Repository navigation
build(deps): Dolt 2.5.1 engine and pins + go-deps group, with compatibility fixes - #7646
Conversation
|
Triage: |
bee-ghosttrack
left a comment
There was a problem hiding this comment.
Approved on the code at 366f829; the merge itself waits, as the description says, on the Bazel / release cross-compile (freebsd/amd64) check, red on dolthub/file-locks v0.1.1 until a v0.1.2 tag exists.
Read every non-vendor hunk in a probe clone at that head. go build ./... ok; go test ./internal/doltversion/ ok; ./internal/ui -run TestTerminalStackPackageInitStaysUnderBudget pass; ./internal/storage/schema -run Prepared ok; BEADS_TEST_EMBEDDED_DOLT=1 go test ./internal/storage/embeddeddolt/ -run 'TestEmbeddedFirstMintOnEmptyHistoryIsRevisionOne|TestEmbeddedGuardedUpdateMintsOnceAfterItsPatches' both pass. Six mutants, all killed: next revision = MAX instead of MAX+1 (new test: 'first mint revision = 0, want 1'); the old query shape restored (all four tests: 'sql: no rows in result set', i.e. the bug reproduced on the 2.4.1 engine); subquery made global with 'OR 1=1' (killed by the three sibling tests only, the new test seeds one issue); RecommendedMin moved to 2.4.0, 2.3.3 and 2.5.0 (policy tests). A throwaway engine probe confirmed the go-mysql-server replaceAgg change: MAX under a constant primary-key prefix on an empty prefix now returns zero rows, while the scalar-subquery form returns one row with 1; the journal prune MIN(seq) still returns one NULL row; SHOW CREATE TABLE keeps created_at DESC.
Non-blocking gaps:
- The "2.3.3 write boundary" lives in prose only. The one code threshold is RecommendedMin=2.4.1, a once-a-day warning that runs only for managed proxied-server mode and
bd init --proxied-server, against the LOCAL dolt binary; plain server mode and external-proxied mode never read the server's version, andbd doctorprints dolt_version() as OK whatever it is. A database created by an embedded 2.4.1 bd and later served by a pre-2.3.3 Dolt is accepted silently and fails at its first write with 'table has unknown fields'. A doctor check comparing dolt_version() against 2.3.3 when idx_comp_snap_issue is descending would close it. - The new behaviour test is single-writer, single-issue; seeding a second issue with history would let it kill the global-subquery mutant on its own.
- Nothing guards the next query of this shape: replaceAgg now fires behind any constant key prefix or key range, so a future MAX/MIN of a key column under one would silently return zero rows. I confirmed version_history.go is the only affected production query today.
- policy.go says 'measured 2026-10-10, 2.2.0 and 2.3.1 fail'; troubleshooting.md says 'measured 2026-10-17, 2.2.0 and 2.3.2 fail; 2.3.3/2.4.0/2.4.1 succeed'. Consistent, but worth one wording.
|
Downstream integration results that may help the Dolt follow-up: we have qualified and installed an embedded Dolt v2.5.0-based engine in our Beads fork while retaining driver v2.4.1. Beads integration and lockfiles. The driver, embedded engine and external Dolt CLI remain separately versioned. This is a downstream qualification, not a proposal to merge this dependency group unchanged. Relevant results:
For the FreeBSD blocker, our fork retains its separately pinned BSD We did not change the external server/CLI policy, enable multi-branch commit options, or qualify standalone server/replication behavior with these embedded tests. No request to merge before your stated blockers are resolved. BlueboxC |
bee-ghosttrack
left a comment
There was a problem hiding this comment.
R2 — APPROVED at a46e441 (incremental over the R1 approval at 366f829).
Range-diff against the reviewed head: the eight reviewed commits are byte-identical (= on every row, same base ae693ef, no rebase); one commit is appended, build(deps): bump dolthub/file-locks to v0.1.2 for BSD builds.
Verified:
go.sumh1 lines for file-locks v0.1.2 match sum.golang.org (D0eaII+…HfJU=zip,8B7bWbbkj…kA5b1w=go.mod, the go.mod hash unchanged from v0.1.1 as expected for a tag that only edits a build constraint).- proxy.golang.org resolves v0.1.2 to tag 5b3ec47edfdd, which is the merge commit of dolthub/file-locks#6 (shrisha, merged). v0.1.1…v0.1.2 is
lock_unix.go//go:build linux || darwin→darwin || dragonfly || freebsd || linux || netbsd || openbsd, plus CI/README; no behavior change on linux/darwin. - Motivation holds: on 366f829 the
Bazel / release cross-compiletier was red (run 39011233701). - Bazel reads
go.modthroughgo_deps.from_file(go_work = "//:gazelle.go.work");MODULE.bazel.lockwas not touched by the earlier go.mod-only commits in this PR either, so no lock update is owed here.
Not verified: the recomputed Nix vendorHash (no nix toolchain on my side); the Nix build is the check for that.
CI on this head is still running (14 pass / 11 pending / 0 fail at 10:0xZ); the five Bazel tiers that stalled on #6943's farm this morning ([Sched] … 0 running, BEP lists no test results) are a farm condition, not a PR signal, if they recur here.
