fix: preserve quoted Literal delimiters during type deserialization - #13055

Related Issues

No issue is linked. Independently reproduced on main at a5432b976f888678b61902a7fb999654a6384ef6.

This follows the Literal support added in #12286. The open #12339 handles Annotated metadata; it does not fix the existing generic-argument and PEP 604 scanners changed here.

Proposed Changes:

Type deserialization counts brackets and splits separators before reaching the quote-aware Literal parser. A bracket inside a string value can therefore be mistaken for type syntax. For example, this public component round trip currently fails:

from typing import Literal
from haystack.components.converters import OutputAdapter

adapter = OutputAdapter(template="{{ value }}", output_type=tuple[Literal["["], int])
restored = OutputAdapter.from_dict(adapter.to_dict())
# Before: DeserializationError: Could not import 'typing.Literal['['], int'
# After: restored.output_type == adapter.output_type

The same problem affects Literal["]|"], nested Callable and Optional types, and PEP 604 unions containing these values.

• Make the two existing delimiter scanners respect quoted strings and escaped quotes/backslashes before counting brackets or splitting commas/pipes
• Preserve previously accepted triple-quoted strings in direct deserialize_type() inputs
• Add focused type round-trip regressions, direct triple-quote compatibility tests, and two OutputAdapter.to_dict()/from_dict() regressions
• Add a Reno release note

The serialized format and deserialization allowlist are unchanged.

How did you test it?

Locally with Python 3.12.14 and Hatch 1.14.2:

• On unchanged main, the existing type-serialization and OutputAdapter suites passed: 217 passed
• Adding the initial 16 regression cases without the production fix produced 15 failed, 1 passed
• After the fix and six triple-quote compatibility cases, the related serialization/component suites passed: 538 passed

hatch run test:unit \
  test/utils/test_type_serialization.py \
  test/utils/test_callable_serialization.py \
  test/utils/test_base_serialization.py \
  test/core/test_serialization.py \
  test/core/test_serialization_security.py \
  test/components/converters/test_output_adapter.py \
  test/components/routers/test_conditional_router.py --no-cov

Also passed:

• hatch run fmt-check across the repository
• hatch run test:types across all 550 source files
• hatch run pre-commit run --files for the four changed files
• hatch run reno lint . and the release-note backtick check
• git diff --check
• An additional local exhaustive check of 28,086 string/bytes/nested/union round trips, including bracket, quote, backslash, comma and pipe combinations

The full unit suite, live integrations, Docker-based license-header CI job, and other Python versions were not run locally. No provider credentials or model downloads were used.

Notes for the reviewer

This PR was fully generated with an AI assistant. AI was used for the investigation, implementation, testing, independent review, and this write-up. The reported checks were run locally. The independent review identified the triple-quoted input compatibility case, which is covered by the added tests.

Checklist

• I have read the contributors guidelines and the code of conduct.
• I have updated the related issue with new insights and changes. (Not applicable: no linked issue.)
• I have added unit tests and updated the docstrings.
• I've used one of the conventional commit types for my PR title: fix:.
• I have documented my code.
• I have added a release note file, following the contributors guidelines.
• I have run pre-commit hooks and fixed any issue.