ejected

ejected — ejected from a sqlb schema

This package was written by sqlb eject. It imports pgx and the standard library, and nothing else — no sqlb, no huma, no router. Deleting sqlb from go.mod after taking this is a supported end state, which is the entire point of the command.

Nothing here is precious. Edit it, delete the parts you do not serve, or keep running sqlb eject -check in CI for as long as you want the exit kept current — and drop that gate on the day you stop.

What is here

The endpoints

What came out whole

• CRUD and list, at the same paths, with the same status codes and the same JSON envelope — items, page, per_page, has_more, and total when ?count=exact was asked for.
• The filter operators that are one SQL fragment each: eq, ne, lt, lte, gt, gte, in, nin, isnull, notnull, between, like, ilike, contains, startswith, endswith — and the bare ?column=value shorthand for equality.
• ?sort, ?search and ?page/?per_page, with the ceilings the schema declared.
• Capabilities as refusals. A column that never declared Filterable cannot be filtered here either, and the rejection lists the ones that can be. That is a security property, not a convenience: a column left out of the grammar cannot be probed through it. Hidden columns are absent from the column table entirely.
• The error shape. RFC 9457 problem documents, with the allowed list on each detail, so a client's error handling does not change.
• The constraint mapping. A duplicate unique value is still a 409, and a foreign-key, check or not-null violation still a 422, classified off SQLSTATE class 23 exactly as before — so a retry loop keyed on 409 keeps working. The detail text is generic where the API named the resource; the status, which is what clients branch on, is identical.
• The request budgets. MaxFilters and MaxSortTerms come from the schema; the list cap (100 values in one in/nin) and the value-length cap (256 bytes) are constants at the top of support.go, edit them there. ?search escapes % and _ in the term, so a search for a literal percent sign still matches literally.
• The obligation. A table that declared Scoped or SoftDelete refuses to register without a Confine hook, and a scoped table with a create endpoint refuses without an Assign hook. Startup errors, exactly as before.

What did not come out, by name

• Keyset pagination (?cursor). Offset paging is here; the cursor is not. ?cursor is refused with a message saying so rather than ignored.
• Sparse projections (?select). Every read returns the full row.
• Relation expansion (?expand). One statement that joined a target and built a JSON object for it was the engine, not the surface. Fetch the related row from its own endpoint.
• The JSON filter tree (?filter=). Arbitrary and/or/not nesting is gone; the query-parameter operators are not. Negation survives only where an operator spells it — ne, nin, notnull — so a filter that leaned on a not group has to be restated as one of those or moved into the handler.
• Array and document operators (has, hasany, hasall, hasdoc, and their negations nhas, nhasany, nhasall, nhasdoc). The columns are still there and still returned; the containment operators are not.
• The OpenAPI document, and with it the generated TypeScript, Dart and CLI clients. They were emitted from the schema, and the schema is what you are leaving. The wire format they speak is unchanged, so a committed client keeps working — it just has no generator behind it any more.
• Hooks other than the two seams above. BeforeCreate, AfterUpdate and the rest were registrations on a runtime that is no longer here; the handler is a function, so the code that ran in a hook goes in it.
• Transactions across handlers. Each handler runs one statement. DB is an interface a pgx.Tx satisfies, so wrapping is yours to arrange.
• Type overrides. The models use the default type mapping; a column that had a Types override in the generator has its default Go type here. Enums are plain strings, and the CHECK constraint in schema.sql is what still enforces the value set.

Notes for this schema

• authors expanded org through ?expand. The foreign key is still returned; the joined row is not.
• posts expanded author through ?expand. The foreign key is still returned; the joined row is not.

Wiring it up

mux := http.NewServeMux()
pool, err := pgxpool.New(ctx, dsn)
if err != nil {
	log.Fatal(err)
}
if err := ejected.Register(mux, pool, ejected.Options{
	Posts: ejected.PostsHooks{
		// Required: deleted_at declares a soft delete.
		Confine: func(r *http.Request) ([]ejected.Condition, error) {
			return []ejected.Condition{
				{Column: "deleted_at", Op: ejected.OpIsNull},
			}, nil
		},
	},
}); err != nil {
	// A resource whose schema declared Scoped or SoftDelete fails here rather
	// than serving unconfined rows.
	log.Fatal(err)
}
log.Fatal(http.ListenAndServe(":8080", mux))