Narrow by property assignment in an object literal

[sandersn]

This PR narrows the type of variables that (1) have a type annotation and (2) are initialized with an object literal:

// @strict: true
type A = {
    x?: string[]
    y?: number[]
}
let a: A = {
    x: [],
}
a.x.push('hi') // ok!
a.y.push(4) // error, object is possibly undefined

This works recursively, and for shorthand property assignments:

type A = {
    x?: string[]
    y?: number[]
    z?: {
        ka?: boolean
        ki?: boolean
    }
}
const y = [1, 2, 3]
let a: A = {
    x: [],
    y,
    z: {
        ka: false
    },
}
a.x.push('hi')
a.y.push(4)
let b = a.z.ka // ok
b = a.z.ki // error, object is possibly undefined

But not for spread assignments — it's too complicated to track reference in the control flow graph — or string-literal/number-literal/computed properties — they are accessed via element access, which doesn't participate in control flow.

Note that FlowInitializer looks very similar to FlowAssignment. I tried re-using the existing structures, but the code size to constantly decide between the two uses was actually quite a bit bigger. Handling FlowInitializer is still very similar to FlowAssignment, but mostly simpler.

Fixes #20219

[sandersn]

@RyanCavanaugh I think you were interested in this.

[weswigham]

Just because it's non-obvious if it works, can we add a test with a binding pattern, ie:

// @strict: true
type A = {
    x?: string[]
    y?: number[]
    z?: {
        ka?: boolean
        ki?: boolean
    }
    extra?: string
    0?: string
    'two words'?: string
}
const y = [1, 2, 3]
const wat = { extra: "life" }
let {x, y, z, extra, 0: zero, ["two words"]: words}: A = {
    x: [],
    y,
    z: {
        ka: false
    },
    ...wat,
    0: 'hi',
    'two words': 'ho'
}
x.push('hi')
y.push(4)
let b = z.ka
b = z.ki // error, object is possibly undefined
extra.length // error, reference doesn't match the spread
zero.length
words.length

[typescript-bot]

Thanks for your contribution. This PR has not been updated in a while and cannot be automatically merged at the time being. For housekeeping purposes we are closing stale PRs. If you'd still like to continue working on this PR, please leave a message and one of the maintainers can reopen it.