You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Should awaiting a non-Promise value be an error? #8310
Currently this compiles just fine and behaves correctly at runtime. But is there a situation where the user would want to deliberately await a non-Promise? Or should it be a compiler error?
I thought of:
Normalizing a sync-or-async value:
declarefunctionmayReturnSyncOrAsyncResult(): number|Promise<number>;varresult=awaitmayReturnSyncOrAsyncResult();// number
Either await could still be allowed for unions that contain Promise, or the user could be required to use await Promise.resolve instead of just await.
Extracting the value of a thenable:
declarefunctionfoo(): bluebird.Promise<number>;varresult=awaitfoo();// number
Again await could still be allowed, or the user could be required to use await Promise.resolve
Deliberately introducing a yield point to make the rest of the function be a separate async continuation. I'm not sure if such a use actually exists in the wild. If it does exist, it too can be handled by await Promise.resolve
The downside of await Promise.resolve is that it introduces one extra Promise, since await will wrap the awaited value in one anyway.
I'm not completely convinced one way or the other. Opening this issue to get TS team's and other people's thoughts.
Alternatively I suppose we can make a tslint rule for "stricter awaits" or something.
I believe it is not an error to await a non-thenbale non-promise object. awaiting a literal as in the example should be equivalent to await Promise.resolve(3). Brian Terlson (@bterlson) can point to the spec details.
one thing to note, var x = await 5; is not the same as var x = 5;; var x = await 5; will assign x5 in the next tern, where as var x = 5; will evaluate immediately.
Me : I can't think of a good reason. The only use case (next tern) is not one the user should be depending upon. There might be some obscure debounce style stuff that one can cook up 🌹
Just some anecdotal info: I got unwanted behavior when writing my first TypeScript program, since in many libs the TS version uses *Async names for the Promise-returning version of the function.
So since the callback was optional, I "awaited" on void and then the program would continue too early...
So if there no real reason to keep this behavior, I can assure you that this has already created some bug (even though luckily only during development).
The goal of the current semantics isn't to allow uselessly awaiting known synchronous values but instead to allow awaiting maybe-promises and to align with the semantics of various existing promise utilities including Promise.resolve (promises and thenables unwrapped, values passed through) and then handlers (returned promises and thenables unwrapped, values passed through).
Thinking about this more, this could be enforced under a flag, by default if the compiler can attest that the type of await expression can not be a promise, then it should report an error, if the flag is set, the error reporting is skipped.
Not necessarily clear this an error (maybe you intend to effectively yield)
What if the awaited expression is a type parameter? Not clear how to determine if that's an error or not
We clearly want to allow await T | Promise<T> (for reasons Brian Terlson (@bterlson) mentioned above) but then it's weird to allow await Promise<T> but disallow await T since we're effectively saying either might happen
We can't tell that you definitely meant to not await any arbitrary value, so disallowing it isn't something we can do for certain.
Indeed it is. The reason why I've dropped my usage of ntypescript (which could support ts extension) and moved to a simpler automation of bring your own typescript : https://github.com/basarat/byots (as the extension can now be done using tslint) and also working on first class integration into alm : alm-tools/alm#155 🌹
Activity
basarat commentedon Apr 26, 2016
Yes. When the API consumer wants a promise but the API provider doesn't need to be async. The provider code can then just
await🌹basarat commentedon Apr 26, 2016
instead of
you get to
Arnavion commentedon Apr 26, 2016
Basarat Ali Syed (@basarat) That should be
basarat commentedon Apr 26, 2016
Indeed. Sorry. I know I should have been sleeping at that time 🌹
mhegazy commentedon Apr 26, 2016
// CC Brian Terlson (@bterlson)
I believe it is not an error to await a non-thenbale non-promise object. awaiting a literal as in the example should be equivalent to
await Promise.resolve(3). Brian Terlson (@bterlson) can point to the spec details.one thing to note,
var x = await 5;is not the same asvar x = 5;;var x = await 5;will assignx5in the next tern, where asvar x = 5;will evaluate immediately.Arnavion commentedon Apr 26, 2016
Yes, I said all that in the OP.
basarat commentedon Apr 26, 2016
This is the heart of the question by Arnav Singh (@Arnavion) ^.
Me : I can't think of a good reason. The only use case (next tern) is not one the user should be depending upon. There might be some obscure debounce style stuff that one can cook up 🌹
mhegazy commentedon Apr 26, 2016
Brian Terlson (@bterlson), can you shed some light on the rational/motivating use cases for the current design?
tp commentedon Apr 27, 2016
Just some anecdotal info: I got unwanted behavior when writing my first TypeScript program, since in many libs the TS version uses *Async names for the Promise-returning version of the function.
So since the callback was optional, I "awaited" on void and then the program would continue too early...
So if there no real reason to keep this behavior, I can assure you that this has already created some bug (even though luckily only during development).
bterlson commentedon Apr 28, 2016
The goal of the current semantics isn't to allow uselessly awaiting known synchronous values but instead to allow awaiting maybe-promises and to align with the semantics of various existing promise utilities including Promise.resolve (promises and thenables unwrapped, values passed through) and then handlers (returned promises and thenables unwrapped, values passed through).
mhegazy commentedon Apr 28, 2016
Thinking about this more, this could be enforced under a flag, by default if the compiler can attest that the type of await expression can not be a promise, then it should report an error, if the flag is set, the error reporting is skipped.
RyanCavanaugh commentedon Jun 9, 2016
Main points from discussion:
yield)await T | Promise<T>(for reasons Brian Terlson (@bterlson) mentioned above) but then it's weird to allowawait Promise<T>but disallowawait Tsince we're effectively saying either might happenWe can't tell that you definitely meant to not await any arbitrary value, so disallowing it isn't something we can do for certain.
Arnavion commentedon Jun 9, 2016
I dunno, atleast to me it doesn't seem weird.
tinganho commentedon Jul 25, 2016
Would be great if there was a flag for raising an error.
RyanCavanaugh commentedon Jul 25, 2016
Rumor has it
tslintis getting type-based rules in the near future thanks to some work from the Angular team, so that might be a good place to go.basarat commentedon Jul 25, 2016
Indeed it is. The reason why I've dropped my usage of ntypescript (which could support ts extension) and moved to a simpler automation of bring your own typescript : https://github.com/basarat/byots (as the extension can now be done using tslint) and also working on first class integration into alm : alm-tools/alm#155 🌹