Steven Troxler

This commit **reverts the default behavior** of **CinderX reporting** within `pyrefly`, now only dumping explicitly checked files rather than all analyzed dependencies. This **performance optimization** makes CinderX integration faster and more lightweight by default, which is crucial for rapid development and an OSS-suitable MVP, especially in large projects with extensive dependencies. A new command-line argument, `--cinderx-include-deps`, has been added to `pyrefly/lib/commands/check.rs` to allow users to explicitly include transitive dependencies in the report when needed. The `write_results` function in `pyrefly/lib/report/cinderx/mod.rs` was refactored to accept a slice of handles, enabling this granular control, with corresponding updates to CinderX tests.

This commit introduces a significant **performance optimization** for the **`pyrefly` type checker's CinderX reporting subsystem** by inlining per-module type file generation. Instead of accumulating all data in memory for post-processing, the new `CinderxReporter` now writes individual module JSON and optional text reports as soon as type-checking answers are available, significantly reducing memory footprint and improving execution speed for the `--report-cinderx` feature. This **refactoring** impacts the `check` command and internal **state management** (`TransactionData`), ensuring project-level files are still written at the end of the run while individual module reports are processed efficiently.

This commit introduces a **performance optimization** for the **CinderX reporting** pipeline within Pyrefly. It now stores compact CinderX class metadata directly within the `Solutions` structure, specifically in `pyrefly/lib/alt/answers.rs`, by adding new `CinderxClassRef` and `CinderxClassInfo` types. This **refactoring** allows `class_metadata.json` generation to consume these lightweight, per-module summaries instead of requiring full retained `Solutions` for every defining module. The change significantly reduces memory usage and improves performance for `--report-cinderx` by streamlining post-processing, with a fallback path for tests and non-reporter callers.

This commit introduces a significant **performance optimization** and **memory reduction** for **CinderX reporting** within the **Pyrefly** type checker. It refactors the reporting mechanism to stop retaining full intermediate state, instead processing and writing reports inline during type checking, similar to a previous improvement for Pysa. This change allows modules to evict large intermediate data, substantially reducing RAM consumption and improving performance, particularly when compiling entire applications. The `pyrefly/lib/commands/check.rs` file was adjusted to implement this new data retention strategy.

This commit **adds a new test case** to `pyrefly/lib/test/generic_callable_degeneracy.rs` to reproduce a known **type-checking bug** (issue #2105). The test demonstrates an issue where `ParamSpec` incorrectly pins to the first function overload, rather than preserving the intended overload structure. This diagnostic addition helps to clarify the conceptual problem within the **type checker's handling of generic callable types and overloads**, specifically concerning `ParamSpec` behavior. It provides a concrete example for future debugging and development, paving the way for a **correctness fix** in how `ParamSpec` interacts with overloaded functions.

This commit introduces a **new test case** in `pyrefly/lib/test/generic_callable_degeneracy.rs` to explore the interaction between `ParamSpec` and **generic functions**. Specifically, it demonstrates current **type inference** behavior when `ParamSpec` wraps a generic function whose return type is not fully determined by the call. This **exploratory analysis** helps understand how type parameters currently resolve to partial types, either inferred from first use or falling back to concrete types. The insights gained from this test will inform the future design and improvement of `ParamSpec` resolution against overloaded or generic functions.

This commit **adds a new test case** to `pyrefly/lib/test/generic_callable_degeneracy.rs` to explore degenerate behaviors of **generic callables** and how `Concatenate` preserves generic parameters. Specifically, it verifies the **partial type resolution** when a generic function's result is called, demonstrating how generic return types might not be fully determined. This **exploratory test** is crucial for understanding existing type system behavior, informing future design decisions for `ParamSpec` when resolved against overloaded or generic functions. It helps ensure the **type checker** correctly handles complex generic scenarios.

This commit introduces a **new test case** in `pyrefly/lib/test/generic_callable_degeneracy.rs` to explore the **type system's behavior** when `Concatenate` strips generic parameters from a `forall` type. Specifically, it examines how a **degenerate callable type** with a dangling return type is handled, observing that the system currently falls back to `Any`. This **exploratory test** helps document existing behavior and will inform future design decisions for `ParamSpec` resolution against complex generic functions. It highlights a scenario where contextual typing might not fully resolve partial types, impacting the precision of type inference.

This commit introduces a **new test case** within `pyrefly/lib/test/generic_callable_degeneracy.rs` to explore **degenerate generic callable behaviors**. It specifically verifies the type inference behavior when a **type variable** is present in a callable's return type but not in any of its parameter types, preventing its resolution from call arguments. This **clarifies existing type system behavior** for such scenarios, ensuring that the type variable remains unsolved in these cases. The added test is crucial for understanding and informing the future design of `ParamSpec` resolution, particularly when dealing with overloaded or generic functions.

This commit introduces **new test cases** to the **Pyrefly type checker** to explicitly document its current behavior when encountering **degenerate generic callable types**, specifically those where type variables are only present in the return type. This **maintenance** work, adding `pyrefly/lib/test/generic_callable_degeneracy.rs`, clarifies how Pyrefly handles such scenarios, which is crucial for understanding existing semantics before tackling complex issues related to `ParamSpec` and `Callable` overloads. The tests do not alter any existing logic but provide a foundational understanding of the system's current state, aiding future **bug fixes** and **feature development** in the **type inference** subsystem.