|
This PR has been added to the review queue. Track progress: https://factory.gascity.com/city/maintainer-city/runs/gcg--9223372036837089440 The review restarted at 2026-10-23 01:55 UTC; the link above follows the current run. |
The update-shaped mint ended with UPDATE issues SET current_revision = ? WHERE id = ?. current_revision always changes there, so the column's ON UPDATE CURRENT_TIMESTAMP clause fired: with versioned history on, every label, dependency or import write moved issues.updated_at to the session's wall clock (local time, read back as UTC). An import with RejectStaleUpserts then stored the wall clock instead of the source's updated_at, and the source's next genuine edit was rejected as stale. The statement now assigns updated_at to itself, the rule gastownhall#4298, blocked_state.go's mark/unmark templates and the create-shaped mint statements already follow, so history on leaves updated_at exactly where history off does. It is the one statement every production update-shaped mint ends with. Tests: - TestEmbeddedMintLeavesUpdatedAtWhereHistoryOffDoes (embeddeddolt) ages a bead, adds and removes a label and a dependency, and runs two RejectStaleUpserts imports, with history off first as the control. With the fix reverted, history on fails all six arms (updated_at reads the wall clock, and the later source edit is rejected as stale); history off passes either way. - RunParticipationGenerationFlagOffSkipsUpdateOnParticipatingRecord, wired on the dolt, embeddeddolt, uow and http legs: a record created with the flag on and updated with it off mints nothing and keeps its current_revision. Every other flag-off case uses a legacy record, which the fence stops first, so with RecordVersionInTx's gate removed this is the only case that fails, on all four legs. That gate is the single entry for every production update-shaped mint, so the one update arm covers label and close writes too. - The sqlmock pin of the statement (version_history_preenable_scan_test.go) and the journal-completeness exemption comment follow the new SQL. - The new embedded test has its line in the storage shard manifest (gen_embedded_storage_shard_manifest.py 40 --weights=duration --write), so TestStorageEmbeddedShardManifestGeneratorNotStale stays green. It sits after TestEmbeddedEveryAcceptedLifecycleMutationMintsOneVersion rather than at the end of the file, where gastownhall#7646 appends its own test, so the two PRs still merge without a conflict in that file. Docs: - CHANGELOG: the feature's two entries under ### Fixed described fixes to a switch no released bd has. One ### Added entry replaces them, covering bd versions, the versioned-history.enabled switch and its per-store semantics, the pre-enable check, the single-writer caveat, migration 0071 and the bd version <arg> refusal. - The bd versions help, its recording-off refusal and empty-listing note, and the opening paragraph of the configuration doc's Versioned Issue History section now state the rule the pre-enable refusal prints: a bead participates if it was created, or entered the issues plane (a wisp promoted or made persistent), while recording was on. bd version: the "did you mean bd versions" refusal now runs after the command's metrics event is created, like the other in-RunE argument refusals, so a mistyped bd version <id> is counted as a version invocation instead of going uncounted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Signature: claude-code-claude-opus-5-5-max on behalf of Julian Knutsen
The update-shaped mint ended with UPDATE issues SET current_revision = ? WHERE id = ?. current_revision always changes there, so the column's ON UPDATE CURRENT_TIMESTAMP clause fired: with versioned history on, every label, dependency or import write moved issues.updated_at to the session's wall clock (local time, read back as UTC). An import with RejectStaleUpserts then stored the wall clock instead of the source's updated_at, and the source's next genuine edit was rejected as stale. The statement now assigns updated_at to itself, the rule gastownhall#4298, blocked_state.go's mark/unmark templates and the create-shaped mint statements already follow, so history on leaves updated_at exactly where history off does. It is the one statement every production update-shaped mint ends with. Tests: - TestEmbeddedMintLeavesUpdatedAtWhereHistoryOffDoes (embeddeddolt) ages a bead, adds and removes a label and a dependency, and runs two RejectStaleUpserts imports, with history off first as the control. With the fix reverted, history on fails all six arms (updated_at reads the wall clock, and the later source edit is rejected as stale); history off passes either way. - RunParticipationGenerationFlagOffSkipsUpdateOnParticipatingRecord, wired on the dolt, embeddeddolt, uow and http legs: a record created with the flag on and updated with it off mints nothing and keeps its current_revision. Every other flag-off case uses a legacy record, which the fence stops first, so with RecordVersionInTx's gate removed this is the only case that fails, on all four legs. That gate is the single entry for every production update-shaped mint, so the one update arm covers label and close writes too. - The sqlmock pin of the statement (version_history_preenable_scan_test.go) and the journal-completeness exemption comment follow the new SQL. - The createbatchequiv issues_updated_at goldens of small, depadd, import458 and import458-reject-stale are re-recorded. They pinned the wall-clock bump, so each row minted a second time (eq-n2; eq-db1 and eq-dl; eq-b3 and eq-b250) had dropped out of the compared set; CI's embedded and server legs both printed the new digests. Recorded on embedded Dolt, the fix-reverted tree reproduces every committed golden byte for byte, HEAD moves only these four digests, and its rows equal the history-off rows in every scenario. GoldenDirEnv's comment lists it as the goldens' third deliberate change. - The new embedded test has its line in the storage shard manifest (gen_embedded_storage_shard_manifest.py 40 --weights=duration --write), so TestStorageEmbeddedShardManifestGeneratorNotStale stays green. It sits after TestEmbeddedEveryAcceptedLifecycleMutationMintsOneVersion rather than at the end of the file, where gastownhall#7646 appends its own test, so the two PRs still merge without a conflict in that file. Docs: - CHANGELOG: the feature's two entries under ### Fixed described fixes to a switch no released bd has. One ### Added entry replaces them, covering bd versions, the versioned-history.enabled switch and its per-store semantics, the pre-enable check, the single-writer caveat, migration 0071 and the bd version <arg> refusal. - The bd versions help, its recording-off refusal and empty-listing note, and the opening paragraph of the configuration doc's Versioned Issue History section now state the rule the pre-enable refusal prints: a bead participates if it was created, or entered the issues plane (a wisp promoted or made persistent), while recording was on. bd version: the "did you mean bd versions" refusal now runs after the command's metrics event is created, like the other in-RunE argument refusals, so a mistyped bd version <id> is counted as a version invocation instead of going uncounted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Agent-Signature: claude-code-claude-opus-5-5-max on behalf of Julian Knutsen
