Danny Yang

This commit **refactors** the handling of **module-level ignore directives** (`ignore_all`) within the `pyrefly` linter's **error collection and suppression logic**. It moves the `ignore_all` computation from the `Ignore` struct's initialization to a later stage in `get_errors`/`get_all_errors`, where AST-derived multi-line string ranges are available. This **pure refactoring** extracts `parse_ignore_all` as a standalone function and introduces a new `ModuleRanges` struct to bundle these directives and ranges. The `ignore_all` state is now threaded through the error collection pipeline, ensuring `UnusedIgnore` errors are exempt from suppression. This change enables future improvements to directive recognition, such as after module docstrings, without altering current behavior or duplicating string-tracking logic.

This commit **fixes a bug** in **PyreFly's error suppression mechanism** by adding robust support for **backslash continuation comments**. Previously, suppression comments were not correctly applied to code using backslash continuations, especially when placed after the `\` or between lines. New logic in `pyrefly/lib/state/errors.rs` (e.g., `sorted_backslash_continuation_ranges`) now accurately identifies these blocks, and `pyrefly/lib/state/state.rs` integrates these ranges into the overall multi-line processing. The **error suppression logic** in `pyrefly/lib/error/suppress.rs` (`add_suppressions`) is enhanced to correctly remap errors and apply suppressions, ensuring a comment above the first line covers the entire continued expression. This **bug fix** significantly improves the reliability and usability of error suppression for Python code utilizing backslash continuations, addressing issue #2935.

This commit **enhances Pyrefly's ignore directive parsing** by allowing `# pyrefly: ignore-errors` directives to be placed after a module docstring. It modifies the `parse_ignore_all` function in `crates/pyrefly_python/src/ignore.rs` to correctly skip over docstrings when scanning for top-level ignores, leveraging recently added multi-line string range information. This **feature enhancement relaxes placement requirements** for these directives, addressing a specific user request for increased flexibility in code suppression. New test cases in `pyrefly/lib/test/suppression.rs` validate this behavior, ensuring that only explicit `ignore-errors` are recognized in this position, not ambiguous bare `# type: ignore`.

This commit implements a **bug fix** and **validation improvement** within the `pyrefly` type checker, specifically addressing the use of `Self` return annotations. It now **rejects `Self` as a return type** when the `self` parameter of a method is explicitly annotated with a `TypeVar`, as this constitutes an invalid semantic usage. This change impacts the **function validation logic** in `pyrefly/lib/alt/function.rs` and extends the **core type system** in `crates/pyrefly_types/src/types.rs` with a new `is_explicit_type_variable` method. The update introduces a **new type-checking error** for previously accepted, but semantically incorrect, code, thereby enhancing the robustness and correctness of type inference.

This commit **fixes a bug in the type binding logic** that incorrectly allowed `ParamSpec` components like `P.args` and `P.kwargs` to be resolved even when the `ParamSpec` `P` was not properly defined or in scope. The **`pyrefly/lib/binding` module** now includes `try_intercept_lookup` and modifies `Expr::Attribute` binding to ensure `P.args` and `P.kwargs` are only recognized if `P` is an existing `ParamSpec` type parameter, preventing its implicit introduction. This **improves the accuracy of `ParamSpec` resolution** and provides more precise error messages for incorrect usage. Conformance tests and `ParamSpec` unit tests have been updated to reflect this corrected behavior and clearer error reporting.

This commit **implements core assignability rules and corrects the behavior of `TypeForm`** within the **PyRefly type system**, specifically addressing aspects of **PEP 747**. It updates the `pyrefly/lib/alt/specials.rs` module to ensure `TypeForm[T]` correctly produces the `Type::TypeForm(T)` variant and modifies `pyrefly/lib/alt/solve.rs` to interpret bare `TypeForm` as `TypeForm[Any]`. Furthermore, the `pyrefly/lib/solver/subset.rs` module now includes comprehensive assignability rules for `TypeForm`, covering covariance and its interactions with `type`, `ClassDef`, and `Annotated` types, significantly enhancing type checking accuracy. This **feature implementation** also includes new tests and updates to conformance expectations to reflect these changes.

