Uh oh!
There was an error while loading. Please reload this page.
fix(driver-mongodb): take a structured GroupByNode, and answer count/count_distinct like every other backend - #7550
Merged
Conversation
…count_distinct like every other backend (#6850, #6814) `driver-mongodb` is now enrolled in the shared `AGGREGATION_CASES` standard. Clearing that cell fixed three divergences, all of the kind that ANSWER rather than fail — which is why none of them ever surfaced as an error: 1. #6850 — a structured `GroupByNode` had no lowering at all. `groupBy` was annotated `string[]` and `groupId[field] = '$' + field` stringified the object, so the `$group._id` key became the literal `"[object Object]"` and its value a field path matching nothing: rows grouped by a nonexistent path, under a column of that name. `MongoDBDriver.aggregate` passed the value through an `any` cast, which is why the declared union never met that annotation at `tsc`. Both sides now spell `GroupByNode[]`, and the `_id` keys on `alias ?? field` with the FIELD as its value — the #6401 rule. 2. #6814 — `count_distinct` sized a `$addToSet` that keeps explicit nulls, so a nullable column answered 3 where the standard says 2. The sizing now excludes null, matching `COUNT(DISTINCT col)` and the in-memory fallback. 3. Measured here, named by neither card — `count(col)` ignored `field` and emitted `{ $sum: 1 }`, so `count(stage)` came back 6 where the standard says 4. It now counts non-null values, a missing field reading as null. A `dateGranularity` node is refused with NOT_IMPLEMENTED/501 in the ADR-0112 envelope rather than silently ignored — the `driver-sql` / `driver-turso` refusal, first sentence for first sentence (#6212). A native `$dateTrunc` lowering needs the engine's bucket labels, a published capability record and `date-bucket-parity.test.ts`, so it is its own change. A `groupBy` entry that is neither half of the union is refused with INVALID_QUERY/400. The suite is server-free (the real-mongod halves are opt-in since #5517): it drives the EMITTED pipeline through a strict in-process evaluator that refuses every shape it does not model, and each pre-fix lowering is replayed through it and must fail the case it broke. It holds the LOWERING to the table and does not answer "does MongoDB agree?" — recorded as open rather than implied. The `driver-mongodb` x `AGGREGATION_CASES` DEBT row goes with it, in this PR, per the gate's rule. driver-memory is untouched: it stays under the #5499 freeze and keeps its row, so #6814 stays open for its half. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PeD85qA1JNLXodmYdnu4Zm
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
Contributor
📓 Docs Drift CheckThis PR changes 2 package(s): 107 hand-written doc(s) reference the affected code and may need an implementation-accuracy re-verification:
⛔ 7 release-owned page(s) also reference the affected code. These are read-only:
|
huangyiirene
marked this pull request as ready for review
August 11, 2026 05:35
Uh oh!
There was an error while loading. Please reload this page.
This was referenced Aug 11, 2026
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 freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Fixes#6850
Part of #6814 — the driver-mongodb half only.
driver-memorystays under the #5499 freeze and keeps itsAGGREGATION_CASESDEBT row, so #6814 must stay OPEN after this lands.Authorized by the maintainer's 2026-08-11 partial unfreeze on #5499 (comment
5249019855), which is the restart condition both findings-round gradings named.What was wrong — three divergences, all of which ANSWER
driver-mongodbis now enrolled in the sharedAGGREGATION_CASESstandard (@objectstack/spec/data). Clearing the cell fixed three things. None of them ever surfaced as an error, which is the point: each returned plausible rows.1. A structured
GroupByNodehad no lowering at all (#6850).GroupByNodeSchemadeclares a union — a bare field name, or{ field, dateGranularity?, alias? }.buildAggregationPipelineannotatedgroupByasstring[]and didgroupId[field] = '$' + field. A structured node is an object in that loop, so the$group._idkey became the literal"[object Object]"and its value the field path"$[object Object]", which matches nothing.MongoDBDriver.aggregatepassed the value through(query as any).groupBy, which is why the declared union never met that annotation attsc.Both sides now spell
GroupByNode[]. The_idkeys onalias ?? fieldwith the field as its value, so the projected column is renamed and the grouping does not move — the rule #6401 converged the three SQL faces onto, and the onein-memory-aggregation.tshas always applied. The bare-string spelling emits exactly what it emitted before (pinned).Verified the type guard is now real rather than nominal: a probe assigning
DriverQuery['groupBy']tostring[]failstscwith the full union, so the next drift between the two is a compile error.2.
count_distinctcounted NULL as a distinct value (#6814).$addToSetkeeps an explicitnull, so a nullable column sized one higher thanCOUNT(DISTINCT col)— 3 where the standard says 2.3.