a46e441 to
2068dbe
Compare
…scending twin (0071) Review fixes for #7646, iterations 1, 2 and 3. This replaces the unpushed iteration-1 commit 89f67d6bbe. The nine contributor commits are rebased onto main at ce70b26, and the migration moves from 0070 to 0071 because #6943 landed its participation_generation migration at 0070 first. With two 0070s, the schema test package does not compile, the hygiene check fails, and every bd command that opens a store panics with "schema: duplicate migration version 70". The Dolt 2.4.1 driver stores the created_at DESC that migration 0010 declares on idx_comp_snap_issue as descending (dolthub/dolt#11746), so every database it creates was unreadable by Dolt older than 2.3.3: released bd (v1.3.x links driver v2.2.0) failed on open with "table has unknown fields", which BD_IGNORE_SCHEMA_SKEW=1 does not get past. Migration 0071 adds idx_comp_snap_issue_asc (issue_id, compaction_level, created_at) and then drops idx_comp_snap_issue, each step guarded on INFORMATION_SCHEMA.STATISTICS the way 0042 is. fk_comp_snap_issue keeps ON DELETE/ON UPDATE CASCADE. The descending order never served a query. Measured 2026-10-18 on a database bd init built from this tree (schema_migrations at 71, participation_generation present, no descending column at the root): dolt 2.1.8 and 2.2.0 read, write, push, clone and pull it, and the installed v1.3.0 bd (driver v2.2.0) stops at its structured "database is at v71, binary knows up to v69" refusal, which BD_IGNORE_SCHEMA_SKEW=1 gets past. With 0071 reverted, information_schema queries and writes fail with "table has unknown fields" (push and clone still copy the chunks), on 2.1.8 an UPDATE or DELETE on issues reports a fk_comp_snap_issue violation, 2.2.0's pull fails, and the v1.3.0 bd fails with the engine error that BD_IGNORE_SCHEMA_SKEW=1 does not get past. History keeps the descending column: bd init commits each migration separately, so the commits from before 0071 still hold it. On 2.1.8 and 2.2.0, reading compaction_snapshots at one of them (AS OF, dolt_history_compaction_snapshots, dolt_diff_compaction_snapshots) fails, and so does any dolt diff with one as an endpoint, even a diff limited to issues. issues AS OF, dolt_history_issues and dolt_diff_issues work. The up.sql header, the CHANGELOG and docs/architecture/dolt.md say so. Migrations take slots in landing order, and versions must stay gap-free (TestAllMigrationsSQLAppliesThroughDoltCLIAndRecordsLatestVersion). #6661 also numbers a migration 0071; whichever lands second renumbers. Tests: TestLatestVersionIncludesMigration0071 is the slot pin, and main's TestLatestVersionIncludesMigration0070 is deleted, as each earlier pin was when the next migration landed. A pure-Go test checks the frozen text and the down file, and a dolt CLI test checks the bundle leaves no descending column. An embedded test covers the upgrade from v69 through 0070 and 0071, a replay, the foreign key's cascade, resume after the add, a store without the table, and a fresh store. cli_prepared_ddl.go's inventory justifies the prepared ADD/DROP INDEX, measured to execute through the CLI batch path on 2.1.8 and 2.4.1. The example modules replace github.com/steveyegge/beads with the repo root, so they have to follow its dependency bump. On the iteration-2 head 72f5a6794e, scripts/build-examples.sh failed both with "updates to go.mod needed". go mod tidy in examples/library-usage and examples/bd-example-extension-go moves them to the root's versions (driver v2.4.1, dolthub/file-locks v0.1.2 and the go-deps group). Every requirement they share with the root go.mod now matches it except golang.org/x/exp, which both already pinned to a newer version on main. The script now prints "2 example module(s) type-check cleanly". Minor and nit findings: - The proxied-server old-version warning scopes each reason to the releases it covers, without naming a schema version: the CALL DOLT_RESET('--hard') regression measured on 2.3.0 and 2.3.1 (2.3.2 through 2.4.0 unmeasured, 2.4.1 measured clean), the join mis-plan before 2.4.0, and "table has unknown fields" before 2.3.3. A 2.4.0 user now reads why 2.4.1 is still recommended. TestWarningMessage pins all three; dropping the DOLT_RESET clause fails it. - scripts/ci/install-dolt.sh no longer says 2.4.1-engine storage is unwritable by 2.2.0. It names the join bug and the split between the current root, which 2.1.8 and 2.2.0 read and write, and the commits from before 0071. - scripts/test-git-remote.sh keeps its dolt >= 2.2.0 floor, backed by a 2.2.0 run of its write, push, clone and pull steps. - The dated cross-version measurements live in docs/architecture/dolt.md; policy.go and troubleshooting.md point there. - The CHANGELOG scopes the fk_comp_snap_issue violation to 2.1.8. - The localdoltserver port comment covers 2.4.1 as well as 2.2.0; both print "Port N already in use." on a held port (measured). - TestEmbeddedFirstMintOnEmptyHistoryIsRevisionOne pins that revisions count per issue. - docs/architecture/dolt.md says 2.4.1 is the only release after 2.3.1 measured for the DOLT_RESET regression and that 2.3.2, 2.3.3 and 2.4.0 have not been. It had said no release after 2.3.1 was, right under the 2.4.1 0 / 250 row. - cli_migrations.go puts the parity oracle's green run against the then-pinned 2.2.0 sql-server in the past tense. Probed 2026-10-18, a prepared ADD COLUMN vanishes on the 2.2.0 CLI and lands on 2.4.1, so on the 2.4.1 pin the oracle cannot catch a dropped override; the comment names the two binary-free tests that still can. Agent-Signature: claude-code-claude-opus-5-5-unknown-reasoning on behalf of Julian Knutsen Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bee-ghosttrack
left a comment
There was a problem hiding this comment.
R3 — incremental over R2 (a46e441 → 2068dbe): APPROVE stands.
The force-push (16:54Z) is a rebase from ae693ef onto current main ce70b26 (post-#6943). git range-diff origin/main a46e441341 2068dbee97: 8 of 9 commits =, one ! — fix(storage): compute the next version revision with a scalar subquery — and that row differs in context only:
version_history.go: the same ±hunk at a new offset (L371 vs L278), now insiderecordVersionInTx(…, isCreate bool)— #6943's shared body — rather than the pre-splitRecordVersionInTx. Identical patch text. The oldCOALESCE(MAX(revision), 0) + 1 … WHERE issue_id = ?is still what main L376 runs, so the fix lands on the live site.version_completeness_behaviour_test.go: byte-identical patch (diffof the twogit showoutputs is empty).embedded-storage-test-shards.txt: same one added line (40 4 TestEmbeddedFirstMintOnEmptyHistoryIsRevisionOne), neighbours moved by the rebase.
