Stack overflow with incorrect mapped type as constraint of type parameter

[sandersn]

function sum<T extends { [P in T]: number }, K extends keyof T>() {
    n += v[k];
}

This incorrect code puts the compiler in an infinite loop:

     RangeError: Maximum call stack size exceeded
      at Object.contains (src/compiler/core.ts:276:5)
      at getBaseConstraint (src/compiler/checker.ts:5955:21)
      at getResolvedBaseConstraint (src/compiler/checker.ts:5949:36)
      at hasNonCircularBaseConstraint (src/compiler/checker.ts:5936:20)
      at getConstraintOfTypeParameter (src/compiler/checker.ts:5906:20)
      at getConstraintTypeFromMappedType (src/compiler/checker.ts:5761:56)
      at isGenericMappedType (src/compiler/checker.ts:5798:84)
      at computeBaseConstraint (src/compiler/checker.ts:5997:21)
      at getBaseConstraint (src/compiler/checker.ts:5960:32)
      at computeBaseConstraint (src/compiler/checker.ts:5969:38)
      at getBaseConstraint (src/compiler/checker.ts:5960:32)

[arciisine]

I'm having a similar issue to this, but can't reproduce the issue in a minimal case. In the minimal case it works properly, but inside of my project, I have this same stack trace, but my code is:

export type MatchQuery<T> = {
  [P in keyof T]?: T[P] | MatchQuery<T[P]>
};

[sandersn]

Timothy Soehnlin (@arciisine) for that small of a repro, there's nothing to cause the compiler to start checking the insides of the mapped type. As soon as you add some code that uses the mapped type you should see the crash.

[arciisine]

I actually have a larger minimal case, that doesn't cause it to happen. It appears to be something along the way, that I'm going to try to narrow down and see if I can hit that a minimal repro.

Btw, I have many versions of the recursive mapped type that do not throw any errors, but this one in specific seems to be problematic.

[sandersn]

Fixed by #17912

[sandersn]

Note that the fix in #17912 is specifically the deletion of if (isGenericMappedType(t)) { return emptyObjectType; } at the end of computeBaseConstraint. It looks like I will probably not delete this code in #17912, so this bug will remain open.

[sandersn]

The call to isGenericMappedType might be correct if it would always return, but it indirectly calls getResolvedBaseConstraint(type) from within a call to getResolvedBaseConstraintType(type), so the process never makes progress.

I have a fix that clones isGenericMappedType and its callees inside of getResolvedBaseConstraint and replaces the recursive call, which fixes the bug in an ugly way.