pg-topo: CREATE TYPE AS RANGE is treated as unsupported

[xmh1011]

Summary

pg-topo currently treats PostgreSQL range type creation as an unsupported statement class. This leaves downstream objects that use the range type with an unresolved type:* dependency, even though packages/pg-topo/docs/architecture.md lists types/domains/collations/sequences under implemented coverage.

This is a separate issue class from #281. #281 is about expression dependencies inside ALTER TABLE subcommands; this one is about missing statement classification/extraction for the CreateRangeStmt AST root.

Minimal repro

Input statements, intentionally shuffled:

create type app.int_range as range (subtype = int4);

create table app.bookings(
  id int primary key,
  span app.int_range not null
);

create schema app;

Actual result on current main

Current main: 9f01826b00b64e7f9628844ecb6f69a07f53c1f8

pg-topo ordered the table before the range type and classified the range statement as UNKNOWN:

{
  "ordered": [
    {
      "class": "CREATE_SCHEMA",
      "sql": "create schema app;"
    },
    {
      "class": "CREATE_TABLE",
      "sql": "create table app.bookings(id int primary key, span app.int_range not null);"
    },
    {
      "class": "UNKNOWN",
      "sql": "create type app.int_range as range (subtype = int4);"
    }
  ],
  "diagnostics": [
    {
      "code": "UNRESOLVED_DEPENDENCY",
      "message": "No producer found for 'type:app:int_range:'."
    },
    {
      "code": "UNKNOWN_STATEMENT_CLASS",
      "message": "Unsupported statement AST root 'CreateRangeStmt'."
    }
  ]
}

Runtime validation then reports the table's range type as missing:

RUNTIME_ASSUMED_EXTERNAL_DEPENDENCY: type "app.int_range" does not exist
sqlstate: 42704
missingObjectKey: type:app:int_range:

Expected result

The range type should be recognized as a supported type producer and ordered before objects that use it:

CREATE_SCHEMA
CREATE_TYPE  -- create type app.int_range as range (...)
CREATE_TABLE -- span app.int_range

No UNKNOWN_STATEMENT_CLASS or UNRESOLVED_DEPENDENCY diagnostic should be emitted for the range type in this repro.

Root cause

classifyStatement does not map the parser's CreateRangeStmt AST root. extractDependenciesForStatement therefore never reaches a range-type extractor, so CREATE TYPE ... AS RANGE does not provide the type:app:int_range object that the table column requires.

Relevant code paths:

• packages/pg-topo/src/classify/classify-statement.ts: CLASS_BY_AST_NODE handles CompositeTypeStmt and CreateEnumStmt as CREATE_TYPE, but not CreateRangeStmt.
• packages/pg-topo/src/extract/extract-dependencies.ts: the CREATE_TYPE branch handles enum/composite-like providers through existing AST roots, but has no CreateRangeStmt extraction path.

Suggested fix direction

1. Classify CreateRangeStmt as CREATE_TYPE.
2. Add extraction for CreateRangeStmt so it provides the created range type stable ref, e.g. type:app:int_range.
3. Add dependencies for:
   • the containing schema;
   • user-defined subtype types, when present;
   • range support functions/options such as canonical and subtype_diff, if they are schema-qualified or otherwise resolvable from the AST.
4. Add focused pg-topo tests covering:
   • simple CREATE TYPE ... AS RANGE (subtype = int4) before a table column that uses it;
   • a range type with support functions, if the AST exposes those options cleanly.

Local validation

I reproduced this twice with the focused local probe below. Both runs produced the same UNKNOWN_STATEMENT_CLASS / UNRESOLVED_DEPENDENCY diagnostics.

./node_modules/.bin/bun run --cwd packages/pg-topo scripts/run-tests.ts \
  --test-name-pattern "CREATE TYPE AS RANGE" \
  test/deep-probe.test.ts

[xmh1011]

@avallete I can take this one as a small pg-topo coverage PR.

Proposed fix plan

This should be a classifier + extractor addition for the existing type object family.

1. Classify CreateRangeStmt as CREATE_TYPE.
2. In the CREATE_TYPE extraction branch, add a CreateRangeStmt path that provides the created range type ref, e.g. type:app:int_range.
3. Add a schema dependency when the range type is schema-qualified.
4. Inspect range DefElem options and add dependencies where the AST gives structured refs:
   • subtype -> referenced type, ignoring built-in types through the existing built-in filtering;
   • canonical / subtype_diff -> referenced function refs;
   • subtype_opclass / collation-related options only if they are resolvable without guessing.
5. Keep the extractor best-effort and catalog-independent, consistent with pg-topo's static-analysis model.

Regression coverage I would add

I would add focused pg-topo tests for:

• CREATE TYPE app.int_range AS RANGE (subtype = int4) ordered before a table column using app.int_range;
• schema-qualified range type creation requiring CREATE SCHEMA app first;
• range support functions, e.g. canonical = app.canonical_range or subtype_diff = app.diff, ordered after those functions when the parser exposes the function refs cleanly;
• no UNKNOWN_STATEMENT_CLASS and no unresolved type:app:int_range diagnostic.

This is distinct from #281: no ALTER TABLE expression extraction is involved here; the missing AST root is CreateRangeStmt.

[avallete]

Sounds good to me, thank's for reporting ! Feel free to take it up.