Ran at 2068dbe in a fresh worktree:
go build ./internal/storage/... ./cmd/bdok (pulls file-locks v0.1.2);go vetissueops + embeddeddolt ok.go test ./internal/storage/issueops -run 'Version|Revision|Mint': 10 PASS (the static mint-completeness suite still recognizes bothRecordVersionInTxand the create-shaped sibling).BEADS_TEST_EMBEDDED_DOLT=1 go test ./internal/storage/embeddeddolt -run 'FirstMint|GuardedUpdateMintsOnce|ParticipationGeneration|CreateBatchFastPathsMatchPerRow': 8 PASS, 0 SKIP (51s) — includingTestEmbeddedFirstMintOnEmptyHistoryIsRevisionOneand the #6943 fence/participation tests on the composed tree.- Mutant at the rebased head: restoring the old
MAX(revision)query makes bothTestEmbeddedFirstMintOnEmptyHistoryIsRevisionOneandTestEmbeddedGuardedUpdateMintsOnceAfterItsPatchesfail withcompute next revision …: sql: no rows in result set— the regression test still bites at the relocated site.
Nit (non-blocking): the header comment at L62–63 now says "in RecordVersionInTx"; the query lives in recordVersionInTx after #6943. Cosmetic.
CI at post time: 25 success / 10 skipped / 1 in progress (CI Gate / Required). vendorHash still unverified here (no nix), as in R2.
Maintainer Adoption ReviewThanks for the contribution, @julianknutsen! This PR moves beads onto the Dolt 2.4.1 engine. It carries Dependabot's go-deps group, including
Credit also goes to @coffeegoddd, whose #7254 pin raise is carried in This PR was reviewed and adopted with maintainer fixes pushed directly to the PR branch. Original PR ReviewDecision: request_changes resolved to approve. Iterations 1 and 3 requested changes, iteration 2 blocked, and iteration 4 approved. Specific gaps fixed:
The review also checked the contributor's own fixes and confirmed them without changes:
Review findings addressed:
Non-blocking follow-ups (shared for visibility — NOT completed in this PR):
@BlueboxC, thank you for the downstream qualification. The upstream file-locks release your FreeBSD note refers to has shipped. Maintainer Changes
Final Review StatusReady for the merge queue: final head Since #6661 merged, GitHub reports this PR conflicting with main. The merge queue owns the merge attempt, and the repair it needs is the bd-jnaarc item above. CI: https://github.com/gastownhall/beads/actions/runs/39147155734 Review Iterations4 review passes were performed, each by three reviewers (Claude, Codex and Gemini).
The fixes from iterations 1 to 3 were never pushed separately: they are one commit, @bee-ghosttrack's R3 noted that the Nix vendorHash went unverified locally. CI covers it: Adopted via claude-code-claude-opus-5-5-max on behalf of Julian Knutsen |
…e-way (floor 2.3.3) Maintainer decision on #7646: the descending-index storage change is accepted as one-way. Dolt 2.3.3 and later store migration 0010's created_at DESC on compaction_snapshots' idx_comp_snap_issue as descending (dolthub/dolt#11746), and Dolt older than 2.3.3 cannot read such a table ("table has unknown fields"). bd does not convert the index back, so every database bd creates from this release on needs Dolt 2.3.3 or newer, and a bd built on it, for good. Databases created earlier keep an ascending index and stay readable and writable by older Dolt even after this release migrates and writes to them (measured 2026-10-21 with Dolt 2.2.0 and 2.3.2). - CHANGELOG: the entry now says the embedded engine is Dolt 2.5.1 (driver v2.4.1 with dolt/go at the v2.5.1 release), that the change is one-way, that the recommended dolt is 2.5.0 (CI pins 2.5.1), and lists the defects below it: dolthub/dolt#12093 on 2.4.0-2.4.2, the join bug before 2.4.0, the DOLT_RESET('--hard') regression on 2.3.0 and 2.3.1. - docs/architecture/dolt.md: the pin paragraph says embedded mode links the 2.5.1 engine and the CLI pin tracks it; a one-way statement; and one version table merging the DOLT_RESET measurements with the 2.3.3 storage floor and the MAX/MIN engine bug (2.4.0-2.4.2 failing, 2.5.0 and later correct; 2026-10-21: 2.3.1 4/300, 2.5.0 and 2.5.1 0/500 each, 2.4.2 0/250). - troubleshooting.md: the old-version warning section recommends 2.5.0 and lists the three defect ranges, pointing at dolt.md. - scripts/test-git-remote.sh: dolt >= 2.3.3. Its 2.2.0 floor rested on migration 0071, which this branch does not carry. - sync-setup.md scopes the DOLT_RESET regression to 2.3.0 and 2.3.1 (2.3.2 carries the fix); upgrading.md scopes the #7037 embedded-engine caveat to bd releases that link Dolt older than 2.4.0. The MAX/MIN column is bee-ghosttrack's reproduction from #7646. Agent-Signature: claude-agent-sdk-claude-opus-5-5-unknown-reasoning on behalf of Julian Knutsen Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2068dbe to
68a3184
Compare
Maintainer decisions (2026-10-23): branch rebuilt on current main, Dolt 2.5.1The branch has been force-pushed. It is now the nine contributor commits, rebased onto current
Credits and replies:
Re-review needed. The earlier approvals were on the 2.4.1 shape. The head, the engine and the version policy have all changed, so this needs a fresh review before it merges. claude-agent-sdk-claude-opus-5-5-unknown-reasoning on behalf of Julian Knutsen |
bee-ghosttrack
left a comment
There was a problem hiding this comment.
Reviewed at 68a318409adb (14 commits on 3b4906684948, the current main; PR CI 28 pass / 10 skipping, nothing pending). Approving the rebuilt head.
Range-diff against my last approved head 2068dbee9796. Commits 5–9 are identical. Commit 1 (go-deps group) lost its x/crypto, x/mod, x/net, x/text hunks because main's #6355 now carries them. Commit 2 only re-bases the old side of the vendorHash. Commit 4 (scalar subquery) gains the two sqlmock expectation updates for #6661's pins and the recordVersionAtInTx comment. The nogo alignment commit is recreated with the same content (root and tools/nogo both at x/mod v0.41.0, x/sync v0.24.0, x/text v0.42.0). 2068dbee9796 with migration 0071 is gone; five new commits replace it (engine 2.5.1, pins 2.5.1, comment fixes, per-issue first-mint test, docs table).
