Skip to content

Commit

Permalink
Auto merge of #134748 - DianQK:rollup-3y5fzcx, r=DianQK
Browse files Browse the repository at this point in the history
Rollup of 3 pull requests

Successful merges:

 - #134525 (Arbitrary self types v2: unstable doc updates.)
 - #134735 (Consider arm to diverge if guard diverges)
 - #134741 (Actually print all the relevant parts of a coroutine in verbose mode)

r? `@ghost`
`@rustbot` modify labels: rollup
  • Loading branch information
bors committed Dec 25, 2024
2 parents 41f2f5c + 1d10117 commit 68b9e4f
Show file tree
Hide file tree
Showing 9 changed files with 261 additions and 10 deletions.
12 changes: 10 additions & 2 deletions compiler/rustc_hir_typeck/src/_match.rs
Original file line number Diff line number Diff line change
Expand Up @@ -77,12 +77,20 @@ impl<'a, 'tcx> FnCtxt<'a, 'tcx> {
let mut prior_non_diverging_arms = vec![]; // Used only for diagnostics.
let mut prior_arm = None;
for arm in arms {
self.diverges.set(Diverges::Maybe);

if let Some(e) = &arm.guard {
self.diverges.set(Diverges::Maybe);
self.check_expr_has_type_or_error(e, tcx.types.bool, |_| {});

// FIXME: If this is the first arm and the pattern is irrefutable,
// e.g. `_` or `x`, and the guard diverges, then the whole match
// may also be considered to diverge. We should warn on all subsequent
// arms, too, just like we do for diverging scrutinees above.
}

self.diverges.set(Diverges::Maybe);
// N.B. We don't reset diverges here b/c we want to warn in the arm
// if the guard diverges, like: `x if { loop {} } => f()`, and we
// also want to consider the arm to diverge itself.

let arm_ty = self.check_expr_with_expectation(arm.body, expected);
all_arms_diverge &= self.diverges.get();
Expand Down
6 changes: 6 additions & 0 deletions compiler/rustc_middle/src/ty/print/pretty.rs
Original file line number Diff line number Diff line change
Expand Up @@ -841,6 +841,12 @@ pub trait PrettyPrinter<'tcx>: Printer<'tcx> + fmt::Write {
p!(
" upvar_tys=",
print(args.as_coroutine().tupled_upvars_ty()),
" resume_ty=",
print(args.as_coroutine().resume_ty()),
" yield_ty=",
print(args.as_coroutine().yield_ty()),
" return_ty=",
print(args.as_coroutine().return_ty()),
" witness=",
print(args.as_coroutine().witness())
);
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,57 @@
# `arbitrary_self_types_pointers`

The tracking issue for this feature is: [#44874]

[#38788]: https://github.com/rust-lang/rust/issues/44874

------------------------

This extends the [arbitrary self types] feature to allow methods to
receive `self` by pointer. For example:

```rust
#![feature(arbitrary_self_types_pointers)]

struct A;

impl A {
fn m(self: *const Self) {}
}

fn main() {
let a = A;
let a_ptr: *const A = &a as *const A;
a_ptr.m();
}
```

In general this is not advised: it's thought to be better practice to wrap
raw pointers in a newtype wrapper which implements the `core::ops::Receiver`
trait, then you need "only" the `arbitrary_self_types` feature. For example:

```rust
#![feature(arbitrary_self_types)]
#![allow(dead_code)]

struct A;

impl A {
fn m(self: Wrapper<Self>) {} // can extract the pointer and do
// what it needs
}

struct Wrapper<T>(*const T);

impl<T> core::ops::Receiver for Wrapper<T> {
type Target = T;
}

fn main() {
let a = A;
let a_ptr: *const A = &a as *const A;
let a_wrapper = Wrapper(a_ptr);
a_wrapper.m();
}
```

[arbitrary self types]: arbitrary-self-types.md
154 changes: 154 additions & 0 deletions src/doc/unstable-book/src/language-features/arbitrary-self-types.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,154 @@
# `arbitrary_self_types`

The tracking issue for this feature is: [#44874]

[#38788]: https://github.com/rust-lang/rust/issues/44874

------------------------

Allows any type implementing `core::ops::Receiver<Target=T>` to be used as the type
of `self` in a method belonging to `T`.

