New type inference: add support for upper bounds and values

mypy/solve.py

@@ -10,11 +10,13 @@
 from mypy.expandtype import expand_type
 from mypy.graph_utils import prepare_sccs, strongly_connected_components, topsort
 from mypy.join import join_types
-from mypy.meet import meet_types
+from mypy.meet import meet_type_list, meet_types
 from mypy.subtypes import is_subtype
 from mypy.typeops import get_type_vars
 from mypy.types import (
     AnyType,
+    Instance,
+    NoneType,
     ProperType,
     Type,
     TypeOfAny,
@@ -108,15 +110,15 @@ def solve_constraints(
             else:
                 candidate = AnyType(TypeOfAny.special_form)
             res.append(candidate)
-    return res, [originals[tv] for tv in free_vars]
+    return res, free_vars
 
 
 def solve_with_dependent(
     vars: list[TypeVarId],
     constraints: list[Constraint],
     original_vars: list[TypeVarId],
     originals: dict[TypeVarId, TypeVarLikeType],
-) -> tuple[Solutions, list[TypeVarId]]:
+) -> tuple[Solutions, list[TypeVarLikeType]]:
     """Solve set of constraints that may depend on each other, like T <: List[S].
 
     The whole algorithm consists of five steps:
@@ -135,23 +137,24 @@ def solve_with_dependent(
     raw_batches = list(topsort(prepare_sccs(sccs, dmap)))
 
     free_vars = []
+    free_solutions = {}
     for scc in raw_batches[0]:
         # If there are no bounds on this SCC, then the only meaningful solution we can
         # express, is that each variable is equal to a new free variable. For example,
         # if we have T <: S, S <: U, we deduce: T = S = U = <free>.
         if all(not lowers[tv] and not uppers[tv] for tv in scc):
-            # For convenience with current type application machinery, we use a stable
-            # choice that prefers the original type variables (not polymorphic ones) in SCC.
-            # TODO: be careful about upper bounds (or values) when introducing free vars.
-            free_vars.append(sorted(scc, key=lambda x: (x not in original_vars, x.raw_id))[0])
+            best_free = choose_free([originals[tv] for tv in scc], original_vars)
+            if best_free:
+                free_vars.append(best_free.id)
+                free_solutions[best_free.id] = best_free
 
     # Update lowers/uppers with free vars, so these can now be used
     # as valid solutions.
-    for l, u in graph.copy():
+    for l, u in graph:
         if l in free_vars:
-            lowers[u].add(originals[l])
+            lowers[u].add(free_solutions[l])
         if u in free_vars:
-            uppers[l].add(originals[u])
+            uppers[l].add(free_solutions[u])
 
     # Flatten the SCCs that are independent, we can solve them together,
     # since we don't need to update any targets in between.
@@ -166,7 +169,7 @@ def solve_with_dependent(
     for flat_batch in batches:
         res = solve_iteratively(flat_batch, graph, lowers, uppers)
         solutions.update(res)
-    return solutions, free_vars
+    return solutions, [free_solutions[tv] for tv in free_vars]
 
 
 def solve_iteratively(
@@ -276,6 +279,61 @@ def solve_one(lowers: Iterable[Type], uppers: Iterable[Type]) -> Type | None:
     return candidate
 
 
+def choose_free(
+    scc: list[TypeVarLikeType], original_vars: list[TypeVarId]
+) -> TypeVarLikeType | None:
+    """Choose the best solution for an SCC containing only type variables.
+
+    This is needed to preserve e.g. the upper bound in a situation like this:
+        def dec(f: Callable[[T], S]) -> Callable[[T], S]: ...
+
+        @dec
+        def test(x: U) -> U: ...
+
+    where U <: A.
+    """
+
+    if len(scc) == 1:
+        # Fast path, choice is trivial.
+        return scc[0]
+
+    common_upper_bound = meet_type_list([t.upper_bound for t in scc])
+    common_upper_bound_p = get_proper_type(common_upper_bound)
+    # We include None for when strict-optional is disabled.
+    if isinstance(common_upper_bound_p, (UninhabitedType, NoneType)):
+        # This will cause to infer <nothing>, which is better than a free TypeVar
+        # that has an upper bound <nothing>.
+        return None
+
+    values: list[Type] = []
+    for tv in scc:
+        if isinstance(tv, TypeVarType) and tv.values:
+            if values:
+                # It is too tricky to support multiple TypeVars with values
+                # within the same SCC.
+                return None
+            values = tv.values.copy()
+
+    if values and not is_trivial_bound(common_upper_bound_p):
+        # If there are both values and upper bound present, we give up,
+        # since type variables having both are not supported.
+        return None
+
+    # For convenience with current type application machinery, we use a stable
+    # choice that prefers the original type variables (not polymorphic ones) in SCC.
+    best = sorted(scc, key=lambda x: (x.id not in original_vars, x.id.raw_id))[0]
+    if isinstance(best, TypeVarType):
+        return best.copy_modified(values=values, upper_bound=common_upper_bound)
+    if is_trivial_bound(common_upper_bound_p):
+        # TODO: support more cases for ParamSpecs/TypeVarTuples
+        return best
+    return None
+
+
+def is_trivial_bound(tp: ProperType) -> bool:
+    return isinstance(tp, Instance) and tp.type.fullname == "builtins.object"
+
+
 def normalize_constraints(
     constraints: list[Constraint], vars: list[TypeVarId]
 ) -> list[Constraint]: