]> git.lizzy.rs Git - rust.git/log
rust.git
2 years agoAdd the new stability attributes, for Windows.
Dan Gohman [Wed, 15 Jun 2022 16:46:56 +0000 (09:46 -0700)]
Add the new stability attributes, for Windows.

2 years agoRevise the documentation for `try_clone`.
Dan Gohman [Thu, 19 May 2022 13:59:12 +0000 (06:59 -0700)]
Revise the documentation for `try_clone`.

On Unix, describe these in terms of the underlying "file description". On
Windows, describe them in terms of the underlying "object".

2 years agoAdd `BorrowedFd::try_clone_to_owned`.
Dan Gohman [Thu, 19 May 2022 13:43:37 +0000 (06:43 -0700)]
Add `BorrowedFd::try_clone_to_owned`.

And `BorrowedHandle::try_clone_to_owned` and
`BorrowedSocket::try_clone_to_owned` on Windows.

2 years agoDocument that `BorrowedFd` may be used to do a `dup`.
Dan Gohman [Thu, 19 May 2022 13:41:35 +0000 (06:41 -0700)]
Document that `BorrowedFd` may be used to do a `dup`.

2 years agodebuginfo: Fix NatVis for Rc and Arc with unsized pointees.
Michael Woerister [Wed, 15 Jun 2022 14:49:52 +0000 (16:49 +0200)]
debuginfo: Fix NatVis for Rc and Arc with unsized pointees.

2 years agoAuto merge of #95897 - AzureMarker:feature/horizon-std, r=nagisa
bors [Wed, 15 Jun 2022 14:21:28 +0000 (14:21 +0000)]
Auto merge of #95897 - AzureMarker:feature/horizon-std, r=nagisa

STD support for the Nintendo 3DS

Rustc already supports compiling for the Nintendo 3DS using the `armv6k-nintendo-3ds` target (Tier 3). Until now though, only `core` and `alloc` were supported. This PR adds standard library support for the Nintendo 3DS. A notable exclusion is `std::thread` support, which will come in a follow-up PR as it requires more complicated changes.

This has been a joint effort by `@Meziu,` `@ian-h-chamberlain,` myself, and prior work by `@rust3ds` members.

### Background

The Nintendo 3DS (Horizon OS) is a mostly-UNIX looking system, with the caveat that it does not come with a full libc implementation out of the box. On the homebrew side (I'm not under NDA), the libc interface is partially implemented by the [devkitPro](https://devkitpro.org/wiki/devkitPro_pacman) toolchain and a user library like [`libctru`](https://github.com/devkitPro/libctru). This is important because there are [some possible legal barriers](https://github.com/rust-lang/rust/pull/88529#issuecomment-919938396) to linking directly to a library that uses the underlying platform APIs, since they might be considered a trade secret or under NDA.

To get around this, the standard library impl for the 3DS does not directly depend on any platform-level APIs. Instead, it expects standard libc functions to be linked in. The implementation of these libc functions is left to the user. Some functions are provided by the devkitPro toolchain, but in our testing, we used the following to fill in the other functions:
- [`libctru`] - provides more basic APIs, such as `nanosleep`. Linked in by way of [`ctru-sys`](https://github.com/Meziu/ctru-rs/tree/master/ctru-sys).
- [`pthread-3ds`](https://github.com/Meziu/pthread-3ds) - provides pthread APIs for `std::thread`. Implemented using [`libctru`].
- [`linker-fix-3ds`](https://github.com/Meziu/rust-linker-fix-3ds) - fulfills some other missing libc APIs. Implemented using [`libctru`].

For more details, see the `src/doc/rustc/src/platform-support/armv6k-nintendo-3ds.md` file added in this PR.

### Notes
We've already upstreamed changes to the [`libc`] crate to support this PR, as well as the upcoming threading PR. These changes have all been released as of 0.2.121, so we bump the crate version in this PR.
Edit: After some rebases, the version bump has already been merged so it doesn't appear in this PR.

A lot of the changes in this PR are straightforward, and follow in the footsteps of the ESP-IDF target: https://github.com/rust-lang/rust/pull/87666.

The 3DS does not support user space process spawning, so these APIs are unimplemented (similar to ESP-IDF).

[`libctru`]: https://github.com/devkitPro/libctru
[`libc`]: https://github.com/rust-lang/libc

2 years agoAdd regression test for #93775
Yuki Okushi [Wed, 15 Jun 2022 12:31:47 +0000 (21:31 +0900)]
Add regression test for #93775

Signed-off-by: Yuki Okushi <jtitor@2k36.org>
2 years agoAuto merge of #98131 - JohnTitor:rollup-c17vjdy, r=JohnTitor
bors [Wed, 15 Jun 2022 11:04:23 +0000 (11:04 +0000)]
Auto merge of #98131 - JohnTitor:rollup-c17vjdy, r=JohnTitor

Rollup of 3 pull requests

Successful merges:

 - #95118 (Implement stabilization of `#[feature(io_safety)]`.)
 - #98110 (Make `ExprKind::Closure` a struct variant.)
 - #98115 (Remove `rustc_deprecated` diagnostics)

Failed merges:

r? `@ghost`
`@rustbot` modify labels: rollup

2 years ago[RFC 2011] Minimal initial implementation
Caio [Wed, 15 Jun 2022 10:37:40 +0000 (07:37 -0300)]
[RFC 2011] Minimal initial implementation

2 years agoRollup merge of #98115 - jhpratt:remove-rustc_deprecated, r=compiler-errors
Yuki Okushi [Wed, 15 Jun 2022 10:37:15 +0000 (19:37 +0900)]
Rollup merge of #98115 - jhpratt:remove-rustc_deprecated, r=compiler-errors

Remove `rustc_deprecated` diagnostics

Follow-up on #95960. The diagnostics will remain until the next bootstrap, at which point people will have had six weeks to adjust.

``@rustbot`` label +A-diagnostics

r? ``@compiler-errors``

2 years agoRollup merge of #98110 - cjgillot:closure-brace, r=Aaron1011
Yuki Okushi [Wed, 15 Jun 2022 10:37:14 +0000 (19:37 +0900)]
Rollup merge of #98110 - cjgillot:closure-brace, r=Aaron1011

Make `ExprKind::Closure` a struct variant.

