ci(postgrest-typegen): drop the nightly parity job against postgres-meta

.github/workflows/postgrest-typegen-validate.yml

@@ -10,9 +10,6 @@ on:
     paths:
       - "packages/postgrest-typegen/**"
       - ".github/workflows/postgrest-typegen-validate.yml"
-  schedule:
-    - cron: "0 6 * * *" # nightly drift check against real postgres-meta
-  workflow_dispatch: {}
 
 defaults:
   run:
@@ -21,10 +18,9 @@ defaults:
 jobs:
   test:
     name: Types, lint, build, test
-    if: github.event_name == 'push' || github.event_name == 'pull_request'
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
       - name: Install bun
         run: curl -fsSL https://bun.sh/install | bash
       - name: Add bun to PATH
@@ -35,92 +31,3 @@ jobs:
       - run: bun run knip
       - run: bun run build
       - run: bun run test
-
-  parity-check:
-    name: Parity vs. real postgres-meta
-    # Nightly/manual only, never on push or pull_request: postgres-meta can
-    # drift for reasons unrelated to any change here, and that shouldn't
-    # block or flake unrelated PRs (same pattern as capability-matrix's
-    # "Tier 2" reference checks).
-    if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
-    runs-on: ubuntu-latest
-    services:
-      postgres:
-        image: postgres:15-alpine
-        env:
-          POSTGRES_USER: postgres
-          POSTGRES_PASSWORD: postgres
-          POSTGRES_DB: postgres
-        ports:
-          - 5432:5432
-        options: >-
-          --health-cmd pg_isready
-          --health-interval 5s
-          --health-timeout 5s
-          --health-retries 10
-    env:
-      DATABASE_URL: postgresql://postgres:postgres@localhost:5432/postgres
-    steps:
-      - name: Checkout supabase/sdk
-        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
-      - name: Checkout supabase/postgres-meta
-        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
-        with:
-          repository: supabase/postgres-meta
-          path: postgres-meta
-      - name: Install bun
-        run: curl -fsSL https://bun.sh/install | bash
-      - name: Add bun to PATH
-        run: echo "$HOME/.bun/bin" >> "$GITHUB_PATH"
-      - run: bun install --frozen-lockfile
-      - name: Load shared fixture schema
-        run: bun scripts/load-fixtures.ts
-      - name: Generate postgrest-typegen output
-        run: |
-          mkdir -p /tmp/ours
-          for lang in typescript go python swift; do
-            bun scripts/print-generated.ts "$lang" > "/tmp/ours/$lang.txt"
-          done
-      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
-        with:
-          node-version: "22"
-      - name: Install postgres-meta dependencies
-        working-directory: postgres-meta
-        run: npm ci
-      - name: Generate postgres-meta output
-        working-directory: postgres-meta
-        env:
-          PG_META_DB_URL: postgresql://postgres:postgres@localhost:5432/postgres
-        run: |
-          mkdir -p /tmp/theirs
-          for lang in typescript go python swift; do
-            PG_META_GENERATE_TYPES="$lang" node --loader ts-node/esm src/server/server.ts > "/tmp/theirs/$lang.txt"
-          done
-      - name: Canonicalize postgres-meta's TypeScript output
-        # postgres-meta formats with prettier; this package defaults to
-        # oxfmt, which has different line-wrap heuristics (e.g. printWidth is
-        # matched at 80, but wrapping decisions for nested object types can
-        # still differ). Reformatting postgres-meta's output through our own
-        # formatter first means the diff below reflects content, not which
-        # formatter produced it.
-        run: bun scripts/canonicalize-typescript.ts /tmp/theirs/typescript.txt > /tmp/theirs/typescript.canonical.txt
-      - name: Compare outputs
-        # Order-independent: sortGeneratorMetadata canonicalizes collection
-        # order (semantic schema+name, not postgres-meta's environment- and
-        # data-dependent heap order) across all four generators, which is a
-        # deliberate improvement over postgres-meta's real behavior, not a
-        # divergence to flag. Comparing declaration BLOCKS (paragraphs
-        # separated by a blank line) as a sorted set, rather than raw bytes,
-        # verifies the two agree on content without being sensitive to that
-        # reordering.
-        run: |
-          mv /tmp/theirs/typescript.canonical.txt /tmp/theirs/typescript.txt
-          normalize() { awk 'BEGIN{RS="";FS="\n"} {gsub(/\n/, "\\n"); print}' "$1" | sort; }
-          failed=0
-          for lang in typescript go python swift; do
-            if ! diff -u <(normalize "/tmp/theirs/$lang.txt") <(normalize "/tmp/ours/$lang.txt"); then
-              echo "::error::$lang output diverges from real postgres-meta"
-              failed=1
-            fi
-          done
-          exit "$failed"

