The `Context` argument is currently smuggled through TLS for
async-generated futures. The current infrastructure is closure-based,
and results in an extra 6 stack frames when .awaiting an async-generated
future!
```
12: foo::async_b::{{closure}}
at src/main.rs:10
13: <std::future::GenFuture<T> as core::future::future::Future>::poll::{{closure}}
at /rustc/4560ea788cb760f0a34127156c78e2552949f734/src/libstd/future.rs:43
14: std::future::set_task_context
at /rustc/4560ea788cb760f0a34127156c78e2552949f734/src/libstd/future.rs:79
15: <std::future::GenFuture<T> as core::future::future::Future>::poll
at /rustc/4560ea788cb760f0a34127156c78e2552949f734/src/libstd/future.rs:43
16: std::future::poll_with_tls_context::{{closure}}
at /rustc/4560ea788cb760f0a34127156c78e2552949f734/src/libstd/future.rs:121
17: std::future::get_task_context
at /rustc/4560ea788cb760f0a34127156c78e2552949f734/src/libstd/future.rs:111
18: std::future::poll_with_tls_context
at /rustc/4560ea788cb760f0a34127156c78e2552949f734/src/libstd/future.rs:121
19: foo::async_a::{{closure}}
at src/main.rs:6
```
While the long (medium?) term solution is to remove the use of TLS
entirely, we can improve things a bit in the meantime. In particular,
this commit does 2 things:
1. `get_task_context` has been inlined into `poll_with_tls_context`,
removing 2 frames (16 and 17 above).
2. `set_task_context` now returns a guard type that resets the TLS
rather than taking a closure, removing 2 frames (13 and 14 above).
We can also remove frame 18 by removing `poll_with_tls_context` in favor
of a `get_task_context` function which returns a guard, but that
requires adjusting the code generated for .await, so I've left that off
for now.
Use ptr::drop_in_place for VecDeque::truncate and VecDeque::clear
This commit allows `VecDeque::truncate` to take advantage of its (largely) contiguous memory layout and is consistent with the change in #64432 for `Vec`. As with the change to `Vec::truncate`, this changes both:
- the drop order, from back-to-front to front-to-back
- the behavior when dropping an element panics
For consistency, it also changes the behavior when dropping an element panics for `VecDeque::clear`.
These changes in behavior can be observed. This example ([playground](https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=d0b1f2edc123437a2f704cbe8d93d828))
```rust
use std::collections::VecDeque;
fn main() {
struct Bomb(usize);
impl Drop for Bomb {
fn drop(&mut self) {
panic!(format!("{}", self.0));
}
}
let mut v = VecDeque::from(vec![Bomb(0), Bomb(1)]);
std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
v.truncate(0);
}));
std::mem::forget(v);
}
```
panics printing `1` today and succeeds. `v.clear()` panics printing `0` today and succeeds. With the change, `v.clear()`, `v.truncate(0)`, and dropping the `VecDeque` all panic printing `0` first and then abort with a double-panic printing `1`.
The motivation for this was making `VecDeque::truncate` more efficient since it was used in the implementation of `VecDeque::clone_from` (#65069), but it also makes behavior more consistent within the `VecDeque` and with `Vec` if that change is accepted (this probably doesn't make sense to merge if not).
This might need a crater run and an FCP as well.
Merge repeated definitions
Step forward on #66149
I may need further context to understand the need for a separate crate.
Also, please tell me if you think of other definitions to merge.
Make error and warning annotations mandatory in UI tests
This change makes error and warning annotations mandatory in UI tests.
The only exception are tests that use error patterns to match compiler
output and don't have any annotations.
Fixes#55596.
This change makes error and warning annotations mandatory in UI tests.
The only exception are tests that use error patterns to match compiler
output and don't have any annotations.
Support registering inert attributes and attribute tools using crate-level attributes
And remove `#[feature(custom_attribute)]`.
(`rustc_plugin::Registry::register_attribute` is not removed yet, I'll do it in a follow up PR.)
