[ty] Allow recursive type expansion a limited number of times during type relation checking

[mtshiba]

Summary

This PR addresses #26503 (comment)

As described in the comment, instead of immediately returning the Cycle state when a same identity is found, we expect the recursive expansion to stabilize by Pending several times. If it does not stabilize even after expanding 10 times, it will be determined as a cycle.

This makes DefinitionReferenceVisitor completely unnecessary. In other words, even if we see signs of a growing pattern, we will try expanding recursive types up to 10 times. At first glance, it might seem better to keep the DefinitionReferenceVisitor guard in place, but doing so would actually cause us to miss patterns that look like growing patterns but ultimately converge in a finite number of steps. For example:

type Left[A, B, C, D] = Left[B, C, D, D]
type Right[A, B, C, D] = Right[B, C, D, D]

Now consider these type relations:

Left[
    int,
    list[int],
    list[list[int]],
    list[list[list[int]]],
]

Right[
    int,
    list[int],
    list[list[int]],
    list[list[list[int]]],
]

CycleDetector observes the increase in specialization as follows. However, this will eventually converge after a finite number of steps. In other words, it is not possible to conclude that "this type relation check will not stop" just from the increasing structure of specialization observed by CycleDetector.

[int, list[int],  list[list[int]], list[list[list[int]]]]
↓
[list[int], list[list[int]], list[list[list[int]]], list[list[list[int]]]]
↓
[list[list[int]], list[list[list[int]]], list[list[list[int]]], list[list[list[int]]]]
↓
[list[list[list[int]]], list[list[list[int]]], list[list[list[int]]], list[list[list[int]]]]
↓
[list[list[list[int]]], list[list[list[int]]], list[list[list[int]]], list[list[list[int]]]] (converged)

Test Plan

mdtest updated

[astral-sh-bot]

Typing conformance results

No changes detected ✅

Current numbers

The percentage of diagnostics emitted that were expected errors held steady at 96.95%. The percentage of expected errors that received a diagnostic held steady at 92.58%. The number of fully passing files held steady at 104/133.

[astral-sh-bot]

Memory usage report

Memory usage unchanged ✅

[astral-sh-bot]

ecosystem-analyzer results

No diagnostic changes detected ✅

Flaky changes detected. This PR summary excludes flaky changes; see the HTML report for details.

Full report with detailed diff (timing results)

[codspeed-hq]

Merging this PR will not alter performance

✅ 111 untouched benchmarks
⏩ 84 skipped benchmarks¹

---

_{Comparing cycle-detector-visit-limit (f507e20) with main (8c30182)}

Footnotes

1. 84 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. ↩

[mtshiba]

This approach has the undesirable side effect that the union type reduction results depend on the order. Therefore, it cannot be adopted.

type Left12[A, B, C, D, E, F, G, H, I, J, K] = tuple[A, Left12[B, C, D, E, F, G, H, I, J, K, L, None]]
type Right12[A, B, C, D, E, F, G, H, I, J, K] = tuple[A, Right12[B, C, D, E, F, G, H, I, J, K, L, None]]

type Source = Left12[int, int, int, int, int, int, int, int, int, int, int, int]
type RightAfterTwo = Right12[int, int, int, int, int, int, int, int, int, int, None, None]
type Shortcut = tuple[int, tuple[int, RightAfterTwo]]
# Shortcut is equivalent to LongPath
type LongPath = Right12[int, int, int, int, int, int, int, int, int, int, int, int]

static_assert(is_subtype_of(Source, Shortcut))
static_assert(is_subtype_of(Source, Shortcut | LongPath))
static_assert(is_subtype_of(Source, LongPath | Shortcut))  # failure!

Instead, it is necessary to redesign DefinitionReferenceVisitor so that the detector does not immediately return Cycle for recursive-but-saturate patterns.