The 2.5.1 pin is the release set. go.mod's dolthub/dolt/go v0.40.5-0.20261020191035-d1f8f8e76386 is the v2.5.1 tag commit, and go-mysql-server 239e0ad6b2f9, vitess 3e903c0f05d3, file-locks v0.1.2, jsonpath 2af4aa8c81af, go-icu-regex 72563bc7ca83, ishell 5f031e3e9037 and eventsapi_schema d9b4a1c6b4d4 are exactly what dolt v2.5.1's own go/go.mod requires; no replace. The four tarball sha256s in tools/bazel/dolt.bzl equal the v2.5.1 release asset digests GitHub reports. install-dolt.sh, pull-dolt-image.sh, DoltDockerImage and DOLT_VERSION all say 2.5.1.
Measured here on this head (macOS arm64, go1.27.1):
go build ./...andgo vet ./...clean;go mod tidyleaves the root module, bothexamples/*modules andtools/nogobyte-identical;go mod verify: all modules verified.go test ./internal/doltversion/ok (the pin-at-least-RecommendedMin guard and the warning text with its four release ranges);./scripts/ok (the install-pin and container-tag drift guards, 361s);./internal/testutil/fingerprint ok;./internal/storage/schema/string guards ok;./internal/storage/issueops/sqlmock ok;./internal/ui/TestTerminalStackPackageInitStaysUnderBudgetPASS.- On the linked 2.5.1 engine (
BEADS_TEST_EMBEDDED_DOLT=1 -tags gms_pure_go):TestEmbeddedFirstMintOnEmptyHistoryIsRevisionOnePASS (5.6s). Mutant dropping theissue_idfilter from the scalar subquery: FAILfirst mint revision for a second issue = 3, want 1, as the commit says. Mutant restoring the oldCOALESCE(MAX(revision), 0) + 1 FROM issue_versions WHERE issue_id = ?shape: PASS on this engine, consistent with the table (2.5.x answers that shape correctly), so the scalar form guards a 2.4.x-class server, not the linked engine; the test pins the per-issue count either way.
Reading. RecommendedMin 2.0.0 → 2.5.0 stays a warning, not a floor, and its message names each range the doc table measures. The one-way decision reads the same in the CHANGELOG and docs/architecture/dolt.md: a database created from this release on needs Dolt ≥ 2.3.3 for good, an older one keeps its ascending index. The MAX/MIN column is my 2026-10-20 reproduction, adopted as stated. The localdoltserver_test SQLMode fingerprint is the 2.3.3+ default the CHANGELOG now names. Nothing from the 2.4.1-shape review is lost: the example-module tidy, the per-issue first-mint test, the scoped warning and the comment fixes are all here without 0071.
Maintainer Adoption ReviewThanks for the contribution, @julianknutsen! This PR moves beads onto the Dolt 2.5.1 engine:
Credit also goes to:
This PR was reviewed and accepted as-is. Merging it also resolves two follow-ups from earlier reviews. Both are noted to close when this lands:
Original PR ReviewDecision: approve, on the first review pass of the rebuilt head. Specific gaps fixed:
Non-blocking follow-ups (shared for visibility — NOT completed in this PR):
Maintainer ChangesNo maintainer changes -- PR accepted as-is. Final Review StatusReady for the merge queue: final head CI: https://github.com/gastownhall/beads/actions/runs/39886181447 Review Iterations1 review pass was performed on the rebuilt head, by two reviewers (Claude and Codex). It approved with no required changes. This was a fresh review: the 2026-10-18 adoption review covered the earlier 2.4.1 shape. @bee-ghosttrack separately approved the same head at 02:15 UTC with their own measurements. They included the old-query mutant that passes on the 2.5.1 engine, which is the first follow-up above. Adopted via claude-code-claude-opus-5-5-max on behalf of Julian Knutsen |
|
Follow-up integration qualification: our embedded-engine fork is now based on official Dolt 2.6.0, retaining driver/v2 2.4.1 and the existing workspace formats/customizations. Consumer integration: BlueboxC@0429f95 Upstream corrections already corresponding to carried patches are included once. The new ORT merge-base implementation is reused, while our previously qualified refusal of conflicting virtual bases remains at merge and three-dot diff entry points; automatic reconciliation of such bases is not claimed. Root/example modules require Go 1.27.3 and explicitly pin both replacements. The older linter cannot read its export data, so all lint lanes pin the verified compatible v2.15.0. The Nix compiler-source and dependency-cache hashes were updated; a native Nix build was not run locally. Verified: normal/prepared fork engine regressions; nine storage packages; affected analyzer/types/plan packages and all three query-plan suites; independent example builds/tests; native and Windows cross-lint. The complete Beads attempt passed 108 test-bearing packages and exposed one JSON-escape premise in graphops; that test now checks the decoded U+FFFD property, and the corrected package plus all adapted packages pass. The full make command was not repeated after those narrow adaptations. Four pure-Go regex flag cases reproduce against the unchanged prior analyzer as well; ICU parity is not claimed. Isolated backups from native Dolt 2.4.2 and the preceding embedded tuple pass old/new restore, table/working-root/branch/tag/history equality, new writes/reopen/GC and a second backup readable by both engines. Installation then compared all three local Beads databases before/after under exclusive gates, retaining matched recovery sets. All six owned services resumed; readonly graph/context checks pass for the three workspaces. This does not qualify a separate external Dolt CLI/server upgrade or HAL/openclaw application execution. BlueboxC |
Post-merge review:
|
| Database | Dolt 2.1.8–2.3.2 | Dolt 2.3.3–2.5.1 |
|---|---|---|
created by this head (idx_comp_snap_issue stored created_at D) |
fails as in finding 1. Nothing is damaged: afterwards 2.5.1 reads, writes and commits, dolt fsck is OK, and bd works |
read, write, commit OK |
| created by the 2.4.1 engine, then opened by this head | fails as in finding 1 (measured on 2.3.2) | OK on 2.3.3 and 2.5.1. The head lists, creates, updates and closes; the index stays D |
| created by main (2.2.0) or by v1.3.1 (v66), then migrated to v72 and written by this head | index stays A. 2.2.0 and 2.3.2 read, write and commit. main's bd and v1.3.1 (with BD_IGNORE_SCHEMA_SKEW=1) reopen it |
OK on 2.3.3 and 2.5.1 |
The padded journal (perf: Use fdatasync on linux with a padded journal dolthub/dolt#12031).
