Fix discriminant property narrowing through optional chain with null

[andrewbranch]

Fixes #42480

Got mixed up in #42450 thinking of type as the type of the discriminant property access, which would be undefined even if the union type included null. But type is the union, not the discriminant property, so we need to filter out both null and undefined, but then only mix back in undefined to the discriminant property type to model the behavior of an optional chain.

[DanielRosenwasser]

Need to accept baselines? https://github.com/microsoft/TypeScript/actions?query=workflow%3A%22Accept+Baselines+and+Fix+Lints%22

[ahejlsberg]

Funny, I was looking at this issue and came up with exactly the same fix. I had planned to add the following tests, I suggest you include them in this PR:

type U = { kind: undefined, u: 'u' }
type N = { kind: null, n: 'n' }
type X = { kind: 'X', x: 'x' }

function f1(x: X | U | undefined) {
    if (x?.kind === undefined) {
        x;
    }
    else {
        x;
    }
}

function f2(x: X | N | undefined) {
    if (x?.kind === undefined) {
        x;
    }
    else {
        x;
    }
}

function f3(x: X | U | null) {
    if (x?.kind === undefined) {
        x;
    }
    else {
        x;
    }
}

function f4(x: X | N | null) {
    if (x?.kind === undefined) {
        x;
    }
    else {
        x;
    }
}

function f5(x: X | U | undefined) {
    if (x?.kind === null) {
        x;
    }
    else {
        x;
    }
}

function f6(x: X | N | undefined) {
    if (x?.kind === null) {
        x;
    }
    else {
        x;
    }
}

function f7(x: X | U | null) {
    if (x?.kind === null) {
        x;
    }
    else {
        x;
    }
}

function f8(x: X | N | null) {
    if (x?.kind === null) {
        x;
    }
    else {
        x;
    }
}

[rbuckton]

Optional chaining leverages an optionalType marker type to indicate the type of an expression came from a ?.. The optionalType marker is essentially undefined, but its presence has meaning to completions as well as within the checker itself. The code here (and from #42450), removes undefined (regardless of source), then mixes back in undefinedType. In functions like checkPropertyAccessChain, I use the function propagateOptionalTypeMarker to restore optionalType to the result if it was found in the source type to carry that marker forward.

I'm not certain yet whether this (or #42450) will cause problems as a result of not restoring the optionalType marker, as I need to consider if there is a test case that might trigger unwanted behavior.

[andrewbranch]

@rbuckton it sounds like any potential problem is already in master, so I’m going to go ahead and merge this. Let me know if you have a suggestion for a follow-up and I’ll be happy to add to it.