Simple refactor since we both need it to introduce additional fields in `ExprKind::Closure`.

r? ``@Aaron1011``

2 years agoRollup merge of #98110 - cjgillot:closure-brace, r=Aaron1011
Yuki Okushi [Wed, 15 Jun 2022 10:37:14 +0000 (19:37 +0900)]
Rollup merge of #98110 - cjgillot:closure-brace, r=Aaron1011

Make `ExprKind::Closure` a struct variant.

Simple refactor since we both need it to introduce additional fields in `ExprKind::Closure`.

r? ``@Aaron1011``

2 years agoRollup merge of #95118 - sunfishcode:sunfishcode/stabilize-io-safety, r=joshtriplett
Yuki Okushi [Wed, 15 Jun 2022 10:37:13 +0000 (19:37 +0900)]
Rollup merge of #95118 - sunfishcode:sunfishcode/stabilize-io-safety, r=joshtriplett

Implement stabilization of `#[feature(io_safety)]`.

Implement stabilization of [I/O safety], aka `#[feature(io_safety)]`.

Fixes #87074.

[I/O safety]: https://github.com/rust-lang/rfcs/blob/master/text/3128-io-safety.md

2 years agoFix false positive for `never_loop` struct expression fields
andylizi [Wed, 15 Jun 2022 09:54:43 +0000 (17:54 +0800)]
Fix false positive for `never_loop` struct expression fields

2 years agoTest NLL fix of bad lifetime inference for reference captured in closure.
Frank Steffahn [Wed, 15 Jun 2022 01:12:07 +0000 (03:12 +0200)]
Test NLL fix of bad lifetime inference for reference captured in closure.

2 years agoAuto merge of #98084 - nnethercote:rm-thread-local-IGNORED_ATTRIBUTES, r=michaelwoerister
bors [Wed, 15 Jun 2022 08:20:19 +0000 (08:20 +0000)]
Auto merge of #98084 - nnethercote:rm-thread-local-IGNORED_ATTRIBUTES, r=michaelwoerister

Remove thread-local `IGNORED_ATTRIBUTES`.

It's just a copy of the read-only global `ich::IGNORED_ATTRIBUTES`, and
can be removed without any effect.

r? `@michaelwoerister`

2 years agochange "1" to "c" to pass test
KarlWithK [Wed, 15 Jun 2022 08:19:22 +0000 (03:19 -0500)]
change "1" to "c" to pass test

Incorrectly wrote "1" twice when writing test.

2 years agoAdd examples using `add_modify` to btree
KarlWithK [Wed, 15 Jun 2022 07:04:18 +0000 (02:04 -0500)]
Add examples using `add_modify` to btree

Updated the btree's documentation to include two references to
add_modify.

The first is when the `Entry` API is mentioned at the beginning. With
the same reasoning as HashMap's documentation, I thought it would best
to keep `attack`, but show the `mana` example.

The second is with the `entry` function that is used for the `Entry`
API. The code example was a perfect use for `add_modify`, which is why
it was changed to reflect that.

2 years agoAdd examples using `add_modify` to HashMap
KarlWithK [Wed, 15 Jun 2022 06:41:56 +0000 (01:41 -0500)]
Add examples using `add_modify` to HashMap

Updated the HashMap's documentation to include two references to
add_modify.

The first is when the `Entry` API is mentioned at the beginning. I was
hesitant to change the "attack" example (although I believe that it is
perfect example of where `add_modify` should be used) because both uses
work equally, but one is more idiomatic (`add_modify`).

The second is with the `entry` function that is used for the `Entry`
API. The code example was a perfect use for `add_modify`, which is why
it was changed to reflect that.

2 years agoAuto merge of #8964 - tamaroning:read_zero_byte_vec, r=dswij
bors [Wed, 15 Jun 2022 06:16:14 +0000 (06:16 +0000)]
Auto merge of #8964 - tamaroning:read_zero_byte_vec, r=dswij

Warn about read into zero-length `Vec`

Closes #8886

- \[x] Followed [lint naming conventions][lint_naming]
- \[x] Added passing UI tests (including committed `.stderr` file)
- \[x] `cargo test` passes locally
- \[x] Executed `cargo dev update_lints`
- \[x] Added lint documentation
- \[x] Run `cargo dev fmt`

changelog: none

2 years agoAuto merge of #98123 - JohnTitor:rollup-rfg1a4s, r=JohnTitor
bors [Wed, 15 Jun 2022 05:39:29 +0000 (05:39 +0000)]
Auto merge of #98123 - JohnTitor:rollup-rfg1a4s, r=JohnTitor

Rollup of 7 pull requests

Successful merges:

 - #97822 (Filter out intrinsics if we have other import candidates to suggest)
 - #98026 (Move some tests to more reasonable directories)
 - #98067 (compiler: remove unused deps)
 - #98078 (Use unchecked mul to compute slice sizes)
 - #98083 (Rename rustc_serialize::opaque::Encoder as MemEncoder.)
 - #98087 (Suggest adding a `#[macro_export]` to a private macro)
 - #98113 (Fix misspelling of "constraint" as "contraint")

Failed merges:

r? `@ghost`
`@rustbot` modify labels: rollup

2 years agoRollup merge of #98113 - EdwinRy:constraint-misspelling, r=compiler-errors
Yuki Okushi [Wed, 15 Jun 2022 03:02:06 +0000 (12:02 +0900)]
Rollup merge of #98113 - EdwinRy:constraint-misspelling, r=compiler-errors

Fix misspelling of "constraint" as "contraint"

I misspelled a function name a while back

2 years agoRollup merge of #98087 - TaKO8Ki:suggest-adding-macro-export, r=oli-obk
Yuki Okushi [Wed, 15 Jun 2022 03:02:05 +0000 (12:02 +0900)]
Rollup merge of #98087 - TaKO8Ki:suggest-adding-macro-export, r=oli-obk

Suggest adding a `#[macro_export]` to a private macro

fixes #97628

2 years agoRollup merge of #98083 - nnethercote:rename-Encoder, r=bjorn3
Yuki Okushi [Wed, 15 Jun 2022 03:02:04 +0000 (12:02 +0900)]
Rollup merge of #98083 - nnethercote:rename-Encoder, r=bjorn3

Rename rustc_serialize::opaque::Encoder as MemEncoder.

This avoids the name clash with `rustc_serialize::Encoder` (a trait),
and allows lots qualifiers to be removed and imports to be simplified
(e.g. fewer `as` imports).

(This was previously merged as commit 5 in #94732 and then was reverted
in #97905 because of a perf regression caused by commit 4 in #94732.)

r? ```@bjorn3```

2 years agoRollup merge of #98078 - erikdesjardins:uncheckedsize, r=petrochenkov
Yuki Okushi [Wed, 15 Jun 2022 03:02:03 +0000 (12:02 +0900)]
Rollup merge of #98078 - erikdesjardins:uncheckedsize, r=petrochenkov

Use unchecked mul to compute slice sizes

This allows LLVM to realize that `slice.len() > 0` iff `slice.len() * size_of::<T>() > 0`, allowing a branch on the latter to be folded into the former when dropping vecs and boxed slices, in some cases.

Fixes (partially) #96497

2 years agoRollup merge of #98067 - klensy:compiler-deps2, r=Dylan-DPC
Yuki Okushi [Wed, 15 Jun 2022 03:02:02 +0000 (12:02 +0900)]
Rollup merge of #98067 - klensy:compiler-deps2, r=Dylan-DPC

compiler: remove unused deps

Removed unused dependencies in compiler crates and moves few `libc` under `target.cfg(unix)` .

2 years agoRollup merge of #98026 - c410-f3r:z-errors, r=petrochenkov
Yuki Okushi [Wed, 15 Jun 2022 03:02:02 +0000 (12:02 +0900)]
Rollup merge of #98026 - c410-f3r:z-errors, r=petrochenkov

Move some tests to more reasonable directories

r? ```@petrochenkov```

2 years agoRollup merge of #97822 - compiler-errors:hesitate-to-suggest-intrinsics, r=oli-obk
Yuki Okushi [Wed, 15 Jun 2022 03:02:01 +0000 (12:02 +0900)]
Rollup merge of #97822 - compiler-errors:hesitate-to-suggest-intrinsics, r=oli-obk

Filter out intrinsics if we have other import candidates to suggest

Fixes #97618

Also open to just sorting these candidates to be last. Pretty easy to modify the code to do that, too.

2 years agoAuto merge of #97474 - compiler-errors:if-cond-and-block, r=oli-obk
bors [Wed, 15 Jun 2022 02:58:44 +0000 (02:58 +0000)]
Auto merge of #97474 - compiler-errors:if-cond-and-block, r=oli-obk

Improve parsing errors and suggestions for bad `if` statements

1. Parses `if {}` as `if <err> {}` (block-like conditions that are missing a "then" block), and `if true && {}` as `if true && <err> {}` (unfinished binary operation), which is a more faithful recovery and leads to better typeck errors later on.
1. Points out the span of the condition if we don't see a "then" block after it, to help the user understand what is being parsed as a condition (and by elimination, what isn't).
1. Allow `if cond token else { }` to be fixed properly to `if cond { token } else { }`.
1. Fudge with the error messages a bit. This is somewhat arbitrary and I can revert my rewordings if they're useless.

----

Also this PR addresses a strange parsing regression (1.20 -> 1.21) where we chose to reject this piece of code somewhat arbitrarily, even though we should parse it fine:

```rust
fn main() {
    if { if true { return } else { return }; } {}
}
```

For context, all of these other expressions parse correctly:

```rust
fn main() {
    if { if true { return } else { return } } {}
    if { return; } {}
    if { return } {}
    if { return if true { } else { }; } {}
}
```

The parser used a heuristic to determine if the "the parsed `if` condition makes sense as a condition" that did like a one-expr-deep reachability analysis. This should not be handled by the parser though.

2 years ago`BitSet` perf improvements
Jakob Degen [Sun, 5 Jun 2022 23:22:54 +0000 (16:22 -0700)]
`BitSet` perf improvements

This commit makes two changes:
 1. Changes `MaybeLiveLocals` to use `ChunkedBitSet`
 2. Overrides the `fold` method for the iterator for `ChunkedBitSet`

2 years agoRefactor path segment parameter error
EdwinRy [Wed, 15 Jun 2022 01:50:34 +0000 (02:50 +0100)]
Refactor path segment parameter error

2 years agorename function and remove return type
EdwinRy [Wed, 15 Jun 2022 00:06:40 +0000 (01:06 +0100)]
rename function and remove return type

2 years agoRemove `rustc_deprecated` diagnostics
Jacob Pratt [Tue, 14 Jun 2022 23:11:24 +0000 (19:11 -0400)]
Remove `rustc_deprecated` diagnostics

2 years agoAdd a stability attribute to WASI's `try_clone()`.
Dan Gohman [Tue, 14 Jun 2022 21:46:10 +0000 (14:46 -0700)]
Add a stability attribute to WASI's `try_clone()`.

2 years agoImplement stabilization of `#[feature(io_safety)]`.
Dan Gohman [Tue, 8 Mar 2022 23:38:08 +0000 (15:38 -0800)]
Implement stabilization of `#[feature(io_safety)]`.

Implement stabilization of [I/O safety], aka `#[feature(io_safety)]`.

Fixes #87074.

[I/O safety]: https://github.com/rust-lang/rfcs/blob/master/text/3128-io-safety.md

2 years agoAuto merge of #96285 - flip1995:pk-vfe, r=nagisa
bors [Tue, 14 Jun 2022 21:37:11 +0000 (21:37 +0000)]
Auto merge of #96285 - flip1995:pk-vfe, r=nagisa

Introduce `-Zvirtual-function-elimination` codegen flag

Fixes #68262

This PR adds a codegen flag `-Zvirtual-function-elimination` to enable the VFE optimization in LLVM. To make this work, additonal  information has to be added to vtables ([`!vcall_visibility` metadata](https://llvm.org/docs/TypeMetadata.html#vcall-visibility-metadata) and a `typeid` of the trait). Furthermore, instead of just `load`ing functions, the [`llvm.type.checked.load` intrinsic](https://llvm.org/docs/LangRef.html#llvm-type-checked-load-intrinsic) has to be used to map functions to vtables.

