test(conformance): take the schema fingerprint from @akalforge/pg-conformance - #196
Merged
Merged
Conversation
…formance Four separate definitions of "are these schemas the same?" had accumulated across this project and SupaForge, and they had drifted. One of them called two schemas identical when a view's predicate had been inverted, a function body had been replaced, or a trigger had moved from AFTER INSERT to BEFORE UPDATE. This was the last of the four still carrying its own. The shared query is stronger than the one it replaces. It adds view and materialised view bodies, routine bodies, trigger definitions, identity sequence options, per-column generated/storage/compression/collation, constraint validity and deferrability, inheritance and comments — none of which this harness could previously see. Conformance still exits 0 on PostgreSQL 17 with the same 15 passes and 5 known unmodelled kinds, so the extra coverage costs nothing today and closes the gap that let the identity bug sit in the baseline as thirteen separate "wrong output" entries. Mismatch reporting improves as a side effect. It used to inspect only $fp['columns'], so a difference in a view body or a partition bound printed "schema mismatch" with nothing to point at. The fingerprint is now one entry per line, so the report names the objects that actually differ. getSchemaFingerprint treats a missing `fingerprint` column as a fault rather than as "no differences": an empty fingerprint would make every comparison succeed, and the failure would look exactly like a clean run. The package is consumed from npm rather than Composer. It ships a PHP accessor, so composer.json is untouched and no registry credentials are needed — the conformance job runs `npm ci` and requires the file by path. Pinned ~0.0.1, because a caret on a 0.0.z version permits no updates at all and npm silently rewrites a tilde to a caret on install. Verified on live PostgreSQL 17, and the full suite re-run because the harness is shared: Unit 749, MySQL 17, Postgres 47, SQLite 16. Co-Authored-By: Claude <noreply@anthropic.com>
|
jasdeepkhalsa
added a commit
that referenced
this pull request
Aug 31, 2026
Closes the loop on the shared conformance data introduced in #196. ## Why this is needed `npm ci` installs **strictly from the lockfile** and ignores the range in `package.json`. So the shared package auto-publishing patches, plus `~0.0.1` here, achieves nothing on its own — the release succeeds and reaches nobody. Something has to move the lockfile, and dropping `npm ci` would trade reproducible installs for it. ## Dependabot, not Renovate Renovate needs a GitHub App installed against the org. Dependabot is already available and does the same job here. **Allow auto-merge** also had to be enabled on the repo — that is a permission that lets a PR be *marked* merge-on-green, not a behaviour that merges anything by itself. ## Scoped twice, deliberately `dependabot.yml` allows only `@akalforge/pg-conformance` — the sole npm dependency in what is otherwise a Composer project — and the workflow independently refuses anything else that reaches it. Only **patch** updates auto-merge. A minor or major means the fingerprint changed shape, which is exactly when a human should look. ## Nothing merges blind `--auto` waits for every required check, including the conformance matrix across PostgreSQL 16, 17 and 18. An update that broke this harness leaves the PR open and red rather than landing — and the upstream package already tests itself against PostgreSQL 14–18 before publishing, so a bad patch has to get past both. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
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
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.
Four separate definitions of "are these schemas the same?" had accumulated across this project and SupaForge, and they had drifted. This is the last of the four still carrying its own.
Depends on
@akalforge/pg-conformance@~0.0.1.What the shared query adds
The one it replaces could not see any of this:
That blind spot is not theoretical here. The identity-options bug fixed in #195 sat in the baseline as thirteen separate
wrong_outputentries precisely because the fingerprint could not see sequence options and so mis-attributed one root cause across many patterns.Conformance still exits 0 on PostgreSQL 17 with the same 15 passes and 5 known unmodelled kinds, so the extra strength costs nothing today.
Better mismatch reports
Reporting used to inspect only
$fp['columns'], so a difference in a view body or a partition bound printedschema mismatchwith nothing to point at. The fingerprint is now one entry per line, so the report names what actually differs:A silent failure guarded against
getSchemaFingerprinttreats a missingfingerprintcolumn as a fault rather than as "no differences". An empty fingerprint would make every comparison succeed, and a fully green run would be indistinguishable from a correct one.Why npm and not Composer
The package ships a PHP accessor, so
composer.jsonis untouched and no registry credentials are involved — the conformance job runsnpm ciand requires the file by path. DBDiff uses node for nothing else, sopackage.jsonis explicitly test-only tooling.Pinned
~0.0.1, not^0.0.1: under semver a caret on a0.0.zversion permits no updates at all, andnpm installsilently rewrites a tilde to a caret via its defaultsave-prefix.Verification
Run against live PostgreSQL 17, and the full suite re-run because the harness is shared rather than Postgres-only:
🤖 Generated with Claude Code