count(col)counted ROWS, not values — named by neither card. Measured while writing the suite: thecountarm ignoredfieldand emitted{ $sum: 1 }for both spellings, socount(stage)answered 6 — the numbercount(*)already has — where the standard says 4. This is the case-set doing exactly what its own note says it is for ("three different numbers over one column"). It now counts non-null values, with a missing field reading as null (the SQL reading: an absent value is NULL, there is no third state).Decisions worth reviewing
dateGranularityis REFUSED, not implemented and not ignored.NOT_IMPLEMENTED/ 501 in the ADR-0112 envelope —driver-sql's anddriver-turso's refusal, first sentence for first sentence (#6212). The reasoning is measured, not assumed:MongoDBDriver.supportspublishes noqueryDateGranularity, so the engine buckets every granularity in memory and never pushes a bucketed node down here; the refusal fires only for a caller that reached the builder directly — who previously got the"[object Object]"grouping. A native$dateTrunclowering is buildable and this is not a verdict that it cannot be: it has to emit the engine's bucket labels ('2026-01','2026-Q1', ISO'2026-W03'), publish the capability record, and be held todate-bucket-parity.test.ts— a card of its own. Silently ignoring a declared key was the one option not available.The
count_distinctnull exclusion is inpostProcessAggregation, not in the pipeline. Both server-side spellings the finding sketched were measured against that and not taken:$ne: nullbefore the$addToSet— as a$matchit drops the row from the whole pipeline, so acount(*)orsum()sharing it would silently lose the null rows too. It is correct only for a pipeline carrying nothing else, which this builder cannot assume.$sizeof a$setDifferenceagainst[null]— sound, and it would size server-side instead of shipping the array. It needs a$projectstage the builder does not emit when there is nogroupBy, i.e. a pipeline shape change no suite here can execute. Worth doing when this cell gains a live half; the existingArray.isArrayguard would let it fall through harmlessly.Excluding it in post-processing is exact, needs no unobserved server semantics to be true, and is pinned directly.
The suite, and what it does not claim
mongodb-aggregation-translation.test.ts— the server-free half #5517 requires, mirroringmongodb-filter-logic-translation.test.ts. It drives the emitted pipeline through a strict in-process evaluator of the stages the builder emits; every stage, accumulator and expression it does not model is a thrown error, never a tolerated no-op. Its discrimination is proved rather than asserted: each of the three pre-fix lowerings is replayed through it and must FAIL the case it broke, including the"[object Object]"$group._id— executed, it collapses all six rows into one group under a column literally named[object Object], holding null.It also runs every grouped case a second time through the plain structured spelling
{ field: 'region' }— the half of the union that no capability bit guards and that the engine pushes down to every driver.What it deliberately does not answer: whether a real mongod agrees.
$cond/$ifNull/$addToSetare modelled from the documentation, not observed — this environment cannot fetch a mongod binary (proxy 403), and a suite nobody has executed is a claim, not a check. That bound is written into the suite header, into theAGGREGATION_CASESenrolment list, and into the ledger note, because a green cell reads as more than it is.Ledger and table
scripts/check-driver-conformance.mjs: thedriver-mongodb×AGGREGATION_CASESDEBT row is deleted in this PR, with the suite that replaces it — the gate fails either half alone.driver-memory's row is untouched. The section note records that the deleted row was right about what it measured and under-counted what it had not (defect 3), which is the argument for the suite rather than against the ledger.packages/spec/src/data/aggregation-conformance.ts: comment-only —driver-mongodbmoves from the DEBT table to the enrolled list, with the bound above stated.Gates run locally
pnpm --filter @objectstack/driver-mongodb testpnpm --filter @objectstack/driver-mongodb typecheckpnpm --filter @objectstack/spec typecheckpnpm check:driver-conformance(+--self-test)eslint --no-inline-configon every changed file--filter @objectstack/driver-mongodb...)The rest of the lint farm is left to CI.
Deliberately untouched
packages/drivers/driver-memory— frozen under [裁决] driver-memory / driver-mongodb 投入冻结 —— 维护者 2026-08-05 口径（跨单锚点） #5499; the 2026-08-11 ruling lifted the freeze fordriver-mongodbalone.string_aggarm — retired fromAggregationFunctionat [spec] AggregationFunction 声明 8 个,SQL 族只实现 5 个 —— count_distinct / array_agg / string_agg 按 ADR-0049 enforce-or-remove 定去留 #6188 and unreachable through a spec-valid request. [drivers] driver-memory / driver-mongodb disagree with the SQL family oncount_distinct— the two AGGREGATION_CASES cells #6409 left open #6814 names it as not-a-defect; cleaning it is not this card.(query as any).aggregations || (query as any).aggregateread — that cast also carries the undeclaredquery.aggregatelimb [finding]query.aggregate与agg.func是 spec 从未声明的宽容别名,唯一书写者是驱动自己的 fixture(#4984 家族) #6321 deleted from the SQL faces. A separate convergence, noted in place.🤖 Generated with Claude Code
https://claude.ai/code/session_01PeD85qA1JNLXodmYdnu4Zm
Generated by Claude Code