[ty] Simplify Self for final types

[sharkdp]

Summary

For final classes, we simplify Self to an instance of the class.

Ideally, we would also do that for other type variables with a final bound, but those seem like a rare edge-case, because there's not really any benefit to using a type variable in the first place, if it's bound to a final type (thank you, @AlexWaygood).

closes astral-sh/ty#1404

Test Plan

Added regression test

[github-actions]

Diagnostic diff on typing conformance tests

No changes detected when running ty on typing conformance tests ✅

[github-actions]

mypy_primer results

Changes were detected when running on open source projects

mitmproxy (https://github.com/mitmproxy/mitmproxy)
- mitmproxy/proxy/layers/http/_events.py:127:17: error[type-assertion-failure] Argument does not have asserted type `Never`
- Found 1886 diagnostics
+ Found 1885 diagnostics

sphinx (https://github.com/sphinx-doc/sphinx)
+ sphinx/util/osutil.py:84:10: error[invalid-argument-type] Argument to bound method `stat` is incorrect: Expected `Bottom[DirEntry[Unknown]]`, found `Top[DirEntry[Unknown]]`
+ sphinx/util/osutil.py:91:10: error[invalid-argument-type] Argument to bound method `stat` is incorrect: Expected `Bottom[DirEntry[Unknown]]`, found `Top[DirEntry[Unknown]]`
- Found 531 diagnostics
+ Found 533 diagnostics

antidote (https://github.com/Finistere/antidote)
- src/antidote/core/_catalog.py:214:65: error[invalid-argument-type] Argument is incorrect: Expected `ReferenceType[CatalogImpl]`, found `ReferenceType[Self@__init__]`
- Found 307 diagnostics
+ Found 306 diagnostics

scikit-learn (https://github.com/scikit-learn/scikit-learn)
- sklearn/preprocessing/_data.py:3422:31: error[no-matching-overload] No overload matches arguments
- sklearn/preprocessing/_data.py:3455:27: error[no-matching-overload] No overload matches arguments
- sklearn/preprocessing/_data.py:3506:31: error[no-matching-overload] No overload matches arguments
- Found 2477 diagnostics
+ Found 2474 diagnostics

jax (https://github.com/google/jax)
- jax/_src/pallas/mosaic_gpu/core.py:1478:24: error[invalid-return-type] Function can implicitly return `None`, which is not assignable to return type `TMEMLayout`
- Found 2476 diagnostics
+ Found 2475 diagnostics

Memory usage changes were detected when running on open source projects

trio (https://github.com/python-trio/trio)
-     memo metadata = ~30MB
+     memo metadata = ~28MB

[github-actions]

ecosystem-analyzer results

Lint rule	Added	Removed	Changed
invalid-argument-type	2	1	0
no-matching-overload	0	3	0
invalid-return-type	0	1	0
type-assertion-failure	0	1	0
Total	2	6	0

Full report with detailed diff (timing results)

[codspeed-hq]

CodSpeed Performance Report

Merging #21042 will degrade performances by 5.41%

_{Comparing david/self-for-final-classes (5a6bfca) with main (e92fd51)}

Summary

❌ 1 regression
✅ 20 untouched
⏩ 30 skipped¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Mode	Benchmark	BASE	HEAD	Change
❌	Simulation	ty_micro[many_tuple_assignments]	61.3 ms	64.8 ms	-5.41%

Footnotes

1. 30 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[AlexWaygood]

Thinking about this some more, I'm not totally sure this is sound for generic methods. Just because an instance type comes from a @final class doesn't mean that there aren't subtypes of that instance type, and the Self type variable could be solved to be one of those subtypes. For example:

import enum
from typing import Self

class F(enum.Enum):
    X = 0
    Y = 1

    def method(self) -> Self:
        return self

reveal_type(F.X.method())  # `Literal[F.X]` on `main`, `F` on this branch