Type of variable isn't widened inside a conditional inside a loop

[icecream17]

Bug Report

🔎 Search Terms

ts2367 widen loop

🕗 Version & Regression Information

• This is a crash
• This changed between versions ______ and _______
• This is the behavior in every version I tried, and I reviewed the FAQ for entries about _________
• I was unable to test this on prior versions because _______

⏯ Playground Link

Playground link

💻 Code

Edit: shorter version by gabritto

let code: 0 | 1 = 0
while (true) {
    code = code === 1 ? 0 : 1
}

And in practice:

🙁 Actual behavior

Error

🙂 Expected behavior

Since code could be changed in previous iterations of the loop, the condition is not always false, so there shouldn't be any error.

[fatcerberus]

No, there are actually two variables: code and code2. The latter is a loop variable.

[fatcerberus]

Clarification on above comment: I was responding to someone who mistakenly thought code and code2 in the repro code were the same variable. That comment has since been deleted, but I left my response in case someone else makes the same mistake.

[andrewbranch]

A little more reduced: https://www.typescriptlang.org/play?ts=4.6.0-dev.20211105#code/FAGwpgLgBAxg9gEzALigBigHygRi1AJigF51hgAzOAJygAoBuBgSigG9govZEwSekJYqSIB+XFFRpy3KAEsK9AIRKKAQxABnMKw6zZ8QSM7cAvsHNA

The code in the screenshot is kind of easier to follow even though it’s longer since it’s real.

[jameswilddev]

Question: I've just encountered this which looks similar to me, the root problem in both cases being that TypeScript is very quick to narrow types on let statements when it should probably leave them as the type they were declared as:

https://www.typescriptlang.org/play?#code/DYUwLgBGAWCWB2BzAXBABgQTRAPugQtnmgMLYC86WA3ALABQDAxgPbwDOk74ElAFGBapMRAqNJoAlKgBuLWABNeAPggBvBhC1Q4SXlBZ16AXyMNuYPhMln6sAGYQBuxL3KVr6zdtYcWoADpgFkQ+AHJEFjCbBmMGBiA

Would you say this is the same issue or that I should raise a seperate bug?

[Zzzen]

@jameswilddev it's duplicate of #44824

[gabritto]

Even smaller: https://www.typescriptlang.org/play?ts=4.5.4#code/DYUwLgBAxg9gJiAXBADBAPhAjBAvKgKAHcALAS1AgAowAnAVxAEoIBvAiT6eEPDrgQNgI8ufFn6CpXAPyFpCiMiwBuAgF8gA

[MartinJohns]

@RyanCavanaugh Probably want to update the milestone again. :-)

[gabritto]

@ahejlsberg I started investigating this bug a while back but couldn't come up with a fix, so it was reassigned to you, but now I'm wondering if I could fix this along with a fix for #47539. Are you planning on working on this?

[ahejlsberg]

The issue here is that during control flow analysis of the loop, the type of the code variable is seen as 0 during an intermediate step. That's to be expected, but unfortunately it triggers an error that isn't actually correct once we arrive at the full type 0 | 1. We need to suppress that error, which we can do by tracking the fact that checkExpression was called from getTypeOfExpression (in which we care only about the type and not the diagnostics). I will put up a PR to that effect.