- A 2.5.x writer keeps about 4 MiB of trailing zeros while open.
- A clean close (bd's
Connector.Close, or a SIGTERM to the server) truncates them. A crash, or an exit without Close, leaves them. - On crash copies, these all read every commit, wrote, committed and trimmed the padding, and
fsckwas OK afterwards:- CLIs 2.2.0 and 2.3.2 (ascending database);
- CLIs 2.3.3, 2.4.1, 2.4.2 and 2.5.1;
- this head, the 2.4.1-engine build and main's bd;
- v1.3.1.
- 2.2.0's
journal_record.godiffers from 2.5.1's only in its lock import, and it already skips zero padding.
Engine behavior, 2.4.1 → 2.5.1.
- Settings. sql_mode, collation and isolation are unchanged on the CLI, the server and embedded. The only new variable is
dolt_multi_branch_commit, which defaults to OFF. - Fixed shapes. The
MAX/MINno-row shape and the beads-mcp admin validate action calls unsupported bd command #4037 panic are fixed on 2.5.0, 2.5.1 and embedded, including IN-list, range, GROUP BY, JOIN and secondary-index variants. - SQL replay. I ran 966 SQL templates from bd's non-test code on 2.2.0, 2.4.1 and 2.5.1. There were no new errors, and only the one result change in finding 2.
- bd commands. 53 read commands and 21 writes gave identical output on the 2.2.0, 2.4.1 and 2.5.1 builds. Version-control queries were identical on a 102-commit history.
- The
DOLT_MERGEguard (in 2.5.0).DOLT_MERGEand the merge insideDOLT_PULLnow refuse when the session has a pending write in the same explicit transaction ("cannot merge with uncommitted changes"). I measured it: 2.4.1 merges that shape, 2.5.1 refuses it. bd is unaffected:- every
DOLT_MERGEandDOLT_PULLit issues runs under autocommit or after onlySETstatements (store.go:5348-5407,mergesettle.go,fastforward.go,version_control.go); - the GH#3144 shape (
SET, thenDOLT_FETCH, thenDOLT_MERGEin one transaction) merges on both versions.
- every
- Stricter aggregate binding (go-mysql-server test: let bash canonicalize formula path in release script tests #3797). An ambiguous unqualified aggregate argument over a join now errors, and an aggregate over only outer columns now binds to the outer query. No bd production query has either shape.
Lanes and tests at this head.
- Embedded lane, uncached: 10 of 10 targets pass, with 1781 pass, 57 skip and 0 fail top-level testcases. The same tree on the 2.4.1 engine (
f0263d61cc) gave 1781/57/0, with zero per-test differences.TestEmbeddedFirstMintOnEmptyHistoryIsRevisionOne,TestEmbeddedGuardedUpdateMintsOnceAfterItsPatches,TestFirstMintLeavesTheEpochTablesCommittedandTestGraphWalkPlansHonorJoinHintsPASS rather than skip. dolt-race, uncached: 12 of 12 targets pass, with 953 pass, 8 skip and 0 fail, includingTestDoltServerFingerprint/localon the hermetic 2.5.1.- Fingerprint, locally: both the local CLI and the container subtest pass on
dolthub/dolt-sql-server:2.5.1(sha256:3b35e5ee…). - Mint mutants on the 2.5.1 engine:
OR 1=1is killed: "first mint revision for a second issue = 3, want 1".MAXwithout+1is killed.- The old flat query is caught only by the sqlmock pins (finding 3).
- Guards pass:
TestPinnedDoltCLIMatchesContainerImage,TestPinnedDoltCLIAtLeastRecommendedMin,TestWarningMessage,TestCheckRecommended,TestDoltVersionPinsAgreeandTestInstallDolt*. - Build and vet:
CGO_ENABLED=0 go build ./...andgo vet -tags gms_pure_go ./...are clean. dolt.bzl: all four 2.5.1 digests match the release assets.
Version policy, measured on 13 release CLIs (2.1.8 to 2.5.1).
MAX/MIN: correct on 2.1.8–2.3.5; no row plus the beads-mcp admin validate action calls unsupported bd command #4037 panic (index out of range [2] with length 2) on 2.4.0, 2.4.1 and 2.4.2; correct on 2.5.0 and 2.5.1.- The stricter default sql_mode starts exactly at 2.3.3, as the CHANGELOG says.
RecommendedMin2.5.0 is consistent with the pin, 2.5.1.
Rebase.
git range-diff ce70b261c5..2068dbee97 origin/main..68a318409ashows only expected differences:- commits 5–9 are
=; - commits 1–3 lost only the bumps fix(deps): raise x/net and x/crypto past every fixable advisory that reaches the built binary #6355 had already landed on main;
- commit 4 gained the two
QuoteMetapin updates and therecordVersionAtInTxcomment; 2068dbee97was dropped.
- commits 5–9 are
- The branch adds no migration.
- It was based on the then-tip of main, and the diff against main touches only this PR's 37 files, so nothing from main was dropped.
Label state
The PR is merged, and status/merge-queued and status/reviewing were already removed, so I changed no labels. The findings above are docs-only and can go in a follow-up next to bd-sa43it and bd-qtin5n.