This commit **enhances the `pyrefly` type solver** by introducing a new rule within the `is_subset_eq` function to ensure the class representation of `Any` is compatible with all `TypeForm` instances. This **new capability** improves the type system's flexibility and correctness, allowing `Any` to be properly recognized in various type comparisons. The change is validated by **new test cases** added to `pyrefly/lib/test/typeform.rs` that specifically verify `Any`'s recognition as a `TypeForm`. This update has a positive downstream impact, as reflected by a **reduction in reported differences and failures** in the `conformance/third_party/results.json` results, leading to more accurate type checking and inference within the `pyrefly` system.

This commit introduces a **new type rule** and a **critical bug fix** to enhance the `pyrefly` type checker's handling of `TypeForm` types. It **implements a new subset rule** in `pyrefly/lib/solver/subset.rs` that allows `None` to be considered a valid `TypeForm[T]` when `None` is a subset of `T`, improving the expressiveness of type definitions. Concurrently, a **bug in type promotion** within `pyrefly/lib/alt/solve.rs` is addressed, ensuring that `TypeForm[T]` remains unwrapped in annotation positions to prevent incorrect assignment checking. These changes collectively refine the **type form semantics** and **type solving logic**, leading to more accurate and robust type inference and conformance.

This commit introduces the `Type::TypeForm` variant to the core `Type` enum within the **`pyrefly_types` crate**, representing `TypeForm[T]` values as part of **PEP 747 support**. This **new capability** involves adding a `mk_typeform` helper, updating `Display` formatting, and extending `Visit`/`VisitMut` implementations to handle the inner type. It also updates **type resolution** in `attr.rs` and **type conversion** in `convert.rs` to recognize the new variant. However, this variant is currently only defined and not yet fully utilized in annotation processing, as `apply_special_form` will still produce `Type::Type` until assignability rules are implemented.

This commit introduces the **`TypeForm` special form** to the type system, a foundational step towards implementing PEP 747. It **adds `TypeForm` to the `SpecialForm` enum** within the `pyrefly_types` crate and integrates its recognition and processing logic into the `apply_special_form` function in `pyrefly/lib/alt/specials.rs`. This **new capability** allows the system to correctly parse and handle `type[T]` constructs, expanding the expressiveness of type annotations. Extensive **conformance and unit tests** have been updated and added to validate its behavior and error handling, ensuring proper recognition and error reporting for incorrect specializations.

This commit **fixes a conformance issue** in the **type checker** by **rejecting `ClassVar[Final]` annotations in non-dataclass classes**, aligning with the Python typing specification. Previously, this invalid type combination was silently accepted, while `Final[ClassVar]` was already correctly flagged. The change introduces a new check within the `calculate_class_field` function in the **`pyrefly` type analysis module**, ensuring that this error is now reported for regular classes but remains permitted for dataclasses, which explicitly allow it. This improves the **accuracy of type analysis** and is accompanied by updated conformance tests and new unit tests for both `ClassVar[Final]` and `Final[ClassVar]` scenarios.

This commit **refactors** the **`pyrefly` library** by **renaming** an internal function responsible for tracking multi-line string positions. The function `sorted_multi_line_fstring_ranges` in `pyrefly/lib/state/errors.rs` was inaccurately named, as it handles all multi-line string types, not solely f-strings. It has been renamed to `sorted_multi_line_string_ranges` to accurately reflect its broader scope and improve code clarity. This **internal refactoring** updates all references within `pyrefly/lib/error/suppress.rs` and `pyrefly/lib/state/state.rs`, along with its documentation. The change has **no functional impact** on existing suppression logic or user-facing behavior.

This commit delivers a **bug fix** to resolve a **false positive for `__slots__` validation** within the **`pyrefly` static analyzer**. It specifically corrects an issue where attributes defined on a **metaclass** were erroneously flagged, failing to recognize them as valid data descriptors. The fix involves modifying the **attribute lookup logic** in `pyrefly/lib/alt/attr.rs`, particularly the `get_class_attribute` function, and introducing a new `is_data_descriptor` method in `pyrefly/lib/alt/class/class_field.rs` to accurately identify these attributes. This enhancement ensures `pyrefly`'s static analysis correctly handles `__slots__` in conjunction with metaclasses, aligning its behavior with `mypy` and preventing incorrect error reporting. New test cases have been added to `pyrefly/lib/test/slots.rs` to verify this corrected behavior.

