Fix negative type narrowing for isinstance/issubclass with type[T] variable or tuple by hsusul · Pull Request #11616 · microsoft/pyright

Henry Su (hsusul) · 2026-08-10T05:16:55Z

Summary

Fixes an issue where isinstance() and issubclass() type narrowing failed to extract class filters when passed a variable of type type[T] or tuple[type[T], ...]. Also enables sound negative type narrowing when T is a @final class.

Problem & Root Cause

getIsInstanceClassTypes extracts element class types for isinstance/issubclass. When given a variable of type type[T] or tuple[type[A], type[B]], the elements in tupleTypeArgs are ClassInstances of type (e.g. type[A]), rather than instantiable class types A. Previously, getIsInstanceClassTypes set foundNonClassType = true on type[T] instances, failing to extract the target filter types and resulting in no type narrowing.
In negative narrowing (isPositiveTest = false), non-final classes typed as type[T] must preserve includeSubclasses = true because type[T] may hold a runtime subclass of T. However, when T is @final (e.g. @final class or built-in final type), T cannot have runtime subclasses, making negative narrowing sound.

Fix

In getIsInstanceClassTypes:
- Specifically check for ClassInstances of type (ClassType.isBuiltIn(subtype, 'type')) and extract their type arguments (type[T] -> T).
- Reject type[Any] or type[Unknown] as concrete class filters.
- Preserve includeSubclasses = true on extracted instantiable class types.
In narrowTypeForInstanceOrSubclassInternal:
- Allow negative elimination when ClassType.isFinal(concreteFilterType) is true, permitting sound negative narrowing for final class type[T] variables.

Verification

All 160 unit tests in typeEvaluator1.test.ts pass cleanly (160/160).
git diff --check, npm run check, and npm run typecheck pass with 0 errors.

…filter is a type[T] variable or tuple of type[T]

Stella Huang (StellaHuang95) · 2026-08-10T05:17:27Z

🔒 Automated review in progress — Stella Huang (@StellaHuang95) is auto-reviewing this PR.

Stella Huang (StellaHuang95) · 2026-08-10T06:02:32Z

Clearing includeSubclasses makes negative narrowing for type[T] unsound: a runtime strict subclass can cause an A value to enter an else branch where this change removes A.

Stella Huang (StellaHuang95) · 2026-08-10T06:03:23Z

Clearing includeSubclasses makes negative narrowing for type[T] filters unsound and regresses existing subclass behavior.

Stella Huang (StellaHuang95) · 2026-08-10T06:04:44Z

Negative narrowing becomes unsound because type[A] may hold a strict subclass of A, and the new conversion also accepts non-type[...] class instances as valid filters.

… soundness

Stella Huang (StellaHuang95) · 2026-08-10T17:20:04Z

+                        if (isInstantiableClass(typeArg)) {
+                            subtype = typeArg;
+                        } else if (isClass(typeArg) && TypeBase.isInstance(typeArg)) {
+                            subtype = convertToInstantiable(typeArg);


Info · Optional note

The instantiable typeArg branch assigns the filter directly, while the instance-form branch converts it and thereby preserves includeSubclasses. Normalize the instantiable branch as well so any future representation reaching it cannot bypass the conservative non-final negative-narrowing behavior. [verified]

Stella Huang (StellaHuang95) · 2026-08-10T17:20:07Z

+    if isinstance(x, cls):
+        reveal_type(x, expected_text="FinalClass")
+    else:
+        reveal_type(x, expected_text="B")


Warning · Non-blocking recommendation

Please add a non-final negative case. cls: type[A] may hold a subclass of A, so if not isinstance(x, cls) must retain A and reveal A | B; this is the soundness boundary protected by the new final-class guard. [verified]

Stella Huang (StellaHuang95) · 2026-08-10T17:20:10Z

@@ -0,0 +1,37 @@
+# This sample tests type narrowing for isinstance and issubclass when
+# the class argument is passed as a type[T] variable or tuple of type[T].


Warning · Non-blocking recommendation

The sample and PR claim coverage for both isinstance and issubclass, but this file exercises only isinstance. Add an issubclass(..., cls: type[A]) regression case so its narrowing consumer remains covered. [verified]

Stella Huang (StellaHuang95) · 2026-08-10T17:20:13Z

+    if isinstance(x, cls):
+        reveal_type(x, expected_text="FinalClass")
+    else:
+        reveal_type(x, expected_text="B")


Warning · Non-blocking recommendation

Add regression coverage for type[Any] and type[Unknown]. The new foundNonClassType path intentionally rejects these as concrete filters, but the behavior is currently untested.

Stella Huang (StellaHuang95)

Approved via Review Center.

Rich Chiodo (rchiodo)

Approved via Review Center.

Rich Chiodo (rchiodo)

Approved via Review Center.

Rich Chiodo (rchiodo) · 2026-08-17T11:49:03Z

🔒 Automated review in progress — Rich Chiodo (@rchiodo) is auto-reviewing this PR.

Rich Chiodo (rchiodo) · 2026-08-17T19:36:35Z

+        reveal_type(x, expected_text="A | B")
+
+
+def test_final_class_param(x: FinalClass | B, cls: type[FinalClass]):


Warning · Non-blocking recommendation

Please add regression coverage for issubclass, including final-class negative narrowing, and for a non-final type[A] negative case that remains conservative. The tuple test should also include an excluded alternative so it demonstrates that the tuple filter narrows rather than merely preserving A | B.

Rich Chiodo (rchiodo)

Approved via Review Center.

Heejae Chang (heejaechang)

Approved via Review Center.

Rich Chiodo (rchiodo) · 2026-08-24T17:45:27Z

+    if isinstance(x, cls):
+        reveal_type(x, expected_text="FinalClass")
+    else:
+        reveal_type(x, expected_text="B")


Info · Optional note

The regression sample exercises only isinstance with concrete arguments, despite the change also claiming issubclass and explicit type[Any]/type[Unknown] behavior. Add focused assertions for those paths so the helper's full changed contract is protected.

[verified]

Rich Chiodo (rchiodo) · 2026-08-24T17:45:31Z

Verification: The relevant tests could not be fully run in the isolated environment; this review is not fully verified.

Fix negative type narrowing for isinstance and issubclass when class …

683a924

…filter is a type[T] variable or tuple of type[T]

Stella Huang (StellaHuang95) added the review-auto:changes-requested Automated review: posted blocking findings to address. label Aug 10, 2026

Fix type[T] filter extraction in getIsInstanceClassTypes and preserve…

d023ce2

… soundness

Stella Huang (StellaHuang95) reviewed Aug 10, 2026

View reviewed changes

Stella Huang (StellaHuang95) approved these changes Aug 10, 2026

View reviewed changes

Stella Huang (StellaHuang95) added review-auto:approved Automated review: no blocking findings (approval posted). and removed review-auto:changes-requested Automated review: posted blocking findings to address. labels Aug 10, 2026

Rich Chiodo (rchiodo) approved these changes Aug 10, 2026

View reviewed changes

Rich Chiodo (rchiodo) approved these changes Aug 11, 2026

View reviewed changes

Rich Chiodo (rchiodo) reviewed Aug 17, 2026

View reviewed changes

Rich Chiodo (rchiodo) approved these changes Aug 17, 2026

View reviewed changes

Heejae Chang (heejaechang) approved these changes Aug 17, 2026

View reviewed changes

Rich Chiodo (rchiodo) reviewed Aug 24, 2026

View reviewed changes

Navigation Menu

Uh oh!

Fix negative type narrowing for isinstance/issubclass with type[T] variable or tuple - #11616

Fix negative type narrowing for isinstance/issubclass with type[T] variable or tuple#11616
Henry Su (hsusul) wants to merge 2 commits into
microsoft:mainfrom
hsusul:fix/isinstance-typevar-tuple-negative-narrowing

Henry Su (hsusul) commented Aug 10, 2026 •

edited

Loading

Stella Huang (StellaHuang95) commented Aug 10, 2026 •

edited

Loading

Stella Huang (StellaHuang95) commented Aug 10, 2026

Stella Huang (StellaHuang95) commented Aug 10, 2026

Stella Huang (StellaHuang95) commented Aug 10, 2026

Stella Huang (StellaHuang95) Aug 10, 2026

Stella Huang (StellaHuang95) Aug 10, 2026

Stella Huang (StellaHuang95) Aug 10, 2026

Stella Huang (StellaHuang95) Aug 10, 2026

Stella Huang (StellaHuang95) left a comment

Rich Chiodo (rchiodo) left a comment

Rich Chiodo (rchiodo) left a comment

Rich Chiodo (rchiodo) commented Aug 17, 2026 •

edited

Loading

Rich Chiodo (rchiodo) Aug 17, 2026

Rich Chiodo (rchiodo) left a comment

Heejae Chang (heejaechang) left a comment

Rich Chiodo (rchiodo) Aug 24, 2026

Rich Chiodo (rchiodo) commented Aug 24, 2026

Labels

4 participants

		@@ -0,0 +1,37 @@
		# This sample tests type narrowing for isinstance and issubclass when
		# the class argument is passed as a type[T] variable or tuple of type[T].

		reveal_type(x, expected_text="A \| B")


		def test_final_class_param(x: FinalClass \| B, cls: type[FinalClass]):

Uh oh!

Conversation

Henry Su (hsusul) commented Aug 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Problem & Root Cause

Fix

Verification

Stella Huang (StellaHuang95) commented Aug 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Stella Huang (StellaHuang95) commented Aug 10, 2026

Stella Huang (StellaHuang95) commented Aug 10, 2026

Stella Huang (StellaHuang95) commented Aug 10, 2026

Stella Huang (StellaHuang95) Aug 10, 2026

Choose a reason for hiding this comment

Stella Huang (StellaHuang95) Aug 10, 2026

Choose a reason for hiding this comment

Stella Huang (StellaHuang95) Aug 10, 2026

Choose a reason for hiding this comment

Stella Huang (StellaHuang95) Aug 10, 2026

Choose a reason for hiding this comment

Stella Huang (StellaHuang95) left a comment

Choose a reason for hiding this comment

Rich Chiodo (rchiodo) left a comment

Choose a reason for hiding this comment

Rich Chiodo (rchiodo) left a comment

Choose a reason for hiding this comment

Rich Chiodo (rchiodo) commented Aug 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Rich Chiodo (rchiodo) Aug 17, 2026

Choose a reason for hiding this comment

Rich Chiodo (rchiodo) left a comment

Choose a reason for hiding this comment

Heejae Chang (heejaechang) left a comment

Choose a reason for hiding this comment

Rich Chiodo (rchiodo) Aug 24, 2026

Choose a reason for hiding this comment

Rich Chiodo (rchiodo) commented Aug 24, 2026

Labels

4 participants

Henry Su (hsusul) commented Aug 10, 2026 •

edited

Loading

Stella Huang (StellaHuang95) commented Aug 10, 2026 •

edited

Loading

Rich Chiodo (rchiodo) commented Aug 17, 2026 •

edited

Loading