Prevent crash on unpacking when tuple-returning overload is rejected by outer context

mypy/checker.py

@@ -4199,7 +4199,19 @@ def check_multi_assignment_from_tuple(
                     # inferred return type for an overloaded function
                     # to be ambiguous.
                     return
-                assert isinstance(reinferred_rvalue_type, TupleType)
+                if not isinstance(reinferred_rvalue_type, TupleType):
+                    # Using outer context may transform a tuple type into something
+                    # else when the callee is overloaded. Recheck with the new result
+                    # and bail out.
+                    self.check_multi_assignment(
+                        lvalues,
+                        rvalue,
+                        context,
+                        rv_type=reinferred_rvalue_type,
+                        infer_lvalue_type=infer_lvalue_type,
+                        undefined_rvalue=undefined_rvalue,
+                    )
+                    return
                 rvalue_type = reinferred_rvalue_type
 
             left_rv_types, star_rv_types, right_rv_types = self.split_around_star(

test-data/unit/check-inference-context.test

@@ -126,6 +126,22 @@ class A(Generic[T]): pass
 class B: pass
 [out]
 
+[case testInferUnpackedOverloadWithTupleAndOtherReturns]
+# https://github.com/python/mypy/issues/19920
+from typing import Tuple, TypeVar, overload
+
+_T = TypeVar("_T")
+
+@overload  # type: ignore[no-overload-impl]
+def gather(f1: _T) -> Tuple[_T]: ...
+@overload
+def gather(*fns: object) -> int: ...
+
+def crash() -> None:
+    foo: str
+    (foo,) = gather(0)  # E: "int" object is not iterable
+[builtins fixtures/tuple.pyi]
+
 [case testInferMultipleLocalVariableTypesWithArrayRvalueAndNesting]
 from typing import TypeVar, List, Generic
 T = TypeVar('T')