This commit **refactors** the `scripts/benchmark/typecheck_benchmark.py` script to enhance its type checking configuration and clean up obsolete code. It **adds `check_untyped_defs = True`** to the generated `mypy` configuration, which will enforce stricter type checking for untyped definitions during benchmark runs. Concurrently, the **unused `_write_dummy_zuban_config` function is removed**, streamlining the codebase and improving maintainability. This change primarily impacts the **typechecking benchmark subsystem**, leading to more robust type analysis and a cleaner script.

This commit **fixes a bug** in the **type checker's match narrowing** logic, specifically addressing how **irrefutable keyword patterns** in class patterns (e.g., `B(x=_)`) are processed. Previously, these patterns incorrectly prevented **negative narrowing**, hindering the type checker's ability to eliminate the class type in subsequent `case` branches. The fix modifies `bind_pattern` in `pyrefly/lib/binding/pattern.rs` to skip `Placeholder` insertion and introduces a `strip_placeholders` method in `pyrefly/lib/binding/narrow.rs` to remove extraneous `Placeholder` operations. This ensures that the **type checker** can accurately perform **negative narrowing** for such patterns, significantly improving the **correctness** and **precision** of control flow analysis for `match` statements.

This commit delivers a **bug fix** for **pyrefly**'s **type canonicalization logic**, resolving a **false positive error** when using `isinstance` with union types like `type | T`. Previously, the `type` metatype was incorrectly double-wrapped as `Type::Type(Type::Type(Any))` within unions, causing `unwrap_class_object_silently` to fail. The fix, implemented in the `as_class_info` function within `pyrefly/lib/alt/class/classdef.rs`, now prevents this double-wrapping for union members that are already `Type::Type(...)`. This significantly improves the accuracy of **type checking** for `isinstance` and `issubclass` expressions, ensuring correct behavior for complex type annotations.

This commit **enhances the automated issue labeling system** by adding new keywords to the `.github/workflows/label_new_issues.yml` workflow. It introduces terms like 'basemodel', 'hover', 'editor', 'lspconfig', 'completion', 'highlight', and 'semantic' to improve the accuracy of issue categorization. This **maintenance** update ensures that issues containing these terms are now automatically assigned the **`pydantic`** and **`language-server`** labels. The change streamlines issue triage by leveraging insights from past issue data, making the development workflow more efficient.

This commit **enhances** the **size labeler workflow** by introducing a crucial step to **remove existing size labels** from a pull request before applying new ones. This **maintenance improvement** ensures that if a pull request's size changes, any previously assigned size labels are cleared, preventing the accumulation of multiple or outdated labels. Affecting the automated labeling system, specifically the `.github/workflows/size_labeler.yml` configuration, this change guarantees that pull requests always display the most current and accurate size classification. The primary impact is improved **accuracy and cleanliness** of pull request metadata, making it easier to track changes in PR size.

This commit **adds license metadata to the `pyrefly` project's packaging configuration**, addressing a **bug fix** identified in issue #2909. It specifically **updates the `pyproject.toml` file** to declare the project's license, ensuring proper distribution information. Additionally, this **maintenance** task ensures the `pyrefly/LICENSE` file is correctly set as a symbolic link to the main license file. This change improves the **packaging and distribution of the `pyrefly` library**, making its license information readily available and correctly specified for downstream users and tools.

This commit introduces a **new capability** to the **`pyrefly suppress` command**, allowing users to configure the placement of suppression comments. A new `--comment-location` flag enables choosing between placing comments on the line before or as trailing comments on the same line as the error. This **feature** modifies the core **error suppression logic** within `pyrefly/lib/error/suppress.rs`, specifically affecting functions like `add_suppressions` and `suppress_errors`, and extends string literal handling in `pyrefly/lib/state/errors.rs` via `sorted_multi_line_fstring_ranges`. The default behavior for `pyrefly suppress` remains unchanged, and `pyrefly check --suppress-errors` explicitly retains its existing comment placement.