CLAUDE.md

@@ -9,7 +9,7 @@ This is a lightweight monorepo of shared tooling for the Supabase client SDKs. P
 - `packages/capability-matrix/` — the **canonical feature registry** for all Supabase client SDKs — not the SDKs themselves. It defines what features exist (name, description, grouping) across 7 client SDKs (JavaScript, Flutter, Python, Swift, C#, Go, Kotlin). SDKs declare compliance separately in their own repos via `sdk-compliance.yaml` files. The output is a static capability matrix website at https://supabase.github.io/sdk/ showing which features each SDK implements.
 - `packages/dart-symbol-extractor/` — Dart public API symbol extractor used by the Dart compliance workflow.
 - `packages/go-symbol-extractor/` — Go public API symbol extractor used by the Go compliance workflow.
-- `packages/postgrest-typegen/` — introspects a PostgreSQL schema into the `GeneratorMetadata` contract and generates PostgREST types for TypeScript, Go, Python, and Swift. See its own `CLAUDE.md` for architecture and the byte-parity constraint with postgres-meta.
+- `packages/postgrest-typegen/` — introspects a PostgreSQL schema into the `GeneratorMetadata` contract and generates PostgREST types for TypeScript, Go, Python, and Swift. See its own `CLAUDE.md` for architecture and how postgres-meta and the CLI consume it.
 