```rust
#![register_attr(my_attr)]
#![register_tool(my_tool)]
#[my_attr] // OK
#[my_tool::anything] // OK
fn main() {}
```
---
Some tools (`rustfmt` and `clippy`) used in tool attributes are hardcoded in the compiler.
We need some way to introduce them without hardcoding as well.
This PR introduces a way to do it with a crate level attribute.
The previous attempt to introduce them through command line (https://github.com/rust-lang/rust/pull/57921) met some resistance.
This probably needs to go through an RFC before stabilization.
However, I'd prefer to land *this* PR without an RFC to able to remove `#[feature(custom_attribute)]` and `Registry::register_attribute` while also providing a replacement.
---
`register_attr` is a direct replacement for `#![feature(custom_attribute)]` (https://github.com/rust-lang/rust/issues/29642), except it doesn't rely on implicit fallback from unresolved attributes to custom attributes (which was always hacky and is the primary reason for the removal of `custom_attribute`) and requires registering the attribute explicitly.
It's not clear whether it should go through stabilization or not.
It's quite possible that all the uses should migrate to `#![register_tool]` (https://github.com/rust-lang/rust/issues/66079) instead.
---
Details:
- The naming is `register_attr`/`register_tool` rather than some `register_attributes` (plural, no abbreviation) for consistency with already existing attributes like `cfg_attr`, or `feature`, etc.
---
Previous attempt: https://github.com/rust-lang/rust/pull/57921
cc https://github.com/rust-lang/rust/issues/44690
Tracking issues: #66079 (`register_tool`), #66080 (`register_attr`)
Closes https://github.com/rust-lang/rust/issues/29642
Split libsyntax apart
In this PR the general idea is to separate the AST, parser, and friends by a more data / logic structure (tho not fully realized!) by separating out the parser and macro expansion code from libsyntax. Specifically have now three crates instead of one (libsyntax):
- libsyntax:
- concrete syntax tree (`syntax::ast`)
- definition of tokens and token-streams (`syntax::{token, tokenstream}`) -- used by `syntax::ast`
- visitors (`syntax::visit`, `syntax::mut_visit`)
- shared definitions between `libsyntax_expand`
- feature gating (`syntax::feature_gate`) -- we could possibly move this out to its own crater later.
- attribute and meta item utilities, including used-marking (`syntax::attr`)
- pretty printer (`syntax::print`) -- this should possibly be moved out later. For now I've reduced down the dependencies to a single essential one which could be broken via `ParseSess`. This entails that e.g. `Debug` impls for `Path` cannot reference the pretty printer.
- definition of `ParseSess` (`syntax::sess`) -- this is used by `syntax::{attr, print, feature_gate}` and is a common definition used by the parser and other things like librustc.
- the `syntax::source_map` -- this includes definitions used by `syntax::ast` and other things but could ostensibly be moved `syntax_pos` since that is more related to this module.
- a smattering of misc utilities not sufficiently important to itemize -- some of these could be moved to where they are used (often a single place) but I wanted to limit the scope of this PR.
- librustc_parse:
- parser (`rustc_parse::parser`) -- reading a file and such are defined in the crate root tho.
- lexer (`rustc_parse::lexer`)
- validation of meta grammar (post-expansion) in (`rustc_parse::validate_attr`)
- libsyntax_expand -- this defines the infra for macro expansion and conditional compilation but this is not libsyntax_ext; we might want to merge them later but currently libsyntax_expand is depended on by librustc_metadata which libsyntax_ext is not.
- conditional compilation (`syntax_expand::config`) -- moved from `syntax::config` to here
- the bulk of this crate is made up of the old `syntax::ext`
r? @estebank
Move next node ID to Resolver
This moves the `next_node_id` method(s) and related tracking information to the resolver. By doing so, we also remove the OneThread and Cell on next_node_id in Session in this move, which means that the new code is simpler and less "interesting" as it doesn't tie itself to a single thread.
This required moving some of the pretty-printing logic around, but this was just copying the code without any semantic changes, so it's just a second commit instead of a separate PR; I can polish it up a bit more if desired.
rustc_metadata: don't let LLVM confuse rmeta blobs for COFF object files.
This has likely been a silent issue since 1.10 but only caused trouble recently (see https://github.com/rust-lang/rust/issues/65536#issuecomment-552018224), when recent changes to the `rmeta` schema introduced more opportunities for COFF parse errors.
To prevent any undesired interactions with old compilers, I've renamed the file inside `rlib`s from `rust.metadata.bin` to `lib.rmeta` (not strongly attached to it, suggestions welcome).
Fixes#65536.
<hr/>
Before:
```
$ llvm-objdump -all-headers build/*/stage1-std/*/release/deps/libcore-*.rmeta
build/x86_64-unknown-linux-gnu/stage1-std/x86_64-unknown-linux-gnu/release/deps/libcore-6b9e8b5a59b79a1d.rmeta: file format COFF-<unknown arch>
architecture: unknown
start address: 0x00000000
Sections:
Idx Name Size VMA Type
SYMBOL TABLE:
```
After:
```
$ llvm-objdump -all-headers build/*/stage1-std/*/release/deps/libcore-*.rmeta
llvm-objdump: error: 'build/x86_64-unknown-linux-gnu/stage1-std/x86_64-unknown-linux-gnu/release/deps/libcore-6b9e8b5a59b79a1d.rmeta':
The file was not recognized as a valid object file
```
invalid_value lint: use diagnostic items
This adjusts the invalid_value lint to use diagnostic items.
@Centril @oli-obk For some reason, this fails to recognize `transmute` -- somehow the diagnostic item is not found. Any idea why?
r? @Centril
Cc https://github.com/rust-lang/rust/issues/66075
[mir-opt] Handle return place in ConstProp and improve SimplifyLocals pass
Temporarily rebased on top of #66074. The top 2 commits are new.
r? @oli-obk
Correct deprecated `is_global` IPv6 documentation
This method does currently not return false for the `site_local` unicast address space. The documentation of the `is_unicast_global` method on lines 1352 - 1382 suggests that this is intentional as the site-local prefix must no longer be supported in new implementations, thus the documentation can safely be updated to reflect that information.
If not so, either the `is_unicast_global` method should be updated to exclude the unicast site-local address space, or the `is_global` method itself.
Correct error in documentation for Ipv4Addr method
Correct statement in doctests on line 539 of `is_global` method of the `Ipv4Addr` object, which falsely attributed the tests to the broadcast address.
Refactor sync::Once
`std::sync::Once` contains some tricky code to park and unpark waiting threads. [once_cell](https://github.com/matklad/once_cell) has very similar code copied from here. I tried to add more comments and refactor the code to make it more readable (at least in my opinion). My PR to `once_cell` was rejected, because it is an advantage to remain close to the implementation in std, and because I made a mess of the atomic orderings. So now a PR here, with similar changes to `std::sync::Once`!
The initial goal was to see if there is some way to detect reentrant initialization instead of deadlocking. No luck there yet, but you first have to understand and document the complexities of the existing code 😄.
*Maybe not this entire PR will be acceptable, but I hope at least some of the commits can be useful.*
Individual commits:
#### Rename state to state_and_queue
Just a more accurate description, although a bit wordy. It helped me on a first read through the code, where before `state` was use to encode a pointer in to nodes of a linked list.
#### Simplify loop conditions in RUNNING and add comments
In the relevant loop there are two things to be careful about:
- make sure not to enqueue the current thread only while still RUNNING, otherwise we will never be woken up (the status may have changed while trying to enqueue this thread).
- pick up if another thread just replaced the head of the linked list.
Because the first check was part of the condition of the while loop, the rest of the parking code also had to live in that loop. It took me a while to get the subtlety here, and it should now be clearer.
Also call out that we really have to wait until signaled, otherwise we leave a dangling reference.
#### Don't mutate waiter nodes
Previously while waking up other threads the managing thread would `take()` out the `Thread` struct and use that to unpark the other thread. It is just as easy to clone it, just 24 bytes. This way `Waiter.thread` does not need an `Option`, `Waiter.next` does not need to be a mutable pointer, and there is less data that needs to be synchronised by later atomic operations.
#### Turn Finish into WaiterQueue
In my opinion these changes make it just a bit more clear what is going on with the thread parking stuff.
#### Move thread parking to a seperate function
Maybe controversial, but with this last commit all the thread parking stuff has a reasonably clean seperation from the state changes in `Once`. This is arguably the trickier part of `Once`, compared to the loop in `call_inner`. It may make it easier to reuse parts of this code (see https://github.com/rust-lang/rfcs/pull/2788#discussion_r336729695). Not sure if that ever becomes a reality though.
#### Reduce the amount of comments in call_inner
With the changes from the previous commits, the code pretty much speaks for itself, and the amount of comments is hurting readability a bit.
#### Use more precise atomic orderings
Now the hard one. This is the one change that is not anything but a pure refactor or change of comments.
I have a dislike for using `SeqCst` everywhere, because it hides what the atomics are supposed to do. the rationale was:
> This cold path uses SeqCst consistently because the performance difference really does not matter there, and SeqCst minimizes the chances of something going wrong.
But in my opinion, having the proper orderings and some explanation helps to understand what is going on. My rationale for the used orderings (also included as comment):
When running `Once` we deal with multiple atomics: `Once.state_and_queue` and an unknown number of `Waiter.signaled`.
* `state_and_queue` is used (1) as a state flag, (2) for synchronizing the data that is the result of the `Once`, and (3) for synchronizing `Waiter` nodes.
- At the end of the `call_inner` function we have to make sure the result of the `Once` is acquired. So every load which can be the only one to load COMPLETED must have at least Acquire ordering, which means all three of them.
- `WaiterQueue::Drop` is the only place that may store COMPLETED, and must do so with Release ordering to make result available.
- `wait` inserts `Waiter` nodes as a pointer in `state_and_queue`, and needs to make the nodes available with Release ordering. The load in its `compare_and_swap` can be Relaxed because it only has to compare the atomic, not to read other data.
- `WaiterQueue::Drop` must see the `Waiter` nodes, so it must load `state_and_queue` with Acquire ordering.
- There is just one store where `state_and_queue` is used only as a state flag, without having to synchronize data: switching the state from INCOMPLETE to RUNNING in `call_inner`. This store can be Relaxed, but the read has to be Acquire because of the requirements mentioned above.
* `Waiter.signaled` is both used as a flag, and to protect a field with interior mutability in `Waiter`. `Waiter.thread` is changed in `WaiterQueue::Drop` which then sets `signaled` with Release ordering. After `wait` loads `signaled` with Acquire and sees it is true, it needs to see the changes to drop the `Waiter` struct correctly.
* There is one place where the two atomics `Once.state_and_queue` and `Waiter.signaled` come together, and might be reordered by the compiler or processor. Because both use Aquire ordering such a reordering is not allowed, so no need for SeqCst.
cc @matklad
[mir-opt] Implement pass to remove branches on uninhabited variants
Based on discussion [here](https://github.com/rust-lang/rust/pull/64890#discussion_r333612125), this is a pass to eliminate dead code that is caused by branching on an enum with uninhabited variants.
r? @oli-obk
This function was only ever called with 1 so there's little point in it;
this isn't an expensive operation (essentially a checked add) so we're
not really "reserving" anything either.
Rollup of 6 pull requests
Successful merges:
- #65949 (Move promotion into its own pass)
- #65994 (Point at where clauses where the associated item was restricted)
- #66050 (Fix C aggregate-passing ABI on powerpc)
- #66134 (Point at formatting descriptor string when it is invalid)
- #66172 (Stabilize @file command line arguments)
- #66226 (add link to unstable book for asm! macro)
Failed merges:
r? @ghost
