[ty] Only treat recursive types with unbounded specializations as fallback cycles

[mtshiba]

Summary

Fixes #26503 (comment)

Approach #26989 didn't work. This would be a more legitimate approach.

In #26503, regardless of what the recursive type was, if it appeared again with a different specialization during the visit of CycleDetector, there was a risk that the visit would not stop, so a Cycle state was returned immediately.

This is OK for the growing pattern, but misses the saturating pattern, as shown below.

type Growing[T] = T | Growing[list[T]]

type Saturating1[T, U] = T | Saturating1[U, T]
type Saturating2[T] = tuple[T] | tuple[T, Saturating2[int]]
type Saturating3[T] = tuple[T, Saturating3Helper[list[T]]]
type Saturating3Helper[U] = tuple[U, Saturating3[int]]

Think of this as a graph problem. The type parameter of each definition is used as a node, and the following edges are created from the type parameter of the recursive reference.

• Direct: Type parameters are passed as is.
• Nested: A type parameter is passed inside another type.
• Edgeless if no type parameters appear

If there is even one nested edge in the directed cycle containing the root formal type argument, it is determined that the specialization can increase infinitely.

e.g.

type A[X, Y] = A[list[Y], X]
# A.Y -> A.X  Nested
# A.X -> A.Y  Direct
# => growing

type Left[A, B, C, D] = Left[B, C, D, None]
# Left.B -> Left.A  Direct
# Left.C -> Left.B  Direct
# Left.D -> Left.C  Direct
# => saturating

type A[X, Y] = A[list[Y], None]
# A.Y -> A.X  Nested
# => saturating (no cycle)

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 96.96%. The percentage of expected errors that received a diagnostic held steady at 92.96%. The number of fully passing files held steady at 106/133.

[astral-sh-bot]

Memory usage report

Summary

Project	Old	New	Diff	Outcome
sphinx	168.71MB	168.72MB	+0.01% (11.16kB)	⏫
prefect	451.60MB	451.60MB	+0.00% (456.00B)	⏫
flake8	40.31MB	40.31MB	-	✅
trio	93.47MB	93.47MB	-0.00% (928.00B)	⬇️

Significant changes

Click to expand detailed breakdown

sphinx

Name	Old	New	Diff	Outcome
Type<'db>::cached_materialization_	247.39kB	250.33kB	+1.19% (2.94kB)	⏫
may_have_unbounded_specialization_inner	0.00B	2.31kB	+2.31kB (new)	⏫
infer_expression_types_impl	11.42MB	11.42MB	+0.02% (1.78kB)	⏫
may_have_unbounded_specialization_inner::interned_arguments	0.00B	1.75kB	+1.75kB (new)	⏫
infer_definition_types	10.78MB	10.78MB	+0.01% (1.20kB)	⏫
all_narrowing_constraints_for_expression	2.24MB	2.24MB	+0.02% (392.00B)	⏫
is_redundant_with_impl	764.24kB	764.52kB	+0.04% (288.00B)	⏫
infer_statement_types_impl	504.58kB	504.77kB	+0.04% (192.00B)	⏫
infer_scope_types_impl	6.07MB	6.07MB	+0.00% (112.00B)	⏫
StaticClassLiteral<'db>::try_mro_	1.87MB	1.87MB	+0.00% (64.00B)	⏫
infer_unpack_types	259.11kB	259.18kB	+0.02% (64.00B)	⏫
analyze_non_terminal_call	454.88kB	454.94kB	+0.01% (64.00B)	⏫
loop_header_reachability	243.32kB	243.35kB	+0.01% (32.00B)	⏫
member_lookup_with_policy_inner	4.80MB	4.80MB	+0.00% (8.00B)	⏫

prefect

Name	Old	New	Diff	Outcome
may_have_unbounded_specialization_inner::interned_arguments	0.00B	384.00B	+384.00B (new)	⏫
may_have_unbounded_specialization_inner	0.00B	352.00B	+352.00B (new)	⏫
Type<'db>::cached_materialization_	1.03MB	1.03MB	+0.02% (240.00B)	⏫
is_redundant_with_impl	3.35MB	3.35MB	-0.00% (144.00B)	⬇️
infer_expression_types_impl	23.35MB	23.35MB	-0.00% (136.00B)	⬇️
member_lookup_with_policy_inner	10.75MB	10.75MB	-0.00% (120.00B)	⬇️
Type<'db>::apply_specialization_inner_	3.42MB	3.42MB	-0.00% (24.00B)	⬇️
try_call_dunder_get_inner	1.08MB	1.08MB	-0.00% (24.00B)	⬇️
infer_definition_types	34.54MB	34.54MB	-0.00% (24.00B)	⬇️
all_narrowing_constraints_for_expression	4.52MB	4.52MB	-0.00% (16.00B)	⬇️
StaticClassLiteral<'db>::implicit_attribute_inner_	803.80kB	803.79kB	-0.00% (16.00B)	⬇️
infer_expression_type_impl	250.37kB	250.35kB	-0.01% (16.00B)	⬇️
FunctionType<'db>::signature_	4.38MB	4.38MB	+0.00% (16.00B)	⏫
infer_unpack_types	426.79kB	426.78kB	-0.00% (8.00B)	⬇️
infer_deferred_types	4.88MB	4.88MB	+0.00% (8.00B)	⏫
...			2 more

trio

Name	Old	New	Diff	Outcome
Type<'db>::apply_specialization_inner_::interned_arguments	1.18MB	1.18MB	-0.03% (400.00B)	⬇️
Type<'db>::apply_specialization_inner_	791.83kB	791.55kB	-0.04% (288.00B)	⬇️
FunctionType	1.23MB	1.23MB	-0.02% (288.00B)	⬇️
may_have_unbounded_specialization_inner	0.00B	136.00B	+136.00B (new)	⏫
infer_definition_types	4.43MB	4.43MB	-0.00% (128.00B)	⬇️
may_have_unbounded_specialization_inner::interned_arguments	0.00B	64.00B	+64.00B (new)	⏫
infer_deferred_types	1.28MB	1.28MB	-0.00% (24.00B)	⬇️

[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)