Control flow analysis for element access with variable index doesn't account for async race condition

[rixtox]

🔎 Search Terms

Control flow analysis async race condition

🕗 Version & Regression Information

• This changed between versions 5.4.5 and 5.5.0
• This changed in Control flow analysis for element access with variable index #57847

⏯ Playground Link

https://www.typescriptlang.org/play/?ts=5.5.1-rc#code/IYZwngdgxgBAZgV2gFwJYHsIwOYFNkBywAtrgBQCUAXDAAoBO6xqIuAPCMvahNgHwwA3gCgYYmPXwJ6WAOQB3dPQA2AE1kBuYQF9hw0JFiIUGLHACMZdACMAVjQBKuKEtUcuPbABoYZMgA8aTm5eChgAXgEAN3RUVTCAHxgkVVw4HlxVPh8Aa1wwII9QoVFxVDhfZDAAB1x0CptbAG08sABdCPDwmAAiYyg0TB6wkXExmBcIThgIElwImGB5YFRkHHwiUkotcfFGlvy2slnSCh3xXTH91o7ulLSM1S1dYUnp1oXZbEl8T01XzDTRoLUbiA7tY5zQohbDUGAxOIlXZiN7oZS4AB0ynQ2DIAAMABK4ZTYnwAEkEJ1w2gAhHizqUxLoXgwmCxMcASWQmhYrHZcvkKD5eY0BWAKG0GUA

💻 Code

async function getName(): Promise<string> {
    return 'world';
}

async function f1(obj: Record<string, ((x: string) => void) | undefined>, key: string) {
    if (typeof obj[key] === "function") {
        const name = await getName();
        obj[key](name);
    }
    obj[key] = undefined;
}

const key = 'greeting';
const obj = {
    [key](name: string): void {
        console.log(`Hello, ${name}!`);
    }
}

Promise.all([f1(obj, key), f1(obj, key)]);

🙁 Actual behavior

No type error, but runtime failure:

Unhandled Promise Rejection: TypeError: obj[key] is not a function. (In 'obj[key](name)', 'obj[key]' is undefined)

🙂 Expected behavior

Type checker should account for potential race condition and do not persist the narrowed type after an await statement.

Additional information about the issue

No response

[nmain]

This is a duplicate of #9998; see also the many other duplicates linked from that issue.

[rixtox]

This is a duplicate of #9998; see also the many other duplicates linked from that issue.

But this is a regression as 5.4.5 would not make the optimistic type narrowing on this path.

[Josh-Cena]

How is this relevant to computed keys at all? Your code would have the same unsoundness if you substitute all [key] with .greeting.

[rixtox]

How is this relevant to computed keys at all? Your code would have the same unsoundness if you substitute all [key] with .greeting.

It's not about computed keys. I just took the example from this blog and extended a bit. The main concern is that when there's an await in the control flow, you can no longer assume "neither obj nor key are ever mutated".

[MartinJohns]

The main concern is that when there's an await in the control flow, you can no longer assume "neither obj nor key are ever mutated".

Which is exactly what #9998 is about. And it's unrelated to await, a sync function has the same issue.

[fatcerberus]

The analogy for a synch function would be an intervening function call that causes the same mutation.

This issue is Exhibit A of "why TS team was so hesitant to add narrowing by indexed access for so long". The bottom line is the compiler doesn't know whether any given await (or function call) is going to change types, so it's just a question of whether it should be optimistic or pessimistic when it encounters one. Here TS chose to be optimistic. If it was pessimistic, people would complain about unaffected variables being widened (leading to unsafe casts and/or redundant additional type checks), so neither option is really "better" than the other--it's a tradeoff.

[rixtox]

neither option is really "better" than the other--it's a tradeoff.

Maybe worth mentioning the tradeoffs in the blog post. I read the blog post and felt it might not be as reliable as it seems. Which is why I wrote this issue. I don't think I would be the only one with this thought.

[typescript-bot]

This issue has been marked as "Duplicate" and has seen no recent activity. It has been automatically closed for house-keeping purposes.