For technical details of the changes, see the commit messages.

I also tested this flag on https://github.com/tock/tock on different boards to verify that this fixes the issue https://github.com/tock/tock/issues/2594. This flag is able to improve the size of the resulting binary by about 8k-9k bytes by removing the unused debug print functions.

[Rendered documentation update](https://github.com/flip1995/rust/blob/pk-vfe/src/doc/rustc/src/codegen-options/index.md#virtual-function-elimination)

2 years agorustdoc: add test case for "variadic tuple" search notation
Michael Howell [Tue, 14 Jun 2022 21:17:20 +0000 (14:17 -0700)]
rustdoc: add test case for "variadic tuple" search notation

2 years agouse unchecked mul to compute slice sizes
Erik Desjardins [Tue, 14 Jun 2022 01:49:59 +0000 (21:49 -0400)]
use unchecked mul to compute slice sizes

...since slice sizes can't signed wrap

see https://doc.rust-lang.org/std/slice/fn.from_raw_parts.html

> The total size len * mem::size_of::<T>() of the slice must be no larger than isize::MAX.

2 years agoSeparate `source_span` and `expn_that_defined` from `Definitions`.
Camille GILLOT [Mon, 30 May 2022 16:49:17 +0000 (18:49 +0200)]
Separate `source_span` and `expn_that_defined` from `Definitions`.

2 years agoDo not modify the resolver outputs.
Camille GILLOT [Sun, 18 Jul 2021 18:09:20 +0000 (20:09 +0200)]
Do not modify the resolver outputs.

2 years agoMake ResolverAstLowering a struct.
Camille GILLOT [Mon, 5 Jul 2021 20:26:23 +0000 (22:26 +0200)]
Make ResolverAstLowering a struct.

2 years agoSeparate Definitions and CrateStore from ResolverOutputs.
Camille GILLOT [Tue, 13 Jul 2021 22:48:51 +0000 (00:48 +0200)]
Separate Definitions and CrateStore from ResolverOutputs.

2 years agorustdoc: change "variadic tuple" notation to look less like real syntax
Michael Howell [Tue, 14 Jun 2022 19:21:38 +0000 (12:21 -0700)]
rustdoc: change "variadic tuple" notation to look less like real syntax

2 years agoAdd GUI test for sidebar items expand/collapse
Guillaume Gomez [Tue, 14 Jun 2022 09:21:28 +0000 (11:21 +0200)]
Add GUI test for sidebar items expand/collapse

2 years agobtreemap-alloc: adjust ui test
Jacob Hughes [Mon, 14 Mar 2022 10:58:58 +0000 (06:58 -0400)]
btreemap-alloc: adjust ui test

2 years agobtreemap-alloc: fix clear impl
Jacob Hughes [Mon, 14 Mar 2022 10:55:36 +0000 (06:55 -0400)]
btreemap-alloc: fix clear impl

2 years agoBTreeMap: Add alloc param
Jacob Hughes [Thu, 10 Mar 2022 09:55:24 +0000 (04:55 -0500)]
BTreeMap: Add alloc param

2 years agoAuto merge of #96591 - b-naber:transition-to-valtrees-in-type-system, r=lcnr
bors [Tue, 14 Jun 2022 17:19:38 +0000 (17:19 +0000)]
Auto merge of #96591 - b-naber:transition-to-valtrees-in-type-system, r=lcnr

Use valtrees as the type-system representation for constant values

This is not quite ready yet, there are still some problems with pretty printing and symbol mangling and `deref_const` seems to not work correctly in all cases.

Mainly opening now for a perf-run (which should be good to go, despite the still existing problems).

r? `@oli-obk`

cc `@lcnr` `@RalfJung`

2 years agoAuto merge of #8999 - Alexendoo:error-pattern, r=xFrednet
bors [Tue, 14 Jun 2022 16:54:13 +0000 (16:54 +0000)]
Auto merge of #8999 - Alexendoo:error-pattern, r=xFrednet

Remove error-pattern comments

The `clippy_lints` one [is unused](https://rust-lang.zulipchat.com/#narrow/stream/257328-clippy/topic/.60error-pattern.60), the others in `ui-toml` also appear not to have an effect

changelog: none

2 years agoRemove error-pattern comments
Alex Macleod [Tue, 14 Jun 2022 16:28:34 +0000 (16:28 +0000)]
Remove error-pattern comments

2 years agorebase
b-naber [Tue, 14 Jun 2022 15:34:37 +0000 (17:34 +0200)]
rebase

2 years agoadd a test case for `decl_macro`
Takayuki Maeda [Tue, 14 Jun 2022 15:42:10 +0000 (00:42 +0900)]
add a test case for `decl_macro`

2 years agoimplement `MacroData`
Takayuki Maeda [Tue, 14 Jun 2022 15:31:21 +0000 (00:31 +0900)]
implement `MacroData`

2 years agoadd lint
tamaron [Mon, 6 Jun 2022 13:48:05 +0000 (22:48 +0900)]
add lint

2 years agoaccount for endianness in debuginfo for const args
b-naber [Tue, 14 Jun 2022 08:50:20 +0000 (10:50 +0200)]
account for endianness in debuginfo for const args

2 years agorebase
b-naber [Tue, 7 Jun 2022 17:29:13 +0000 (19:29 +0200)]
rebase

2 years agocorrectly create Scalar for meta info
b-naber [Sat, 4 Jun 2022 09:24:29 +0000 (11:24 +0200)]
correctly create Scalar for meta info

2 years agofix wrong evaluation in clippy
b-naber [Fri, 3 Jun 2022 19:41:01 +0000 (21:41 +0200)]
fix wrong evaluation in clippy

2 years agofix wrong evaluation in clippy
b-naber [Fri, 3 Jun 2022 19:41:01 +0000 (21:41 +0200)]
fix wrong evaluation in clippy

