feat(python): PostgREST/Supabase query dialect in generated services (parity with TS #36/#56-#60)

[cultron]

Ports the @apso/crud PostgREST/Supabase dialect to the Python (FastAPI + SQLAlchemy) code generator. Generated services now accept the same PostgREST URL conventions the TypeScript services do. The nestjsx dialect is unchanged.

What changed (generated output)

utils/query.py gains a second parser path alongside the incumbent nestjsx parser, plus a dialect detector. The router reads the X-Crud-Dialect header and shapes the response per dialect. The service exposes both retrieval methods.

• src/lib/templates/python/utils/query.eta — PostgREST parser, dialect detection, postgrest-aware QueryBuilder
• src/lib/templates/python/routers/router.eta — header-driven dialect selection + response shaping
• src/lib/templates/python/services/service.eta — get_many_list (bare array) alongside get_many (envelope)
• src/lib/templates/python/utils/init.eta — exports
• test/lib/generators/python.test.ts — 9 new generator tests

Operators / behaviors covered

• select=a,b including alias:col output-key rename (CLI first pivot: new command surface, architecture refactor, tests #57), col::type cast stripping, and rel(cols) embeds (mapped to selectinload eager loads)
• col=op.value filters: eq / neq / gt / gte / lt / lte / like / ilike / in / is and not.<op>, with the TS operator mapping and *→% like/ilike wildcards; is.null / is.true / is.false; in.(1,2,3) with quoted-value support
• or=(...) / and=(...) logical groups, nested or(...)/and(...) (Plat 1269 #56)
• order=col.asc|desc[.nullsfirst|nullslast]
• limit / offset

Dialect detection (mirrors dialect.ts)

1. X-Crud-Dialect header wins (postgrest|nestjsx, any other value → 400)
2. Param-shape: select/order or a bare col=op.value → postgrest; fields/filter/join/sort/s/or/per_page → nestjsx; limit/offset/page/cache neutral
3. Both families present → 400 (refuse to guess)
4. Otherwise default to nestjsx (incumbent — existing requests byte-identical)

Response shaping

• postgrest: bare JSON array, always (even with limit/offset, which just window the array) — feat: MCP server + CLI modernization #58
• nestjsx: {data,count,total,page,pageCount} envelope — unchanged
• Unknown filter/select/order column on the postgrest path → 400 (QueryParseError → HTTPException 400), not a 500 — Feature: auto-detect codegen/schema issues from real usage and (opt-in) auto-file GitHub issues #60

Deferred

• cs / cd / ov / fts and range operators (Plat 1288 #54, as in TS) — rejected clearly with a parse error rather than silently accepted
• Embedded resource filtering (rel.col=op.value, incl. !inner parent filtering — primaryKeyType: "uuid" generates a non-generated @PrimaryColumn() → every insert fails with null id #59): the Python join path uses selectinload (a separate SELECT) and cannot push a predicate into a JOIN ON clause, so embedded filters are rejected with a 400. select embeds still eager-load the relation. Postgrest filters are scoped to root columns.

Tests

• Generator suite: 330 passed / 25 suites (NODE_OPTIONS='--experimental-vm-modules' jest), including 9 new PostgREST-dialect tests
• Semantic battery: the generated query.py was loaded in Python and run against real SQLAlchemy compilation (eq, not/in/is, like/ilike *→%, or=()/nested and=(), order + nulls, select alias projection, unknown-column 400, bare-array limit/offset windowing, deferred-op rejection, embedded-filter rejection, dialect detection incl. header override + collision) — all green. Generated query.py, router.py, service.py all parse.

Do not merge — merging publishes the CLI. Left green for review.

🤖 Generated with Claude Code