For example,

```rust
#![feature(arbitrary_self_types)]

struct A;

impl A {
fn f(self: SmartPtr<Self>) -> i32 { 1 } // note self type
}

struct SmartPtr<T>(T);

impl<T> core::ops::Receiver for SmartPtr<T> {
type Target = T;
}

fn main() {
let smart_ptr = SmartPtr(A);
assert_eq!(smart_ptr.f(), 1);
}
```

The `Receiver` trait has a blanket implementation for all `T: Deref`, so in fact
things like this work too:

```rust
#![feature(arbitrary_self_types)]

use std::rc::Rc;

struct A;

impl A {
fn f(self: Rc<Self>) -> i32 { 1 } // Rc implements Deref
}

fn main() {
let smart_ptr = Rc::new(A);
assert_eq!(smart_ptr.f(), 1);
}
```

Interestingly, that works even without the `arbitrary_self_types` feature
- but that's because certain types are _effectively_ hard coded, including
`Rc`. ("Hard coding" isn't quite true; they use a lang-item called
`LegacyReceiver` to denote their special-ness in this way). With the
`arbitrary_self_types` feature, their special-ness goes away, and custom
smart pointers can achieve the same.

## Changes to method lookup

Method lookup previously used to work by stepping through the `Deref`
chain then using the resulting list of steps in two different ways:

* To identify types that might contribute methods via their `impl`
blocks (inherent methods) or via traits
* To identify the types that the method receiver (`a` in the above
examples) can be converted to.

With this feature, these lists are created by instead stepping through
the `Receiver` chain. However, a note is kept about whether the type
can be reached also via the `Deref` chain.

The full chain (via `Receiver` hops) is used for the first purpose
(identifying relevant `impl` blocks and traits); whereas the shorter
list (reachable via `Deref`) is used for the second purpose. That's
because, to convert the method target (`a` in `a.b()`) to the self
type, Rust may need to be able to use `Deref::deref`. Type conversions,
then, can only proceed as far as the end of the `Deref` chain whereas
the longer `Receiver` chain can be used to explore more places where
useful methods might reside.

## Types suitable for use as smart pointers

This feature allows the creation of customised smart pointers - for example
your own equivalent to `Rc` or `Box` with whatever capabilities you like.
Those smart pointers can either implement `Deref` (if it's safe to
create a reference to the referent) or `Receiver` (if it isn't).

Either way, smart pointer types should mostly _avoid having methods_.
Calling methods on a smart pointer leads to ambiguity about whether you're
aiming for a method on the pointer, or on the referent.

Best practice is therefore to put smart pointer functionality into
associated functions instead - that's what's done in all the smart pointer
types within Rust's standard library which implement `Receiver`.

If you choose to add any methods to your smart pointer type, your users
may run into errors from deshadowing, as described in the next section.

## Avoiding shadowing

With or without this feature, Rust emits an error if it finds two method
candidates, like this:

```rust,compile_fail
use std::pin::Pin;
use std::pin::pin;
struct A;
impl A {
fn get_ref(self: Pin<&A>) {}
}
fn main() {
let pinned_a: Pin<&A> = pin!(A).as_ref();
let pinned_a: Pin<&A> = pinned_a.as_ref();
pinned_a.get_ref(); // error[E0034]: multiple applicable items in scope
}
```

(this is why Rust's smart pointers are mostly carefully designed to avoid
having methods at all, and shouldn't add new methods in future.)

With `arbitrary_self_types`, we take care to spot some other kinds of
conflict:

```rust,compile_fail
#![feature(arbitrary_self_types)]
use std::pin::Pin;
use std::pin::pin;
struct A;
impl A {
fn get_ref(self: &Pin<&A>) {} // note &Pin
}
fn main() {
let pinned_a: Pin<&mut A> = pin!(A);
let pinned_a: Pin<&A> = pinned_a.as_ref();
pinned_a.get_ref();
}
```

