fix: accept JSON schemas in structured output parser

[Iams4kura]

Summary

• Fixes: Passing a supported JSON Schema dictionary into structured-output generation raises TypeError before any model request instead of parsing the returned dictionary.
• Root cause: get_structured_output_parser documents and types dictionaries as valid schemas but calls the class-only issubclass builtin before dispatching dictionary schemas to _dict_output_parser.

Regression evidence

• Before: PYTHONDONTWRITEBYTECODE=1 uv run --frozen pytest -o addopts='' -p no:cacheprovider tests/utils/output_parser_test.py::test_structured_output_parser_accepts_json_schema -q exited 1

• After: PYTHONDONTWRITEBYTECODE=1 uv run --frozen pytest -o addopts='' -p no:cacheprovider tests/utils/output_parser_test.py::test_structured_output_parser_accepts_json_schema -q exited 0

Verification

• PYTHONDONTWRITEBYTECODE=1 .venv/bin/pytest -o addopts='' -p no:cacheprovider tests/utils/output_parser_test.py -q
• PYTHONDONTWRITEBYTECODE=1 .venv/bin/pytest -o addopts='' -p no:cacheprovider tests/test_json_scraper_graph.py -q
• .venv/bin/ruff check scrapegraphai/utils/output_parser.py tests/utils/output_parser_test.py
• .venv/bin/black --check scrapegraphai/utils/output_parser.py tests/utils/output_parser_test.py
• .venv/bin/isort --check-only scrapegraphai/utils/output_parser.py tests/utils/output_parser_test.py
• uv build --no-sources
• offline ChatOpenAI.with_structured_output JSON Schema contract check

Scope

• 2 files changed, +18 / -1 lines

Local full-suite note

I also attempted the repository CI selection, uv run pytest tests/ -m "unit or not integration". It reached 95% with unrelated live-provider, browser, network, and existing assertion failures, then was interrupted when an external proxy request stalled. The deterministic affected suites listed above pass.

[VinciGit00]

Thanks for this — the diagnosis holds up. I reproduced it on pre/beta: get_structured_output_parser({...}) raises TypeError: issubclass() arg 1 must be a class, so _dict_output_parser was unreachable dead code despite the signature advertising Dict[str, Any].

One note on the test before this goes in.

_dict_output_parser is just return x, so assert parser(output) == output is an identity assertion — it holds for any pass-through callable and never checks that the right parser was selected. Swap _dict_output_parser for _base_model_v2_output_parser in the dispatch and the test still passes on the returned value alone; it only fails today because get_structured_output_parser(schema) itself raises before the assertion is reached.

What actually regressed is the dispatch, so I'd assert on that directly:

from scrapegraphai.utils.output_parser import (
    _dict_output_parser,
    get_structured_output_parser,
)


def test_structured_output_parser_accepts_json_schema():
    schema = {
        "title": "Person",
        "type": "object",
        "properties": {"name": {"type": "string"}},
    }

    assert get_structured_output_parser(schema) is _dict_output_parser

That still fails with the TypeError before the fix, and afterwards it pins the behaviour the fix introduces rather than a property of the returned function. Keeping the existing parser(output) == output line alongside it is fine if you want the round-trip covered too.

Worth doing because tests/utils/output_parser_test.py is one of the files in the test-suite.yml run list, so this test is real CI coverage rather than a local-only check.