fix: add 21 plpgsql deparser edge case fixtures and fix 7 schema_rename failures

[pyramation]

Summary

Adds 21 new PL/pgSQL deparser test fixtures to prevent regression of the END; bug fix and improve coverage of previously untested statement types. Also fixes two deparser bugs that caused 7 pre-existing plpgsql_schema_rename-* fixture failures — all 225 fixtures now pass round-trip testing with no allowlist.

5 groups of fixtures (Tests 17–37):

Group	Fixtures	Priority	What's tested
Nested Block Compositions	17–22	CRITICAL	END; followed by RETURN, IF, RAISE, PERFORM, assignment; labeled blocks
Blocks Inside Control Structures	23–25	HIGH	Block+exception inside IF, LOOP, CASE
Deep Nesting & Sequential Blocks	26–28	HIGH	Sequential blocks, triple nesting, block inside exception handler
Untested Statement Types	29–35	MEDIUM	FOR loops (integer/query/labeled), RETURN NEXT, RETURN QUERY, ASSERT, CALL
Real-World Patterns	36–37	CRITICAL	Permission bitnum trigger (the original bug-exposing function), multi-step sign-in

Bug fixes

Two deparser bugs were causing 7 plpgsql_schema_rename-* fixtures to fail on main. Instead of allowlisting them, both root causes were fixed:

1. cleanPlpgsqlTree query normalization (test-utils/index.ts): The transform function applied normalizeQueryWhitespace to all keys named query, but PLpgSQL_stmt_return_query.query is an object (wrapping PLpgSQL_expr), not a string. normalizeQueryWhitespace returned non-strings unchanged without recursing, so the inner PLpgSQL_expr.query string was never normalized — causing false AST mismatches on multi-line SQL. Fixed by making the query handler recurse into objects via transform(val, cleanProps).

2. getDiagItemKindName string/number mismatch (plpgsql-deparser.ts): The parser returns diag_item.kind as a string (e.g. "ROW_COUNT"), but the deparser only handled numeric DiagItemKind enum values — so GET DIAGNOSTICS deleted_count = ROW_COUNT was deparsed as GET DIAGNOSTICS deleted_count = ; (empty kind). Added a string-name lookup path alongside the existing numeric switch. Four pre-existing snapshots were updated to reflect the corrected output.

Review & Testing Checklist for Human

• Verify cleanPlpgsqlTree recursion fix is safe: The new query handler in cleanProps recurses into objects when query is not a string. Confirm this doesn't accidentally normalize non-query objects or cause infinite recursion (it delegates back to transform with the same props, which only recurses on object/array types). Note: this fix is in the test comparison utility, not the deparser itself — multi-line SQL may still be formatted differently after deparsing, but the round-trip AST comparison now correctly normalizes both sides.
• Verify getDiagItemKindName string values match parser output: The string map keys ('ROW_COUNT', 'PG_CONTEXT', etc.) must exactly match what @libpg-query/parser returns in PLpgSQL_diag_item.kind. Cross-check against the parser's protobuf definitions if possible.
• Spot-check updated snapshots: 4 snapshots changed across hydrate-demo.test.ts.snap, schema-rename-mapped.test.ts.snap, and plpgsql-pretty.test.ts.snap — all diffs should only show = ; → = ROW_COUNT;.
• RETURN NEXT limitation: Fixture 32 uses bare RETURN NEXT (with OUT params) instead of RETURN NEXT rec because the deparser drops the expression from RETURN NEXT <variable>. Decide if you'd prefer a deliberately-failing fixture to document this gap instead.
• Regenerated generated.json: The diff includes 12 plpgsql_schema_rename-* entries — these are NOT new fixtures from this PR; they were already in __fixtures__/plpgsql/plpgsql_schema_rename.sql and are now picked up by the regenerated JSON. Verify no fixtures were accidentally removed.

Recommended test plan: cd packages/plpgsql-deparser && pnpm fixtures && npx jest

Notes

• All 225 generated fixtures pass round-trip testing (0 known failures)
• 83 total tests pass across 8 test suites (40 snapshot tests in deparser-fixes.test.ts, 6 in plpgsql-deparser.test.ts, plus hydrate, schema-rename, pretty tests, etc.)
• Each snapshot test in deparser-fixes.test.ts duplicates its SQL inline rather than loading from generated.json — this matches the existing pattern

Link to Devin session: https://app.devin.ai/sessions/7ac7bb4b376e4458b088c01e61913be0
Requested by: @pyramation

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring