This change updates the future and task API as discussed in the stabilization RFC at https://github.com/rust-lang/rfcs/pull/2592.
Changes:
- Replacing UnsafeWake with RawWaker and RawWakerVtable
- Removal of LocalWaker
- Removal of Arc-based Wake trait
Rollup of bare_trait_objects PRs
All deny attributes were moved into bootstrap so they can be disabled with a line of config.
Warnings for external tools are allowed and it's up to the tool's maintainer to keep it warnings free.
r? @Mark-Simulacrum
cc @ljedrz @kennytm
std::ops::Try impl for std::task::Poll
I originally left out the `Try` impl for `Poll` because I was curious if we needed it, and @MajorBreakfast and I had discussed the potential for it to introduce confusion about exactly what control-flow was happening at different points. However, after porting a pretty significant chunk of Fuchsia over to futures 0.3, I discovered that I was *constantly* having to do repetitive matching on `Poll<Result<...>>` or `Poll<Option<Result<...>>>` in order to propagate errors correctly. `try_poll` (propagate `Poll::Ready(Err(..))`s) helped in some places, but it was far more common to need some form of conversion between `Result`, `Poll<Result<...>>`, and `Poll<Option<Result<...>>>`. The `Try` trait conveniently provides all of these conversions in addition to a more concise syntax (`?`), so I'd like to experiment with using these instead.
cc @seanmonstar
r? @aturon
Note: this change means that far more futures 0.1 code can work without significant changes since it papers over the fact that `Result` is no longer at the top-level when using `Stream` and `Future` (since it's now `Poll<Result<...>>` or `Poll<Option<Result<...>>>` instead of `Result<Poll<..>>` and `Result<Poll<Option<...>>>`).
Clarify what a task is
Currently we call two distinct concepts "task":
1. The top-level future that is polled until completion
2. The lightweight "thread" that is responsible for polling the top-level future. What additional data beside the future is stored in this type varies between different `Executor` implementations.
I'd prefer to return to the old formulation by @alexcrichton:
```rust
/// A handle to a "task", which represents a single lightweight "thread" of
/// execution driving a future to completion.
pub struct Task {
```
Source: [`task_impl/mod.rs` in futures-rs 0.1](1328fc9e8a/src/task_impl/mod.rs (L49-L50))
I think that this change will make it much easier to explain everything.
r? @aturon
@cramertj
Since NonNull is Copy the inner field of the cloned Waker was copied for
use in the new LocalWaker, however this left Waker to be dropped. Which
means that when cloning LocalWaker would also erroneously call drop_raw.
This change forgets the Waker, rather then dropping it, leaving the
inner field to be used by the returned LocalWaker.
Closes#52629.
Unpin references
I also considered adding an impl for raw pointers as well, but that makes it easy to accidentally have unsound owning-collections that might otherwise be able to project pinned-ness (e.g. `Box`).
cc @RalfJung
r? @withoutboats
I don't have a strong opinion on the function vs. method, but there's no point in having both. I'd like to make a `repeat` adaptor like Python/Haskell for turning a value into an infinite stream of the value, so this has to at least be renamed.
