Generic argument narrowed from `T | undefined` collapses a dependent conditional return type to `never`/`undefined`
Dieses Issue hat noch niemand übernommen.
Bewertung
- Schwierigkeit
- 4/5
- Geschätzter Aufwand
- 3-5 Tage
- Anfängerfreundlichkeit
- 45/100
- Issue-Typ
- Bug
- Klarheit
- Klar beschrieben
- Aktivitätsstatus
- Aktiv
- Tech-Stack
- typescript
- Bereich
- compilers
Rechercherichtung
Start by running the supplied TypeScript reproduction in the Playground and compare the inferred types with and without the truthy narrowing. Trace the generic narrowing and conditional-type evaluation involved in the get call; done means the narrowed call preserves the value type and reports number | undefined rather than plain undefined, with a regression test covering the case.
Vom Indexierungsmodell aus dem Issue-Text verfasst.
Beschreibung
Bug Report
- This is a bug, not a question.
TypeScript Version
Reproduces on 6.0.3 and on 7.0.2 (npx typescript@latest, checked 2026-09-22) — not a recent regression, present in both.
Search Terms
narrowing generic T & {}, distributive conditional type never, naked type parameter
narrowed loses distributivity, generic argument narrowed conditional return type never,
extends infer R generic narrowing never
Code
type PlainObject<T> = Exclude<T, string | number | boolean | null | undefined> & object;
type Keys<T> = PlainObject<T> extends infer R ? keyof R : never;
type ValueAt<T, K extends Keys<T>> =
PlainObject<T> extends infer R ? (K extends keyof R ? R[K] : never) : never;
declare function get<const T, const K extends Keys<T>>(obj: T, key: K): ValueAt<T, K>;
// Works as expected: no narrowing involved, return type resolves to the real value type.
function withoutNarrowing<T, K extends Keys<T>>(obj: T, key: K) {
return get(obj, key);
}
// Bug: same call, but obj/key were first narrowed down from `T | undefined` via a truthy check.
function withNarrowing<T, K extends Keys<T>>(obj: T | undefined, key: K | undefined) {
if (!obj || !key) return;
return get(obj, key);
}
const ok = withoutNarrowing({ a: 1 }, 'a'); // number -- correct
const bug = withNarrowing({ a: 1 }, 'a'); // undefined -- wrong, should be `number | undefined`
Playground (paste the code above manually if this link doesn't load it correctly)
(The playground snippet includes two extra lines, checkOk/checkBug, that force the compiler to print each variable's real type in an error message — remove them, or just hover over ok/bug instead.)
Expected behavior
bug's type should be number | undefined — same as ok, plus undefined from the
early return. obj/key are proven non-nullable at the get(obj, key) call site (right
after the if (!obj || !key) return; guard), so the call should behave identically to
withoutNarrowing.
Actual behavior
bug's type collapses to plain undefined — the ValueAt<T, K> branch silently
resolves to never, so the union ValueAt<T, K> | undefined simplifies away the real
value type entirely. No error is raised anywhere; the incorrect type just propagates
silently, which is the dangerous part (a consumer reading bug's type has no signal
anything went wrong).
Root cause (as far as I could narrow it down): after if (!obj || !key) return;, obj
narrows to T & {} rather than to T (documented since TS 4.8, to stay sound about
generics that could structurally include null/undefined). T & {} is an intersection
type, not the naked type parameter symbol T anymore — and Keys<T>/ValueAt<T, K>
rely on T being naked for their X extends infer R ? ... : never conditionals to stay
deferred (unevaluated) until a concrete T is substituted. Once inference for the get
call has to work with T & {} (and its constraint, unknown), the checker seems to try
to evaluate the conditional eagerly instead of deferring it, and it collapses to never.
This only reproduces with the X extends infer R ? ... : never indirection present in
both Keys<T> and ValueAt<T, K> — a version without that indirection (e.g. a plain
T[K] lookup instantiated through a similarly-shaped generic call) resolves correctly
even after the same narrowing, so the infer R step specifically seems to be what tips
the checker into eager evaluation.
Workaround found
Passing the type arguments explicitly at the call site avoids the bug entirely, since it
sidesteps inference off the narrowed (T & {}) argument type altogether:
function withNarrowingFixed<T, K extends Keys<T>>(obj: T | undefined, key: K | undefined) {
if (!obj || !key) return;
return get<T, K>(obj, key); // <- explicit type args: back to `number | undefined`
}
Related Issues
- https://github.com/microsoft/TypeScript/issues/57246
- https://github.com/microsoft/TypeScript/issues/57707
- https://github.com/microsoft/TypeScript/issues/46975
- https://github.com/microsoft/TypeScript/issues/54806
- https://github.com/microsoft/TypeScript/issues/52327
Related in that they deal with narrowing + generic conditional types, but none is an exact duplicate of the scenario above, as far as I could tell.
- Vorherrschende Sprache
- Go
- Sterne
- 111k
- Forks
- 14.4k
- Ø Merge
- 2 T. 1 Std.
- Gemergte PRs (30 T.)
- 126
Beitragsleitfaden
Erste Schritte
- Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
- Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
- Forken Sie das Repository und arbeiten Sie in einem Branch.
- Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.
Mehr aus microsoft/TypeScript
-
Possible Improvement
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 78/100
microsoft/TypeScript#64278 · 1 Kommentar · 1 Reaktion ·
-
Docs
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 70/100
microsoft/TypeScript#64118 · 1 Kommentar ·
-
Schwierigkeit 1/5 Unter einer Stunde Anfängerfreundlichkeit 88/100
microsoft/TypeScript#64094 ·
-
Docs
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 76/100
microsoft/TypeScript#63959 · 5 Kommentare ·
-
Domain: lib.d.ts Help Wanted
Schwierigkeit 1/5 Unter einer Stunde Anfängerfreundlichkeit 91/100
microsoft/TypeScript#63722 · 4 Kommentare · 1 Reaktion ·
Alle Issues in microsoft/TypeScript
Ähnliche Issues
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 65/100
-
bug group: validation priority: low
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 75/100
codecheckers/chekhov#51 ·
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 70/100
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 75/100
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 75/100