2 years agoaddress review
b-naber [Fri, 3 Jun 2022 18:42:35 +0000 (20:42 +0200)]
address review

2 years agoaddress review
b-naber [Fri, 3 Jun 2022 18:42:35 +0000 (20:42 +0200)]
address review

2 years agobless 32-bit ui tests
b-naber [Fri, 3 Jun 2022 11:41:53 +0000 (13:41 +0200)]
bless 32-bit ui tests

2 years agomanually bless 32-bit mir-opt tests
b-naber [Fri, 3 Jun 2022 10:47:16 +0000 (12:47 +0200)]
manually bless 32-bit mir-opt tests

2 years agoaddress review
b-naber [Thu, 2 Jun 2022 17:42:29 +0000 (19:42 +0200)]
address review

2 years agofix clippy test failures
b-naber [Thu, 2 Jun 2022 13:15:01 +0000 (15:15 +0200)]
fix clippy test failures

2 years agofix clippy test failures
b-naber [Thu, 2 Jun 2022 13:15:01 +0000 (15:15 +0200)]
fix clippy test failures

2 years agoimplement valtrees as the type-system representation for constant values
b-naber [Wed, 16 Feb 2022 09:56:01 +0000 (10:56 +0100)]
implement valtrees as the type-system representation for constant values

2 years agoimplement valtrees as the type-system representation for constant values
b-naber [Wed, 16 Feb 2022 09:56:01 +0000 (10:56 +0100)]
implement valtrees as the type-system representation for constant values

2 years agoAuto merge of #78781 - eddyb:measureme-rdpmc, r=oli-obk
bors [Tue, 14 Jun 2022 13:37:39 +0000 (13:37 +0000)]
Auto merge of #78781 - eddyb:measureme-rdpmc, r=oli-obk

Integrate measureme's hardware performance counter support.

*Note: this is a companion to https://github.com/rust-lang/measureme/pull/143, and duplicates some information with it for convenience*

**(much later) EDIT**: take any numbers with a grain of salt, they may have changed since initial PR open.

## Credits

I'd like to start by thanking `@alyssais,` `@cuviper,` `@edef1c,` `@glandium,` `@jix,` `@Mark-Simulacrum,` `@m-ou-se,` `@mystor,` `@nagisa,` `@puckipedia,` and `@yorickvP,` for all of their help with testing, and valuable insight and suggestions.
Getting here wouldn't have been possible without you!

(If I've forgotten anyone please let me know, I'm going off memory here, plus some discussion logs)

## Summary

This PR adds support to `-Z self-profile` for counting hardware events such as "instructions retired" (as opposed to being limited to time measurements), using the `rdpmc` instruction on `x86_64` Linux.

While other OSes may eventually be supported, preliminary research suggests some kind of kernel extension/driver is required to enable this, whereas on Linux any user can profile (at least) their own threads.

Supporting Linux on architectures other than x86_64 should be much easier (provided the hardware supports such performance counters), and was mostly not done due to a lack of readily available test hardware.
That said, 32-bit `x86` (aka `i686`) would be almost trivial to add and test once we land the initial `x86_64` version (as all the CPU detection code can be reused).

A new flag `-Z self-profile-counter` was added, to control which of the named `measureme` counters is used, and which defaults to `wall-time`, in order to keep `-Z self-profile`'s current functionality unchanged (at least for now).

The named counters so far are:
* `wall-time`: the existing time measurement
    * name chosen for consistency with `perf.rust-lang.org`
    * continues to use `std::time::Instant` for a nanosecond-precision "monotonic clock"
* `instructions:u`: the hardware performance counter usually referred to as "Instructions retired"
    * here "retired" (roughly) means "fully executed"
    * the `:u` suffix is from the Linux `perf` tool and indicates the counter only runs while userspace code is executing, and therefore counts no kernel instructions
        * *see [Caveats/Subtracting IRQs](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Subtracting-IRQs) for why this isn't entirely true and why `instructions-minus-irqs:u` should be preferred instead*
* `instructions-minus-irqs:u`: same as `instructions:u`, except the count of hardware interrupts ("IRQs" here for brevity) is subtracted
    * *see [Caveats/Subtracting IRQs](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Subtracting-IRQs) for why this should be preferred over `instructions:u`*
* `instructions-minus-r0420:u`: experimental counter, same as `instructions-minus-irqs:u` but subtracting an undocumented counter (`r0420:u`) instead of IRQs
    * the `rXXXX` notation is again from Linux `perf`, and indicates a "raw" counter, with a hex representation of the low-level counter configuration - this was picked because we still don't *really* know what it is
    * this only exists for (future) testing and isn't included/used in any comparisons/data we've put together so far
    * *see [Challenges/Zen's undocumented 420 counter](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Epilogue-Zen’s-undocumented-420-counter) for details on how this counter was found and what it does*

---

There are also some additional commits:
* ~~see [Challenges/Rebasing *shouldn't* affect the results, right?](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Rebasing-*shouldn’t*-affect-the-results,-right) for details on the changes to `rustc_parse` and `rustc_trait_section` (the latter far more dubious, and probably shouldn't be merged, or not as-is)~~
  *  **EDIT**: the effects of these are no long quantifiable, the PR includes reverts for them
* ~~see [Challenges/`jemalloc`: purging will commence in ten seconds](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#jemalloc-purging-will-commence-in-ten-seconds) for details on the `jemalloc` change~~
  * this is also separately found in #77162, and we probably want to avoid doing it by default, ideally we'd use the runtime control API `jemalloc` offers (assuming that can stop the timer that's already running, which I'm not sure about)
  * **EDIT**: until we can do this based on `-Z` flags, this commit has also been reverted
* the `proc_macro` change was to avoid randomized hashing and therefore ASLR-like effects

---

**(much later) EDIT**: take any numbers with a grain of salt, they may have changed since initial PR open.

#### Write-up / report

Because of how extensive the full report ended up being, I've kept most of it [on `hackmd.io`](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view), but for convenient access, here are all the sections (with individual links):
<sup>(someone suggested I'd make a backup, so [here it is on the wayback machine](http://web.archive.org/web/20201127164748/https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view) - I'll need to remember to update that if I have to edit the write-up)</sup>

