-
Notifications
You must be signed in to change notification settings - Fork 341
Changes Extend trait in order to allow streams that yield references #266
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
@sunjay thanks for the detailed write-up! -- It definitely sounds like you've put a good deal of thought into this, and above all: this seems to be working! I'm very excited about this patch set; this is very good! Thanks so much for doing this! |
@yoshuawuyts This is done and ready to review/merge! I used There is still one impl whose body is I think this is in a reasonable state to merge. All of these impls are unstable, so I don't think the unimplemented is a big deal. Up to you! |
Could you restart the build? It seems to have run into an internal compiler error. |
bors r+ |
266: Changes Extend trait in order to allow streams that yield references r=yoshuawuyts a=sunjay This is not ready to merge yet. I am mainly opening it so we can discuss a change I had to make to the `Extend` trait. cc @yoshuawuyts @stjepang (and anyone else interested) ## Before this can be merged - [x] Discuss/Approve changes to `Extend` trait - [x] Change to using `for_each` after #264 is merged - [ ] (optional) Wait until a `copied()` method is added to `StreamExt` so that the `&char` impl can be finished. - We can also just comment out or remove the impl that uses `copied` until that is added ## Changes To The Extend Trait While writing the impls of the `Extend` trait for the `String` type, I noticed that certain impls weren't possible because there is no bound on `Extend` that guarantees that the type `A` being yielded from the stream actually lives long enough. We probably didn't run into this earlier because this usually isn't a problem for owned values since the compiler doesn't have to worry about whether they will out live the stream that they come from. I ran into this because of the `Extend` impls that consume streams of references. The difference between the async `Extend` and the standard library `Extend` is that the async `Extend` returns a value that still references the input stream. That means that if `A` is any reference type, the compiler needs to be able to guarantee that `A` will be around as long as the `Future` returned from the trait method is around. To fix this, I had to add the bound shown below: ```patch pub trait Extend<A> { /// Extends a collection with the contents of a stream. fn stream_extend<'a, T: IntoStream<Item = A> + 'a>( &'a mut self, stream: T, - ) -> Pin<Box<dyn Future<Output = ()> + 'a>>; + ) -> Pin<Box<dyn Future<Output = ()> + 'a>> where A: 'a; } ``` This guarantees that each value of type `A` will last at least as long as our boxed future does. The bound had to be in a where clause on the method (and not on the declaration of `A` because the lifetime `'a` isn't in scope at the trait level. I don't think there are any negative consequences of using a where clause like this, but that's why I wanted to bring it up for discussion. In addition to this, I had to ensure that when writing the `Extend` impls for `String` I appropriately bounded the lifetime of the references from the stream. You can see this in the code below with `where 'b: 'a`. ```rust impl<'b> Extend<&'b str> for String { fn stream_extend<'a, S: IntoStream<Item = &'b str> + 'a>( &'a mut self, stream: S, ) -> Pin<Box<dyn Future<Output = ()> + 'a>> where 'b: 'a { //TODO: This can just be: stream.into_stream().for_each(move |s| self.push_str(s)) Box::pin(stream.into_stream().fold((), move |(), s| self.push_str(s))) } } ``` I should note that initially I tried to make it work with just the impl shown above, without modifying the `Extend` trait. This doesn't work because it would be a stricter bound than what is found in the trait itself. Rust does not allow stricter bounds like that because it could potentially cause unsoundness when dealing with generics. Of course, I am totally open to being completely wrong in my understanding of how to resolve this issue. I tried to solve the problem with as minimal of a change as possible. Please let me know if you have some better ideas or other suggestions. ## `FromStream` impls for String The purpose of adding these `Extend` impls is to continue my work from #129 in adding the rest of the `FromStream` impls. The `Extend` impls are used directly to add all of the `FromStream` impls for `String`. Just like with #207 and #265, this adds a new `string` module that is unstable just like the other modules added for `FromStream`. Co-authored-by: Sunjay Varma <varma.sunjay@gmail.com>
Build succeeded
|
This is not ready to merge yet. I am mainly opening it so we can discuss a change I had to make to the
Extend
trait. cc @yoshuawuyts @stjepang (and anyone else interested)Before this can be merged
Extend
traitfor_each
after Adds for_each stream combinator #264 is mergedcopied()
method is added toStreamExt
so that the&char
impl can be finished.copied
until that is addedChanges To The Extend Trait
While writing the impls of the
Extend
trait for theString
type, I noticed that certain impls weren't possible because there is no bound onExtend
that guarantees that the typeA
being yielded from the stream actually lives long enough. We probably didn't run into this earlier because this usually isn't a problem for owned values since the compiler doesn't have to worry about whether they will out live the stream that they come from. I ran into this because of theExtend
impls that consume streams of references.The difference between the async
Extend
and the standard libraryExtend
is that the asyncExtend
returns a value that still references the input stream. That means that ifA
is any reference type, the compiler needs to be able to guarantee thatA
will be around as long as theFuture
returned from the trait method is around.To fix this, I had to add the bound shown below:
This guarantees that each value of type
A
will last at least as long as our boxed future does. The bound had to be in a where clause on the method (and not on the declaration ofA
because the lifetime'a
isn't in scope at the trait level. I don't think there are any negative consequences of using a where clause like this, but that's why I wanted to bring it up for discussion.In addition to this, I had to ensure that when writing the
Extend
impls forString
I appropriately bounded the lifetime of the references from the stream. You can see this in the code below withwhere 'b: 'a
.I should note that initially I tried to make it work with just the impl shown above, without modifying the
Extend
trait. This doesn't work because it would be a stricter bound than what is found in the trait itself. Rust does not allow stricter bounds like that because it could potentially cause unsoundness when dealing with generics.Of course, I am totally open to being completely wrong in my understanding of how to resolve this issue. I tried to solve the problem with as minimal of a change as possible. Please let me know if you have some better ideas or other suggestions.
FromStream
impls for StringThe purpose of adding these
Extend
impls is to continue my work from #129 in adding the rest of theFromStream
impls. TheExtend
impls are used directly to add all of theFromStream
impls forString
. Just like with #207 and #265, this adds a newstring
module that is unstable just like the other modules added forFromStream
.