ruff

8e5cd421 - [ty] Decide satisfiability of simple typevar-free conjunctions directly (#27178)

[ty] Decide satisfiability of simple typevar-free conjunctions directly (#27178) ## Summary Fixes astral-sh/ty#4089, where `ty check` took O(n²) time in the size of a literal union such as `mypy_boto3_ec2`'s `InstanceTypeType` (~1200 string literals). Checking iteration over `Sequence[Literal[...]]` builds a constraint set that is a single conjunction with one lower-bound constraint per union element. Deciding its satisfiability walks the BDD with `PathAssignments`, whose `discover_constraint` computes sequents for every pair of constraints — quadratic in the number of constraints. For lower-bound-only pairs, each pair computation also eagerly builds a `Literal[a] | Literal[b]` union in `ConstraintId::intersect`, only for the result to be discarded as `CannotSimplify`, so the entire quadratic pass produces empty sequent maps. This PR adds a linear fast path to `is_never_satisfied` for BDDs that are a single all-positive conjunction with typevar-free bounds: per typevar occurrence, check that the union of the lower bounds is assignable to each upper-bound clause. Assignability distributes over the union on the left and the intersection clauses on the right, so this finds exactly the contradictions that the walk's pairwise disjointness sequents detect. Bounds are grouped by occurrence identity so differently materialized instances of the same typevar are handled together. Type aliases and protocols can hide typevars in lazy attributes, so bounds containing either conservatively fall back to the general walk; this also avoids expanding recursively specialized aliases in the fast path. The existing `compute_simple_bound_conjunction` fast path used for solution extraction is updated to use the same identity and lazy-bound handling. For typevars with only upper-bound evidence, it skips quadratic per-clause redundancy pruning and lets the final intersection determine the solution. Timings for the issue's reproducer (debug build): 4.6s → 0.03s against an installed `mypy_boto3_ec2`, 2.3s → 0.03s for a synthetic 1200-literal union, and the runtime is now flat in the union size (5000 literals also check in ~0.03s). ## Test plan - Added benchmarks covering `Sequence[Literal[...]]` access and many contravariant callback arguments that produce upper-bound-only constraints. - Added unit tests covering satisfiable and contradictory simple conjunctions without sequent-cache growth, equivalence with the general path walk, differently materialized instances of the same typevar, hidden typevars in lazy aliases, and large upper-bound-only conjunctions. - The `ty_python_semantic` test suite passes. - Stable type property tests pass with 2000 generated cases. - Manually verified the issue's reproducer and upper-bound-only scaling. --------- Co-authored-by: Carl Meyer <carl@astral.sh> Co-authored-by: Douglas Creager <dcreager@dcreager.net>