* [**Motivation**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Motivation)

* [**Results**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Results)
    * [**Overhead**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Overhead)
    *Preview (see the report itself for more details):*

    |Counter|Total<br>`instructions-minus-irqs:u`|Overhead from "Baseline"<br>(for all 1903881<br>counter reads)|Overhead from "Baseline"<br>(per each counter read)|
    |-|-|-|-|
    |Baseline|63637621286 ±6||
    |`instructions:u`|63658815885 ±2|&nbsp;&nbsp;+21194599 ±8|&nbsp;&nbsp;+11|
    |`instructions-minus-irqs:u`|63680307361 ±13|&nbsp;&nbsp;+42686075 ±19|&nbsp;&nbsp;+22|
    |`wall-time`|63951958376 ±10275|+314337090 ±10281|+165|

    * [**"Macro" noise (self time)**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#“Macro”-noise-(self-time))
    *Preview (see the report itself for more details):*

    || `wall-time` (ns) | `instructions:u` | `instructions-minus-irqs:u`
    -: | -: | -: | -:
    `typeck` | 5478261360 ±283933373 (±~5.2%) | 17350144522 ±6392 (±~0.00004%) | 17351035832.5 ±4.5 (±~0.00000003%)
    `expand_crate` | 2342096719 ±110465856 (±~4.7%) | 8263777916 ±2937 (±~0.00004%) | 8263708389 ±0 (±~0%)
    `mir_borrowck` | 2216149671 ±119458444 (±~5.4%) | 8340920100 ±2794 (±~0.00003%) | 8341613983.5 ±2.5 (±~0.00000003%)
    `mir_built` | 1269059734 ±91514604 (±~7.2%) | 4454959122 ±1618 (±~0.00004%) | 4455303811 ±1 (±~0.00000002%)
    `resolve_crate` | 942154987.5 ±53068423.5 (±~5.6%) | 3951197709 ±39 (±~0.000001%) | 3951196865 ±0 (±~0%)

    * [**"Micro" noise (individual sampling intervals)**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#“Micro”-noise-(individual-sampling-intervals))

* [**Caveats**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Caveats)
    * [**Disabling ASLR**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Disabling-ASLR)
    * [**Non-deterministic proc macros**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Non-deterministic-proc-macros)
    * [**Subtracting IRQs**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Subtracting-IRQs)
    * [**Lack of support for multiple threads**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Lack-of-support-for-multiple-threads)

* [**Challenges**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Challenges)
    * [**How do we even read hardware performance counters?**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#How-do-we-even-read-hardware-performance-counters)
    * [**ASLR: it's free entropy**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#ASLR-it’s-free-entropy)
    * [**The serializing instruction**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#The-serializing-instruction)
    * [**Getting constantly interrupted**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Getting-constantly-interrupted)
    * [**AMD patented time-travel and dubbed it `SpecLockMap`<br><sup>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;or: "how we accidentally unlocked `rr` on AMD Zen"</sup>**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#AMD-patented-time-travel-and-dubbed-it-SpecLockMapnbspnbspnbspnbspnbspnbspnbspnbspor-“how-we-accidentally-unlocked-rr-on-AMD-Zen”)
    * [**`jemalloc`: purging will commence in ten seconds**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#jemalloc-purging-will-commence-in-ten-seconds)
    * [**Rebasing *shouldn't* affect the results, right?**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Rebasing-*shouldn’t*-affect-the-results,-right)
    * [**Epilogue: Zen's undocumented 420 counter**](https://hackmd.io/sH315lO2RuicY-SEt7ynGA?view#Epilogue-Zen’s-undocumented-420-counter)

2 years agoAdd VFE test for 32 bit
flip1995 [Thu, 12 May 2022 10:04:42 +0000 (11:04 +0100)]
Add VFE test for 32 bit

The offset in the llvm.type.checked.load intrinsic differs on 32 bit platforms

2 years agoAdd user documentation for -Zvirtual-function-elimination
flip1995 [Mon, 2 May 2022 09:55:39 +0000 (10:55 +0100)]
Add user documentation for -Zvirtual-function-elimination

2 years agoAdd test for VFE optimization
flip1995 [Thu, 21 Apr 2022 13:16:02 +0000 (14:16 +0100)]
Add test for VFE optimization

2 years agoAdd metadata generation for vtables when using VFE
flip1995 [Thu, 21 Apr 2022 13:02:54 +0000 (14:02 +0100)]
Add metadata generation for vtables when using VFE

This adds the typeid and `vcall_visibility` metadata to vtables when the
-Cvirtual-function-elimination flag is set.

The typeid is generated in the same way as for the
`llvm.type.checked.load` intrinsic from the trait_ref.

The offset that is added to the typeid is always 0. This is because LLVM
assumes that vtables are constructed according to the definition in the
Itanium ABI. This includes an "address point" of the vtable. In C++ this
is the offset in the vtable where information for RTTI is placed. Since
there is no RTTI information in Rust's vtables, this "address point" is
always 0. This "address point" in combination with the offset passed to
the `llvm.type.checked.load` intrinsic determines the final function
that should be loaded from the vtable in the
`WholeProgramDevirtualization` pass in LLVM. That's why the
`llvm.type.checked.load` intrinsics are generated with the typeid of the
trait, rather than with that of the function that is called. This
matches what `clang` does for C++.

The vcall_visibility metadata depends on three factors:

1. LTO level: Currently this is always fat LTO, because LLVM only
   supports this optimization with fat LTO.
2. Visibility of the trait: If the trait is publicly visible, VFE
   can only act on its vtables after linking.
3. Number of CGUs: if there is more than one CGU, also vtables with
   restricted visibility could be seen outside of the CGU, so VFE can
   only act on them after linking.

To reflect this, there are three visibility levels: Public, LinkageUnit,
and TranslationUnit.