 ## Commands
 

packages/postgrest-typegen/CLAUDE.md

@@ -65,21 +65,23 @@ bun run format-and-lint # oxfmt + oxlint check
 bun run knip            # unused-code/deps check
 ```
 
-## Byte-Parity Constraint
-
-This package must produce **byte-identical** output to postgres-meta's
-templates until parity is validated and released. Two consequences:
-
-- `prettier` is pinned **exact** to the version postgres-meta's own lockfile
-  currently resolves (not a caret range, and not just "latest"), since a
-  version mismatch reformats generator output and silently breaks parity with
-  the real upstream CLI, not just with this package's own snapshots. Check
-  postgres-meta's resolved version before bumping. Bumping means regenerating
-  every inline snapshot (`bun test --update-snapshots`) and the parity
-  fixtures under `test/parity/expected/`, and reviewing the diff.
-- Don't "improve" template strings or SQL for existing generators without
-  regenerating snapshots/fixtures first. Byte parity first; behavior-changing
-  cleanups (e.g. oxfmt instead of prettier) come later.
+## Relationship to postgres-meta
+
+postgres-meta deleted its own templates and consumes this package for all four
+languages (supabase/postgres-meta#1084, shipped in postgres-meta v0.99.0 on
+2026-08-31), and so does the CLI (supabase/cli#6404). This package is the
+single source of truth for generator output: there is no upstream to stay
+byte-identical with, and there is no cross-check against postgres-meta (its
+output is whatever version of this package its lockfile pins, so such a check
+only ever measured release lag). Consequences:
+
+- Every generator output change is a behavior change for every consumer.
+  Regenerate the inline snapshots (`bun test --update-snapshots`) and the
+  goldens under `test/parity/expected/` in the same PR, review that diff
+  deliberately, and describe the output change in the PR so it reaches the
+  release notes consumers read before bumping.
+- Don't reformat or "clean up" template strings casually; a formatter-only
+  change still churns every consumer's generated files.
 
 SQL literal quoting: the introspection queries use `literal()` from
 `src/introspection/sql/pg-format.ts`, an inlined port of `pg-format@1.0.4`

packages/postgrest-typegen/README.md

@@ -17,19 +17,19 @@ JSON output, not inside `postgrest-typegen`.
 
 The TypeScript, Go, Python, and Swift generators bundled here are a
 **deliberate transition**, not the target architecture for new languages.
-They were ported byte-parity from postgres-meta's own templates so
-`supabase gen types` keeps working unchanged while postgres-meta's copies get
-deprecated in favor of this package. It's decided that all four eventually
-move out to their own SDK repos too (tracked in SDK-1641), not yet scheduled
-— sequenced after postgres-meta's cutover to this package settles.
+They were ported byte-for-byte from postgres-meta's own templates so
+`supabase gen types` kept working unchanged; postgres-meta has since deleted
+its copies and consumes this package instead. It's decided that all four
+eventually move out to their own SDK repos too (tracked in SDK-1641), not yet
+scheduled — sequenced after postgres-meta's cutover to this package settles.
 
 > [!IMPORTANT]
 > **We are no longer accepting contributions that fix the bundled generators.**
 > Because each of the four is moving out to its own language's SDK repository,
 > a fix landed here would have to be made a second time in the destination
-> repository, and it also breaks byte parity with postgres-meta until a release
-> propagates. Please open an issue describing the bug instead, so it can be
-> tracked and carried across the move.
+> repository, and postgres-meta and the CLI only pick it up once they bump
+> their pinned version of this package. Please open an issue describing the
+> bug instead, so it can be tracked and carried across the move.
 
 ## Design
 

packages/postgrest-typegen/knip.json

@@ -1,11 +1,5 @@
 {
   "$schema": "https://unpkg.com/knip@5/schema.json",
-  "entry": [
-    "test/global-setup.ts",
-    "test/**/*.test.ts",
-    "scripts/load-fixtures.ts",
-    "scripts/print-generated.ts",
-    "scripts/canonicalize-typescript.ts"
-  ],
+  "entry": ["test/global-setup.ts", "test/**/*.test.ts"],
   "ignoreBinaries": ["oxfmt", "oxlint"]
 }

packages/postgrest-typegen/test/parity/parity.test.ts

@@ -20,30 +20,22 @@ import { introspect } from "../../src/introspection/index.ts";
 import type { GeneratorMetadata } from "../../src/types.ts";
 
 /**
- * End-to-end parity gate: introspect the shared postgres-meta fixture DB, run
- * all four generators with their default options, and assert the output matches
- * the committed golden files in `expected/`.
+ * End-to-end golden gate: introspect the shared fixture DB, run all four
+ * generators with their default options, and assert the output matches the
+ * committed golden files in `expected/`.
  *
  * The metadata is passed through `sortGeneratorMetadata` first (as every
  * consumer must), so the golden files reflect the canonical, deterministic
- * ordering rather than the database's heap order. The goldens were originally
- * captured from a real postgres-meta CLI run against this fixture DB; only the
- * ordering of order-sensitive collections (Go/Python/Swift emit tables/views
- * in metadata order) is canonicalized — and postgres-meta applies the same
- * sort pass, so the two stay in lockstep. postgres-meta's CLI prints with
- * `console.log`, which appends exactly one trailing newline to the
- * generator's return value — hence the `+ "\n"` below.
+ * ordering rather than the database's heap order. Consumers print the
+ * generator's return value with `console.log`, which appends exactly one
+ * trailing newline, hence the `+ "\n"` below.
  *
- * CAUTION — this only proves parity at the moment a golden was captured. The
- * established workflow for an *intentional* generator change is to
- * regenerate the goldens from this package itself and review the diff, which
- * means a generator change and its golden update land in the same PR: the
- * gate then only proves internal self-consistency (generator output matches
- * its own golden), not agreement with postgres-meta's actual current output.
- * It cannot catch a regression introduced by that same change. Re-verify
- * against a real postgres-meta run periodically, and always whenever
- * postgres-meta's own templates change (e.g. the upstream fixes ported into
- * this package) rather than trusting a self-regenerated golden alone.
+ * The goldens were originally captured from postgres-meta's own templates
+ * before it cut over to this package; since then they are regenerated from
+ * this package whenever a generator changes on purpose. A generator change
+ * and its golden update land in the same PR, so this gate proves output does
+ * not change *unintentionally*, and the reviewed golden diff is where an
+ * intentional change gets scrutinized.
  */
 const FIXTURE_DIR = join(import.meta.dir, "..", "introspection", "fixtures");
 const EXPECTED_DIR = join(import.meta.dir, "expected");