Add dataclass and pydantic support with simplified implementation

[termoshtt]

This PR adds support for Python dataclasses and Pydantic models, building upon the work in #23 with significant architectural improvements.

Overview

This implementation extends deserialization capabilities to support:

• Python dataclasses (standard library, Python 3.7+)
• Pydantic models (popular data validation library)

Both types are now seamlessly converted to Rust structs via from_pyobject.

Key Differences from #23

Simplified Architecture

• Removed caching mechanism: Eliminated once_cell dependency for simpler, more maintainable code
• Separate modules: Split into dataclass.rs and pydantic.rs for better organization
• Unified API pattern: Both modules follow the same *_as_dict() pattern returning Option<PyDict>

Improved Error Handling

• Replaced .expect() calls with proper ? error propagation
• Added graceful fallback when pydantic is not installed (returns None instead of panicking)
• Added debug logging for better troubleshooting

Better Testing

• Test matrix: CI now tests all combinations of pydantic presence and abi3 feature
• Graceful test skipping: Pydantic tests automatically skip when pydantic is not installed
• Comprehensive documentation: Added detailed module-level docs for each test suite

API Changes

• Updated to PyO3 0.27.0 (from 0.26.0)
• Replaced deprecated downcast() with cast() throughout
• Replaced deprecated DowncastError with CastError

Implementation Details

Dataclass Support (src/dataclass.rs)

pub fn dataclass_as_dict<'py>(
    py: Python<'py>,
    obj: &Bound<'py, PyAny>,
) -> PyResult<Option<Bound<'py, PyDict>>>

• Uses dataclasses.is_dataclass() to check if object is a dataclass
• Converts to dict via dataclasses.asdict()
• Returns None for non-dataclass objects

Pydantic Support (src/pydantic.rs)

pub fn pydantic_model_as_dict<'py>(
    py: Python<'py>,
    obj: &Bound<'py, PyAny>,
) -> PyResult<Option<Bound<'py, PyDict>>>

• Attempts to import pydantic module (returns None if not available)
• Checks if object is instance of BaseModel
• Converts to dict via model_dump() method
• Returns None for non-pydantic objects

Deserialization Flow (src/de.rs)

The deserializer now checks types in this order:

1. Built-in types (dict, list, tuple, string, bool, int, float)
2. Dataclass → convert via dataclass_as_dict()
3. Pydantic model → convert via pydantic_model_as_dict()
4. Objects with __dict__ attribute
5. Objects with __slots__ attribute

Testing

New test files organized by Python object type:

• tests/python_custom_class.rs - Custom classes with __dict__
• tests/python_dataclass.rs - Dataclass tests (simple and nested)
• tests/python_pydantic.rs - Pydantic model tests (auto-skips if not installed)

All tests follow the pattern:

1. Define Python class/model
2. Create instance
3. Deserialize to Rust
4. Compare with expected Rust value

Why Remove Caching?

The caching mechanism from #23 added complexity with minimal benefit:

• Module imports are already cached by Python's import system
• The performance gain was negligible for typical usage
• Simpler code is easier to maintain and understand
• Reduces dependencies (once_cell removed)

Migration from #23

If you were using the internal API from #23:

• is_pydantic_base_model() → pydantic_model_as_dict()
• Return type changed from bool to Option<PyDict>
• Module renamed: py_module_cache → pydantic

The public API (from_pyobject, to_pyobject) remains unchanged.

Version Bump

Version bumped to 0.8.0 to reflect the new features and breaking changes in internal APIs.

[Copilot]

Pull Request Overview

This PR adds support for Python dataclasses and Pydantic models to the deserialization pipeline, enabling seamless conversion of these Python types to Rust structs. The implementation includes infrastructure improvements, better error handling, and comprehensive testing.

Key changes:

• Added dataclass and Pydantic model deserialization support via new dataclass.rs and pydantic.rs modules
• Updated to PyO3 0.27.0, replacing deprecated downcast() with cast() and DowncastError with CastError
• Enhanced CI testing with matrix strategy covering pydantic presence and abi3 feature combinations

Reviewed Changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/dataclass.rs	New module for dataclass-to-dict conversion using dataclasses.asdict()
src/pydantic.rs	New module for Pydantic model-to-dict conversion using model_dump() with graceful fallback
src/de.rs	Updated deserializer to check for dataclass/Pydantic types before __dict__ fallback; replaced deprecated downcast() with cast()
src/error.rs	Replaced deprecated DowncastError with CastError
src/lib.rs	Added module declarations for dataclass and pydantic
tests/python_dataclass.rs	New test suite for dataclass deserialization (simple and nested)
tests/python_pydantic.rs	New test suite for Pydantic model deserialization with auto-skip when unavailable
tests/python_custom_class.rs	New test suite for custom Python classes, extracted from check_revertible.rs
tests/check_revertible.rs	Added comprehensive documentation and removed custom class test (moved to dedicated file)
tests/to_json_to_pyobject.rs	Added comprehensive documentation explaining cross-validation approach
Cargo.toml	Version bump to 0.8.0, updated PyO3 to 0.27.0, added log dependency, removed abi3-py38 feature
.github/workflows/rust.yml	Added test matrix for pydantic and abi3 combinations

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.