2 years agoAdd llvm.type.checked.load intrinsic
flip1995 [Thu, 21 Apr 2022 12:58:25 +0000 (13:58 +0100)]
Add llvm.type.checked.load intrinsic

Add the intrinsic

declare {i8*, i1} @llvm.type.checked.load(i8* %ptr, i32 %offset, metadata %type)

This is used in the VFE optimization when lowering loading functions
from vtables to LLVM IR. The `metadata` is used to map the function to
all vtables this function could belong to. This ensures that functions
from vtables that might be used somewhere won't get removed.

2 years agoAdd typeid_for_trait_ref function
flip1995 [Thu, 21 Apr 2022 12:42:46 +0000 (13:42 +0100)]
Add typeid_for_trait_ref function

This function computes a Itanium-like typeid for a trait_ref. This is
required for the VFE optimization in LLVM. It is used to map
`llvm.type.checked.load` invocations, that is loading the function from
a vtable, to the vtables this function could be from.

It is important to note that `typeid`s are not unique. So multiple
vtables of the same trait can share `typeid`s.

2 years agoAdd LLVM module flags required for the VFE opt
flip1995 [Thu, 21 Apr 2022 12:35:40 +0000 (13:35 +0100)]
Add LLVM module flags required for the VFE opt

To apply the optimization the `Virtual Function Elim` module flag has to
be set. To apply this optimization post-link the `LTOPostLink` module
flag has to be set.

2 years agoAdd -Zvirtual-function-elimination flag
flip1995 [Thu, 21 Apr 2022 12:29:45 +0000 (13:29 +0100)]
Add -Zvirtual-function-elimination flag

Adds the virtual-function-elimination unstable compiler flag and a check
that this flag is only used in combination with -Clto. LLVM can only
apply this optimization with fat LTO.

2 years agoAuto merge of #98091 - Dylan-DPC:rollup-ueb6b5x, r=Dylan-DPC
bors [Tue, 14 Jun 2022 10:51:16 +0000 (10:51 +0000)]
Auto merge of #98091 - Dylan-DPC:rollup-ueb6b5x, r=Dylan-DPC

Rollup of 5 pull requests

Successful merges:

 - #97869 (BTree: tweak internal comments)
 - #97935 (Rename the `ConstS::val` field as `kind`.)
 - #97948 (lint: add diagnostic translation migration lints)
 - #98042 (Fix compat_fn option method on miri)
 - #98069 (rustdoc:  remove link on slice brackets)

Failed merges:

r? `@ghost`
`@rustbot` modify labels: rollup

2 years agoAuto merge of #8947 - Serial-ATA:lint-produces-output, r=xFrednet
bors [Tue, 14 Jun 2022 10:42:09 +0000 (10:42 +0000)]
Auto merge of #8947 - Serial-ATA:lint-produces-output, r=xFrednet

Add lint output to lint list

changelog: Add the ability to show the lint output in the lint list

This just adds the logic to produce the output, it hasn't been added to any lints yet. It did help find some mistakes in some docs though :smile:.

### Screenshots

<details>
<summary>A single code block</summary>

![single-code-block](https://user-images.githubusercontent.com/69764315/172013766-145b22b1-1d91-4fb8-9cd0-b967a52d6330.png)
</details>

<details>
<summary>A single code block with a "Use instead" section</summary>

![with-usage](https://user-images.githubusercontent.com/69764315/172013792-d2dd6c9c-defa-41e0-8c27-8e8e311adb63.png)
</details>

<details>
<summary>Multiple code blocks</summary>

![multi-code-block](https://user-images.githubusercontent.com/69764315/172013808-5328f59b-e7c5-4914-a396-253822a6d350.png)
</details>

This is the last task in #7172 :tada:.
r? `@xFrednet` (?)

2 years agoAuto merge of #8901 - Jarcho:sharing_code, r=dswij
bors [Tue, 14 Jun 2022 08:59:40 +0000 (08:59 +0000)]
Auto merge of #8901 - Jarcho:sharing_code, r=dswij

Rework `branches_sharing_code`

fixes #7378

This changes the lint from checking pairs of blocks, to checking all the blocks at the same time. As such there's almost none of the original code left.

changelog: Don't lint `branches_sharing_code` when using different binding names

2 years agoFix expand/collapse on source viewer sidebar folders
Guillaume Gomez [Tue, 14 Jun 2022 08:59:01 +0000 (10:59 +0200)]
Fix expand/collapse on source viewer sidebar folders

2 years agoRollup merge of #98069 - notriddle:notriddle/square-brackets, r=jsha
Dylan DPC [Tue, 14 Jun 2022 08:35:33 +0000 (10:35 +0200)]
Rollup merge of #98069 - notriddle:notriddle/square-brackets, r=jsha

rustdoc:  remove link on slice brackets

This is #91778, take two.

Fixes #91173

The reason I'm reevaluating this change is #97668, which makes fully-generic slices link to the slice docs page. This fixes some downsides in the original PR, where `Box<[T]>`, for example, was not linked to the primitive.slice.html page. In this PR, the `[T]` inside is still a link.

The other major reason for wanting to reevaluate this is the changed color scheme. When this feature was first introduced in rustdoc, primitives were a different color from structs and enums. This way, eagle-eyed users could figure out that the square brackets were separate links from the structs inside. Now, all types have the same color, so a significant fraction of users won't even know the links are there unless they pay close attention to the status bar or use an accessibility tool that lists all links on the page.

2 years agoRollup merge of #98042 - DrMeepster:winfred_std_changes, r=ChrisDenton
Dylan DPC [Tue, 14 Jun 2022 08:35:32 +0000 (10:35 +0200)]
Rollup merge of #98042 - DrMeepster:winfred_std_changes, r=ChrisDenton

Fix compat_fn option method on miri

This change is required to make `WaitOnAddress` work with rust-lang/miri#2231

2 years agoRollup merge of #97948 - davidtwco:diagnostic-translation-lints, r=oli-obk
Dylan DPC [Tue, 14 Jun 2022 08:35:31 +0000 (10:35 +0200)]
Rollup merge of #97948 - davidtwco:diagnostic-translation-lints, r=oli-obk

lint: add diagnostic translation migration lints

Introduce allow-by-default lints for checking whether diagnostics are written in
`SessionDiagnostic` or `AddSubdiagnostic` impls and whether diagnostics are translatable. These lints can be denied for modules once they are fully migrated to impls and translation.

