
sessions on the engine, so a failed fetch could kill concurrent in-flight connections — and the
tmp_pack_* cleanup it targeted already runs on teardown at BASE. Reverting this removal would
reintroduce that concurrency bug, so it is kept; confirm with the #3326 owner.

9a9da38c4 — docs(maintainer): add Merge Discipline and Review Requirements section — KEEP
Pure docs: +32 lines to PR_MAINTAINER_GUIDELINES.md, no code/behavior. Kept (or re-land under
attribution) — harmless; nothing to test.

[julianknutsen]

serverV2 reverted fixes — audit list

The serverV2 revert (8515879dc) restores the gated, not-running proxied/external-dolt-server
layer (dbproxy + uow + proxied cmd/bd glue) to BASE 9a1c88b63, removing the unreviewed
changes there. Most of these were real defects in the subsystem — they are listed here with
their proof so the owner can decide which to re-apply through proper review when serverV2 is
un-gated. (The domain/db data-layer fixes were kept at HEAD — they are welded to published
migration 0051 — so they are not in this list.)

proven = confirmed a real bug by mutation: reintroduce the defect into current code and the
listed test goes RED; restore the fix and it goes GREEN.

Commit	Area	Real bug?	The defect that was fixed (now removed)	Guard test
1a83cb08c	proxied-cmd	✅ proven	In internal/lockfile/lock.go, reverted IsLocked from return errors.Is(err, errProcessLocked) to the pre-fix pointer-equality return err == errProcessLocked.	TestIsLocked
29df600aa	dbproxy uow	✅ proven	In internal/storage/uow/doltserver_tx.go, releaseConn(): re-added the line t.conn = nil after _ = t.conn.Close() (reverting the fix), so the pinned conn reference is dropped on release/commit. This is exactly the defect: a use-after-com	TestDoltServerTxRunnerAfterCommitErrorsInsteadOfPanicking
4b0509c11	uow	✅ proven	In internal/storage/uow/errors.go, reverted isRetryableWarmupError to the pre-fix behavior: deleted the net.Error/driver.ErrBadConn/mysql.ErrInvalidConn/io.EOF/ECONNREFUSED/ECONNRESET branches so the function returns true only for isSeriali	TestIsRetryableWarmupError
794ff0790	uow	✅ proven	In internal/storage/uow/doltserver_tx.go, reverted doltServerTx.Commit's DOLT_COMMIT error handling back to the pre-fix behavior: replaced the if err != nil { ROLLBACK; poisonConn on rollback failure; return err } return nil block with th	TestDoltServerTxCommitFailureRollsBackBeforeRelease; TestDol
8866e5b2b	dbproxy	✅ proven	Two separate single-spot anti-fixes against current internal/storage/dbproxy/proxy/endpoint.go, each verified independently: / Item 10 (discovery trusting stale pidfiles): in readAndDial(), deleted the dead-writer guard `if pf.Pid <= 0 ||	TestReadAndDial_DeadPidfileWriterRejected; TestSpawnAndHando
99c70904f	uow	✅ proven	In internal/storage/uow/dolt_sql_provider.go, inside doltSQLProvider.initSchema (the proxied-server store-open path), I deleted the remote-migrate gate guard — the entire `if err := schema.CheckRemoteMigrateGateWithRemoteCheck(ctx, conn, ha	TestNewDoltServerUOWProvider_RemoteMigrateGate_BlocksReopen
b5cdb70ab	proxied-cmd	✅ proven	In cmd/bd/proxied_server.go, replaced the body of commandSupportsProxiedServer's final statement return proxiedServerCommands[cmd.Name()] with return true. This reintroduces the original defect: the PersistentPreRun guard no longer reje	TestCommandSupportsProxiedServer
f880a985b	proxied-cmd	✅ proven	In cmd/bd/uow_factory.go, reverted both error-propagation guards to the original swallowing form: persisted, _ := configfile.Load(beadsDir) (dropping the if err != nil { return ... "workspace config ... unreadable" } block) and `info, _	TestNewProxiedServerUOWProvider_CorruptSidecarErrorsInsteadO
1825cf357	proxied-cmd	— plausible-real-bug	Claims the proxied --tree --parent descendants CTE only walked parent-child edges and silently dropped dotted-ID orphan children (classic's ParentID LIKE fallback), and that gatherProxiedHierarchical failed outright when the tree root was a	—

Recommendation: the proven rows are genuine bugs (corrupt-sidecar abort, open-tx connection
recycled into the pool, warmup-retry misclassification, orphaned-dolt reaping after proxy SIGKILL,
the proxied-path migrate-gate hole, nil-store guard, use-after-commit panic) and should come back
through a reviewed PR when the proxied server is un-gated. Nothing here affects the running
classic/embedded path — full make test is green after the revert.

[julianknutsen]

Maintainer decisions — resolved

All flagged decisions are KEEP (no further reverts):

• 33bf6d0c0 (equal-timestamp upsert keeps local row) — KEEP. Carries regression tests demonstrating a real clobber under the old >= comparison; the equal-timestamp-keeps-local behavior is intentional.
• f1c72dbde (read persisted remotes from repo_state.json) — KEEP. Its live fix (dolt/federation.go + doltutil/remotes.go) stays at HEAD, so the live remote-migrate gate keeps the stronger check. Follow-up: verify the dolt-2.x repo_state.json format assumption and add a fallback/test.
• 7ebf4df6a (feat(linear): ambient staleness signal — bd prime auto-pull + --pull-if-stale #3669 Linear-staleness reversal) and 3fa29e325 (fix(fetch): run GC after failed DOLT_FETCH to prevent tmp_pack_* accumulation #3326 DOLT_GC removal) — KEEP (loop in the respective feature owners).
• 9a9da38c4 (maintainer-policy docs) — KEEP.

[julianknutsen]

We are soft reverting changes in the unreleased serverV2 feature in order to expedite development and reduce merge issues. All changes have been thoroughly documented in this PR for review and verification before serverv2 goes live. @coffeegoddd @timsehn

[maphew]

Wow! 😳 What a major effort. Let me know if there's anything concrete I can do to assist in the follow up!