claude-agent-sdk-claude-opus-5-5-max on behalf of Julian Knutsen
…wording (gastownhall#7646 follow-ups) (gastownhall#7807) * docs(dolt): name the fk_comp_snap_issue symptom below Dolt 2.3.3 and scope the descending-index wording Follow-ups from the post-merge review of gastownhall#7646, measured on release CLIs 2.1.8 through 2.5.1 against a database bd 68a3184 created: - Below 2.3.3 a write to issues fails with "cannot add or update a child row: a foreign key constraint fails" on fk_comp_snap_issue, not with "table has unknown fields"; only other writes, information_schema and dolt commit report the latter. Name both symptoms in the CHANGELOG, docs/architecture/dolt.md, troubleshooting.md and the RecommendedMin comment. - Released v1.3.x fails at open as documented, but a bd built from source against driver 2.2.0 at the same schema version opens and reads the workspace and fails only on writes; the CHANGELOG said it could not open. - "Every database bd creates" holds for embedded mode; a server-mode database whose schema a pre-2.3.3 server created stays ascending (bd-qtin5n). - The dolt.md table: 2.3.0 is now measured (no; MAX/MIN correct), every "opens" row is measured, and every release before 2.5.0 returns no row for an ungrouped MAX over an empty table's whole primary key. - The mint comment and its test's doc comment described the zero-row rewrite as the behaviour of the engine bd links; on 2.5.x both query forms pass the embedded tests, so say the sqlmock pins hold the form (bd-sa43it, comments only). - Re-wrap the fresh-bootstrap heal comment. Comments and docs only; TestWarningMessage's substrings are unchanged. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> * fix(doltversion): name both descending-index populations and both write errors below Dolt 2.3.3 Review follow-up for gastownhall#7807 (adopt-pr iteration 1). The proxied-server warning, the RecommendedMin comment and the troubleshooting bullet said only databases bd creates in embedded mode hold the descending index. The warning prints in proxied-server mode, where the probed dolt runs the sql-server that creates the schema, so all three now also name any database whose schema a dolt sql-server 2.3.3 or newer created, as the CHANGELOG body and dolt.md already did. The warning now names the fk_comp_snap_issue error too, and TestWarningMessage pins that error and both populations: the PR head's wording fails it on all three, and dropping only the sql-server population fails it as well. Measured 2026-10-23 with the 2.1.6, 2.2.0 and 2.3.2 release CLIs on a schema the 2.5.1 CLI created (migration 0010's DDL with 0042's ON UPDATE CASCADE): an UPDATE or DELETE on issues, and the INSERT-then-UPDATE transaction bd create runs, fail with "cannot add or update a child row" on fk_comp_snap_issue. A plain INSERT into issues or any other table, information_schema.tables, CALL DOLT_COMMIT and a SELECT from compaction_snapshots fail with "table has unknown fields". Reads of issues, information_schema.schemata and SHOW TABLES succeed. So "a write to issues" becomes "an UPDATE or DELETE on issues" in the CHANGELOG, dolt.md, troubleshooting.md, the warning and both pin comments. Also: - CHANGELOG: scope the headline to embedded mode, and state the error split without "writes to any table ... instead". - CHANGELOG and dolt.md: "a database created earlier keeps its ascending index" becomes "any other database": one that an earlier bd created through a 2.3.3+ server already holds the descending index. - scripts/ci/install-dolt.sh: the pin comment matches policy.go. - dolt.md: drop the 2.3.2 row's "(measured)" now that every row is, and say bd's production queries accept either empty-table answer (a conformance helper does not). Agent-Signature: claude-code-claude-opus-5-5-unknown-reasoning on behalf of Julian Knutsen Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Julian Knutsen <julianknutsen@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Co-authored-by: Eddie the Engineer <338826946+gascity-eddie[bot]@users.noreply.github.com>
Supersedes #7623 (and the earlier #7441, which Dependabot closed on 2026-10-16 after #7495 landed grpc/thrift). This branch carries #7623's go-deps group commit, then the compatibility fixes #7441 needed, then a move of the embedded engine and every Dolt pin to Dolt 2.5.1. Once this merges,
maincontains the updates and Dependabot should close #7623 on its own.Maintainer update (2026-10-23): rebuilt on main, Dolt 2.5.1, one-way storage change accepted
The branch was rebuilt and force-pushed on main
3b49066849. The decisions are from @julianknutsen; the PR comment has them in full.2068dbee97("replace compaction_snapshots' descending index with an ascending twin (0071)") was removed by maintainer decision. The descending-index storage change is accepted as one-way, with a Dolt floor of 2.3.3 (see below). The PR adds no migration, so main's 0071 (bd versions + a config switch for versioned history — the minimum cut that makes versioned beads usable (#6138) #6661) and 0072 (Phase 3 (R20): epoch-transition enforcement + minted-address CAS - gastownhall/beads#6136 #6664) stand and nothing is renumbered. The review fixes2068dbee97carried that did not depend on 0071 are re-applied in commits 10–14.dolthub/driver/v2stays at the tagged v2.4.1.dolthub/dolt/go,go-mysql-serverandvitessare pinned to the exact Dolt v2.5.1 release set, with noreplace. No driver tag exists on 2.5.x, and driver main requires go 1.27.3. The driver changed no Go source between v2.4.1 and its 2.5-era commit89f53f8d3f72, so this pairing is source-identical to DoltHub's own build. Dolt 2.5.0 fixed the ungroupedMAX/MINengine bug (Ungrouped MAX/MIN over a primary-key column returns no row when nothing matches dolthub/dolt#12093) and the Panicindex out of rangeinreplaceAggfor ungrouped MIN/MAX of a non-key column when WHERE fixes the whole primary key dolthub/go-mysql-server#4037 panic.dolthub/dolt-sql-server:2.5.1) and the hermetic Bazel dolt are all at 2.5.1, anddoltversion.RecommendedMinis 2.5.0. The scalar-subquery mint read stays, because server-mode users on dolt 2.4.0 through 2.4.2 still need it.Earlier approvals were on the 2.4.1 shape, so this needs a fresh review.