These lints are intended to be temporary - once all diagnostics have been changed then we can just change the APIs we have and that will enforce these constraints thereafter.

r? `````@oli-obk`````

2 years agoRollup merge of #97935 - nnethercote:rename-ConstS-val-as-kind, r=lcnr
Dylan DPC [Tue, 14 Jun 2022 08:35:29 +0000 (10:35 +0200)]
Rollup merge of #97935 - nnethercote:rename-ConstS-val-as-kind, r=lcnr

Rename the `ConstS::val` field as `kind`.

And likewise for the `Const::val` method.

Because its type is called `ConstKind`. Also `val` is a confusing name
because `ConstKind` is an enum with seven variants, one of which is
called `Value`. Also, this gives consistency with `TyS` and `PredicateS`
which have `kind` fields.

The commit also renames a few `Const` variables from `val` to `c`, to
avoid confusion with the `ConstKind::Value` variant.

r? `@BoxyUwU`

2 years agoRollup merge of #97869 - ssomers:btree_comments, r=Dylan-DPC
Dylan DPC [Tue, 14 Jun 2022 08:35:29 +0000 (10:35 +0200)]
Rollup merge of #97869 - ssomers:btree_comments, r=Dylan-DPC

BTree: tweak internal comments

2 years agoAuto merge of #8997 - Jarcho:clap_deprecate, r=flip1995
bors [Tue, 14 Jun 2022 08:33:52 +0000 (08:33 +0000)]
Auto merge of #8997 - Jarcho:clap_deprecate, r=flip1995

Fix `clap` deprecation warnings

Clap `3.2.0` deprecated a few functions used by lintcheck.

changelog: None

2 years agoAuto merge of #98082 - lnicola:rust-analyzer-2022-06-14, r=lnicola
bors [Tue, 14 Jun 2022 08:08:36 +0000 (08:08 +0000)]
Auto merge of #98082 - lnicola:rust-analyzer-2022-06-14, r=lnicola

:arrow_up: rust-analyzer

r? `@ghost`

2 years agochange edition to 2021
Takayuki Maeda [Tue, 14 Jun 2022 06:46:11 +0000 (15:46 +0900)]
change edition to 2021

2 years agosuggest adding a `#[macro_export]` to a private macro
Takayuki Maeda [Tue, 14 Jun 2022 05:58:46 +0000 (14:58 +0900)]
suggest adding a `#[macro_export]` to a private macro

2 years agoRemove thread-local `IGNORED_ATTRIBUTES`.
Nicholas Nethercote [Tue, 14 Jun 2022 05:20:54 +0000 (15:20 +1000)]
Remove thread-local `IGNORED_ATTRIBUTES`.

It's just a copy of the read-only global `ich::IGNORED_ATTRIBUTES`, and
can be removed without any effect.

2 years agoAuto merge of #98041 - jackh726:remove-regionckmode, r=oli-obk
bors [Tue, 14 Jun 2022 05:07:11 +0000 (05:07 +0000)]
Auto merge of #98041 - jackh726:remove-regionckmode, r=oli-obk

Remove RegionckMode in favor of calling new skip_region_resolution

Simple cleanup. We can skip a bunch of stuff for places where NLL does the region checking, so skip earlier.

r? rust-lang/types

2 years agoRename rustc_serialize::opaque::Encoder as MemEncoder.
Nicholas Nethercote [Tue, 14 Jun 2022 04:52:01 +0000 (14:52 +1000)]
Rename rustc_serialize::opaque::Encoder as MemEncoder.

This avoids the name clash with `rustc_serialize::Encoder` (a trait),
and allows lots qualifiers to be removed and imports to be simplified
(e.g. fewer `as` imports).

(This was previously merged as commit 5 in #94732 and then was reverted
in #97905 because of a perf regression caused by commit 4 in #94732.)

2 years ago:arrow_up: rust-analyzer
Laurențiu Nicola [Tue, 14 Jun 2022 04:43:32 +0000 (07:43 +0300)]
:arrow_up: rust-analyzer

2 years agoDo not try to statically link libstd++ on Solaris
gco [Tue, 14 Jun 2022 04:29:09 +0000 (21:29 -0700)]
Do not try to statically link libstd++ on Solaris

Fixes: Error on bootstrapping : Empty search path given via '-L' (solaris) #97260
2 years agoImprove parsing errors and suggestions for bad if statements
Michael Goulet [Sat, 28 May 2022 04:58:48 +0000 (21:58 -0700)]
Improve parsing errors and suggestions for bad if statements

2 years agoUse a private type definition to reduce cfg noise
Mark Drobnak [Wed, 8 Jun 2022 02:57:15 +0000 (19:57 -0700)]
Use a private type definition to reduce cfg noise

I checked with t-libs to make sure this is OK to do on stable functions:
https://rust-lang.zulipchat.com/#narrow/stream/219381-t-libs/topic/Replacing.20std.20function.20arg.20type.20with.20private.20type.20def.3F

2 years agoDisable `has_thread_local` due to weird issues in some programs
Mark Drobnak [Sun, 22 May 2022 00:11:48 +0000 (17:11 -0700)]
Disable `has_thread_local` due to weird issues in some programs

For example, in the following issue the `thread_info` thread-local is
not correctly initialized in debug builds:
https://github.com/Meziu/ctru-rs/issues/60

2 years agoEnable thread_local_dtor on horizon OS
Ian Chamberlain [Thu, 24 Mar 2022 01:00:38 +0000 (21:00 -0400)]
Enable thread_local_dtor on horizon OS

Always use fallback thread_local destructor, since __cxa_thread_atexit_impl
is never defined on the target.

See https://github.com/AzureMarker/rust-horizon/pull/2

2 years agoAdd platform-support page for armv6k-nintendo-3ds
Ian Chamberlain [Thu, 17 Mar 2022 03:26:22 +0000 (23:26 -0400)]
Add platform-support page for armv6k-nintendo-3ds

Co-authored-by: Mark Drobnak <mark.drobnak@gmail.com>