This commit **refactors the testing methodology** for the `cinderx` module by introducing an **expect-style testing framework** to improve test readability and maintainability. It specifically converts the `function_types` test case within `pyrefly/lib/test/cinderx` to this new format, adding a new Python fixture and its corresponding `.expected` output file. This **test infrastructure improvement** involves adding helper functions like `check_fixture` in `mod.rs` to facilitate fixture-based expect tests. The change aims to enable faster iteration and test-driven development by comparing actual `cinderx` output against predefined expectations.

This commit **refactors** the **`cinderx` type analysis testing suite** by converting simple scalar tests to a new **expect test format**. It introduces new Python fixture files, such as `literal_type.py` and `optional_type.py`, alongside their corresponding `.expected` output files, which are rendered by the Rust-based `format_module_types` tool. This **maintenance** change updates the `pyrefly/lib/test/cinderx/mod.rs` module to utilize a `check_fixture` pattern for these tests. The new format is designed to be **easier to maintain** and facilitates **TDD-style development** for new features or bug fixes within `cinderx`'s type analysis capabilities.

This commit **refactors** the **`pyrefly/lib/test/cinderx`** module by converting existing callable and bound-method tests to an "expect test format". This **maintenance** work introduces new test fixtures, such as `bound_method_defining_class.py` and `callable_defining_func.py`, along with their corresponding expected output files. The new format allows for more human-verifiable output and facilitates quicker test-driven development (TDD) iteration, particularly for `cinderx/pyrefly` integration bugs. Consequently, `pyrefly/lib/test/cinderx/mod.rs` is updated to utilize these new fixture-based tests, improving the overall testing infrastructure for core Python features.

This commit **refactors** the **`pyrefly` testing suite** by converting a generic type test to an expect test format. Specifically, it introduces a new Python test fixture `class_with_type_args.py` and its corresponding expected output `class_with_type_args.expected` to test **CinderX's handling of class type arguments**. The `pyrefly/lib/test/cinderx/mod.rs` file is updated to utilize this new fixture, replacing an older inline test. This **test refactoring** improves the **maintainability and structure of the test suite** for generic type analysis within `pyrefly`.

This commit **refactors** and **enhances the test suite** for **type narrowing** within the `pyrefly` project's `cinderx` integration. It converts existing or adds new callable/bound-method tests to an "expect test format" by introducing multiple Python test fixtures, such as `facet_narrow_chained_attr.py` and `facet_narrow_index.py`, along with their corresponding `.expected` output files. This change significantly improves the **testability** and **debuggability** of type analysis, providing human-verifiable output that clearly displays narrowed/unnarrowed types and mismatch markers. The `pyrefly/lib/test/cinderx/mod.rs` file was updated to integrate these new fixtures, enabling a more robust and TDD-friendly approach to validating complex type narrowing scenarios.

This commit **refactors** the **testing infrastructure** for **CinderX/PyRefly integration** by converting numerous `__static__` contextual-type tests to the "expect-test" format. This **maintenance** effort involves adding new Python test fixtures and their corresponding `.expected` output files, while removing older, less readable inline test functions from `pyrefly/lib/test/cinderx/mod.rs`. Additionally, the redundant `test_literal_promoted_type` is **deleted**, as its functionality is already covered by an existing fixture. This change significantly improves the **developer experience** for TDD and debugging, as expect-style tests provide more human-readable output for quicker iteration on `cinderx/pyrefly` integration issues.

This commit introduces a **new capability** to the **Pyrefly CLI** by adding the `--cinderx-include-readable` flag. This flag enables the generation of **human-readable `.txt` files** for CinderX type reports, alongside the existing machine-optimized `.json` output. The **CinderX reporting subsystem** is enhanced with a new `display` module (`pyrefly/lib/report/cinderx/display.rs`) and modifications to `write_results` in `pyrefly/lib/report/cinderx/mod.rs` to conditionally produce these debug-friendly reports. This significantly improves the **debuggability** of CinderX type data by providing easily parsable output for manual inspection, complementing the structured data used for modeling.

This commit **refactors** the **type inference** engine in `pyrefly/lib/alt/solve.rs` by **unifying and aggressively reducing the maximum width of inferred return type unions to 3**. This **policy refinement** aims to significantly improve **user experience** by **reducing noisy type errors** and false positives, particularly for dynamic code, resulting in 4755 fixed errors. It also serves as a crucial **performance optimization**, preventing "recursive explosions" that previously caused regressions in recursive return type inference. By limiting union complexity, the system now focuses on inferring only simple, unambiguous return types, effectively handling common patterns like `Optional` while avoiding overly complex and often incorrect inferences.

This commit **refactors the testing infrastructure** by introducing a dedicated `cinderx` folder within the `tests` directory. This change aims to **improve the organization and isolation of report-specific tests** for the **CinderX** subsystem. By grouping these tests separately, it facilitates easier management, allows for better fixture integration, and prepares the groundwork for potentially transitioning from unit testing to more scalable expect testing for CinderX bug validation. This is a **maintenance** change that enhances the maintainability and future extensibility of the test suite.

This commit **removes** the `skip_determinism_check` flag from the build configuration for the **`parso` module**. This **maintenance** and **cleanup** task is performed because the `parso` implementation has been deterministic since a previous fixpoint implementation was landed, rendering the explicit skip flag unnecessary. The flag was likely a temporary measure to bypass non-determinism that has since been resolved. This change **simplifies** the build process by removing an obsolete check, ensuring that `parso`'s determinism is now implicitly expected and not explicitly bypassed.