Commits
build(deps): bump the go-deps group across 1 directory with 20 updates: build(deps): Bump the go-deps group across 1 directory with 20 updates #7623's commit (Dependabot authorship kept). On this base its x/net bump is a no-op, because main's fix(deps): raise x/net and x/crypto past every fixable advisory that reaches the built binary #6355 already took x/net to v0.60.0.build(nix): move the flake to nixos-26.05 for go 1.26.5+: driver v2.4.1's go.mod requires go >= 1.26.5, and the frozennixos-25.11branch ships 1.26.4.build(nogo): align tools/nogo … with the root module: fixesTestModuleRequirementsAgreeWithRoot. Dependabot only watches/.fix(storage): compute the next version revision with a scalar subquery: on Dolt 2.4.0–2.4.2, an ungroupedMAX(revision)over an empty primary-key prefix returns zero rows, which broke the first versioned-history mint on every issue. The query is nowSELECT COALESCE((SELECT MAX(revision) …), 0) + 1. This commit addsTestEmbeddedFirstMintOnEmptyHistoryIsRevisionOneand moves bd versions + a config switch for versioned history — the minimum cut that makes versioned beads usable (#6138) #6661's twoQuoteMetapins to the new literal.build(dolt): raise the Dolt CLI, container and Bazel pins to 2.4.1: carried from build(deps): align Dolt driver and test versions to 2.4.1 #7254 (co-authored by @coffeegoddd).docs(changelog): note the Dolt 2.4.1 engine's descending-index storage breakfix(deps): move go-runewidth to v0.0.28: keeps a 2 MB width-table init (about 35 ms) off every bd command.docs(dolt): name 2.3.3 as the write boundary and drop the stale 2.3.x DOLT_RESET claimbuild(deps): bump dolthub/file-locks to v0.1.2 for BSD builds: picks up fslock: build on freebsd, openbsd, netbsd and dragonfly again dolthub/file-locks#6 (by @shrisha), so freebsd/openbsd/netbsd build with noreplace.build(deps): run the embedded Dolt engine on 2.5.1: go.mod/go.sum (4 module lines), thedefault.nixvendorHash, andgo mod tidyin both example modules (without itscripts/build-examples.shfails).build(dolt): move the CLI, container and Bazel pins to 2.5.1 and recommend 2.5.0:install-dolt.sh,pull-dolt-image.sh,testdoltcommon.go,dolt.bzl(digests match the release assets), the test constants, the docs snippet,RecommendedMin2.5.0, and an old-version warning that names each defect range.TestWarningMessagepins each one.docs(code): correct the parity-oracle and port-check comments for the 2.5.1 pin: review fixes re-applied from2068dbee97.test(storage): pin per-issue first-mint revisions: from2068dbee97. It kills a mutant that drops theissue_idfilter.docs(dolt): Dolt 2.5 version table; the descending-index change is one-way (floor 2.3.3): CHANGELOG,docs/architecture/dolt.md, troubleshooting,scripts/test-git-remote.sh(dolt >= 2.3.3).One-way storage change (accepted)
Dolt 2.3.3 and later store migration 0010's
idx_comp_snap_issue (…, created_at DESC)as a real descending index (dolthub/dolt#11746). Dolt older than 2.3.3 answerstable has unknown fieldsfor such a database, which includes every releasedbd(v1.3.x links Dolt 2.2.0 or older), andBD_IGNORE_SCHEMA_SKEW=1does not get past it. Every databasebd initcreates from this release on is affected, and so is any database whose schema adolt sql-server>= 2.3.3 created.bddoes not convert the index back. Databases created earlier keep an ascending index and stay readable and writable by older and newer Dolt, even after this release migrates and writes to them (measured 2026-10-21 with Dolt 2.2.0 and 2.3.2). The CHANGELOG states this.Dolt version table
bdcreatesMAX/MINover an unmatched PK prefixDOLT_RESET('--hard')brokenThe
MAX/MINcolumn is @bee-ghosttrack's repro, re-measured here with the release CLIs on linux/amd64. TestDoltServerFingerprint matches the existing golden on the 2.5.1 CLI and container: the defaultsql_modedid not change from 2.4.1.Verification (local, at this head)
go build ./...andgo vet ./...: clean.go mod tidyis a no-op on the root,tools/nogoand both example modules.scripts/build-examples.sh: "2 example module(s) type-check cleanly".CGO_ENABLED=0 go build ./cmd/bdfor linux/{amd64,arm64}, android/arm64, freebsd/amd64, darwin/{amd64,arm64}, windows/{amd64,arm64}, openbsd/amd64 and netbsd/amd64: all build.scripts/ci/bazel-release-cross-compile.sh: passes for all 8 release targets.--config=ci197/197,--config=integration155/155,--config=dolt-race12/12 (hermetic dolt 2.5.1),--config=dolt-int4/4,--config=embedded10/10 targets with 0 failed testcases. The--config=purebuild,bd versionand the pure test subsets pass, and--config=js-wasm //internal/hooks:hooks_testbuilds.bazel test //tools/bazel:dolt_version_test //tools/nogo:nogo_test //scripts/... //internal/doltversion/... //test/docsync/...: 31/32. The exception is//scripts:doc_freshness_required_test, which fails identically on pristine main3b49066849("compiled inventory missing=…"). It is pre-existing and not in--config=ci.go testforinternal/storage/issueops(bd versions + a config switch for versioned history — the minimum cut that makes versioned beads usable (#6138) #6661's pins),internal/doltversion,scripts(TestPinnedDoltCLIMatchesContainerImage,TestPinnedDoltCLIAtLeastRecommendedMin) andtest/docsync: pass, at the head and at each of commits 10–14.BEADS_TEST_EMBEDDED_DOLT=1:TestEmbeddedFirstMintOnEmptyHistoryIsRevisionOne,TestEmbeddedGuardedUpdateMintsOnceAfterItsPatches,TestFirstMintLeavesTheEpochTablesCommittedandTestGraphWalkPlansHonorJoinHintsall PASS (none skip) on the 2.5.1 engine. With theissue_idfilter mutated away (OR 1=1), the first-mint test fails with "first mint revision for a second issue = 3, want 1".TestDoltServerFingerprint:local(the dolt 2.5.1 release CLI, tarball sha256 matchesdolt.bzl) andcontainer(dolthub/dolt-sql-server:2.5.1, digestsha256:3b35e5ee…) both pass against the existing golden. The 2.5.1 CLI prints "Port N already in use." on a held port.nixos/nixin Docker):nix build .#defaultsucceeds with vendorHashsha256-6FetqEO9fOyj1+0fOQPb5/pybho0yOztwPEbMO4Cxak=,bd --helpruns, andnix flake checkreports "all checks passed".make bazel-sync-checkandmake bazel-syncleave no drift.make check-docspasses.scripts/check-migration-hygiene.shis OK, and the PR adds no file underinternal/storage/schema/migrations/.Related, and left open: #7254, #7045 and #6811 (Dolt-family subsets this PR supersedes). They will be closed separately with credit once this lands. bd-4besq4 (nothing yet checks go.mod's
dolt/gocommit against the CLI pin) stays open.🤖 Generated with Claude Code