This is to guard against the case where an inner (referent) type has a
method of a given name, taking the smart pointer by reference, and then
the smart pointer implementer adds a similar method taking self by value.
As noted in the previous section, the safe option is simply
not to add methods to smart pointers, and then these errors can't occur.
4 changes: 2 additions & 2 deletions tests/ui/async-await/async-closures/def-path.stderr
Original file line number Diff line number Diff line change
Expand Up @@ -5,11 +5,11 @@ LL | let x = async || {};
| -- the expected `async` closure body
LL |
LL | let () = x();
| ^^ --- this expression has type `{static main::{closure#0}::{closure#0}<?17t> upvar_tys=?16t witness=?6t}`
| ^^ --- this expression has type `{static main::{closure#0}::{closure#0}<?17t> upvar_tys=?16t resume_ty=ResumeTy yield_ty=() return_ty=() witness=?6t}`
| |
| expected `async` closure body, found `()`
|
= note: expected `async` closure body `{static main::{closure#0}::{closure#0}<?17t> upvar_tys=?16t witness=?6t}`
= note: expected `async` closure body `{static main::{closure#0}::{closure#0}<?17t> upvar_tys=?16t resume_ty=ResumeTy yield_ty=() return_ty=() witness=?6t}`
found unit type `()`

error: aborting due to 1 previous error
Expand Down
4 changes: 2 additions & 2 deletions tests/ui/coroutine/print/coroutine-print-verbose-2.stderr
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@ LL | | drop(a);
LL | | });
| |______^ coroutine is not `Sync`
|
= help: within `{main::{closure#0} upvar_tys=() witness={main::{closure#0}}}`, the trait `Sync` is not implemented for `NotSync`
= help: within `{main::{closure#0} upvar_tys=() resume_ty=() yield_ty=() return_ty=() witness={main::{closure#0}}}`, the trait `Sync` is not implemented for `NotSync`
note: coroutine is not `Sync` as this value is used across a yield
--> $DIR/coroutine-print-verbose-2.rs:20:9
|
Expand All @@ -34,7 +34,7 @@ LL | | drop(a);
LL | | });
| |______^ coroutine is not `Send`
|
= help: within `{main::{closure#1} upvar_tys=() witness={main::{closure#1}}}`, the trait `Send` is not implemented for `NotSend`
= help: within `{main::{closure#1} upvar_tys=() resume_ty=() yield_ty=() return_ty=() witness={main::{closure#1}}}`, the trait `Send` is not implemented for `NotSend`
note: coroutine is not `Send` as this value is used across a yield
--> $DIR/coroutine-print-verbose-2.rs:27:9
|
Expand Down
2 changes: 1 addition & 1 deletion tests/ui/coroutine/print/coroutine-print-verbose-3.stderr
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@ LL | | };
| |_____^ expected `()`, found coroutine
|
= note: expected unit type `()`
found coroutine `{main::{closure#0} upvar_tys=?4t witness=?6t}`
found coroutine `{main::{closure#0} upvar_tys=?4t resume_ty=() yield_ty=i32 return_ty=&'?1 str witness=?6t}`

error: aborting due to 1 previous error

Expand Down
8 changes: 6 additions & 2 deletions tests/ui/reachable/expr_match.rs
Original file line number Diff line number Diff line change
Expand Up @@ -21,9 +21,13 @@ fn d() {
}

fn e() {
// Here the compiler fails to figure out that the `println` is dead.
match () { () if return => (), () => return }
match () {
() if return => (),
//~^ ERROR unreachable expression
() => return,
}
println!("I am dead");
//~^ ERROR unreachable statement
}

fn f() {
Expand Down
24 changes: 23 additions & 1 deletion tests/ui/reachable/expr_match.stderr
Original file line number Diff line number Diff line change
Expand Up @@ -23,5 +23,27 @@ LL | println!("I am dead");
|
= note: this error originates in the macro `println` (in Nightly builds, run with -Z macro-backtrace for more info)

error: aborting due to 2 previous errors
error: unreachable expression
--> $DIR/expr_match.rs:25:25
|
LL | () if return => (),
| ------ ^^ unreachable expression
| |
| any code following this expression is unreachable

error: unreachable statement
--> $DIR/expr_match.rs:29:5
|
LL | / match () {
LL | | () if return => (),
LL | |
LL | | () => return,
LL | | }
| |_____- any code following this `match` expression is unreachable, as all arms diverge
LL | println!("I am dead");
| ^^^^^^^^^^^^^^^^^^^^^ unreachable statement
|
= note: this error originates in the macro `println` (in Nightly builds, run with -Z macro-backtrace for more info)

error: aborting due to 4 previous errors

0 comments on commit 68b9e4f

Please sign in to comment.