]> git.lizzy.rs Git - rust.git/log
rust.git
2 years agoAuto merge of #87598 - ccqpein:master, r=yaahc
bors [Mon, 23 Aug 2021 05:06:29 +0000 (05:06 +0000)]
Auto merge of #87598 - ccqpein:master, r=yaahc

Add doctests for HashMap's into_values and into_keys methods

Fixes #87591

2 years agoAuto merge of #88210 - spastorino:diff-lifetimes-def-use-test, r=oli-obk
bors [Mon, 23 Aug 2021 02:25:36 +0000 (02:25 +0000)]
Auto merge of #88210 - spastorino:diff-lifetimes-def-use-test, r=oli-obk

Test TAITs different lifetimes in defining uses fail

r? `@oli-obk`

Related to #86727

2 years agoAuto merge of #88200 - pcwalton:no-dso-local-on-mach-o, r=nagisa
bors [Sun, 22 Aug 2021 23:44:48 +0000 (23:44 +0000)]
Auto merge of #88200 - pcwalton:no-dso-local-on-mach-o, r=nagisa

Stop emitting the `dso_local` LLVM attribute for external symbols under the static relocation model on macOS.

This matches Clang's behavior:

https://github.com/llvm/llvm-project/blob/973cb2c326be9f256da0897c4d2ef117dc22761d/clang/lib/CodeGen/CodeGenModule.cpp#L1038-L1040

Even if `dso_local` were properly supported in this way on macOS, it seems
incorrect to add this annotation as liberally as we did. The `dso_local`
annotation is for symbols that ultimately end up in the same linkage unit, but
we were adding this annotation even for `static` values inside `extern` blocks
marked with `#[link(type="framework")]`, which should be considered dynamically
linked.  Note that Clang likewise avoids emitting `dso_local` for `dllimport`
symbols:

https://github.com/llvm/llvm-project/blob/973cb2c326be9f256da0897c4d2ef117dc22761d/clang/lib/CodeGen/CodeGenModule.cpp#L1005-L1007

This issue caused breakage in the `ring` crate, which links to a symbol defined
in `Security.framework` that ultimately resolves to address `0x0`:

https://github.com/briansmith/ring/blob/b94d61e044b42827fefd71d5f61e8c58a7659870/src/rand.rs#L390

For this symbol, the use of `dso_local` causes LLVM to emit a relocation of
type `X86_64_RELOC_SIGNED`, which is a 32-bit signed PC-relative offset. If the
binary is large enough, `0x0` might be out of range, and the link will fail.
Avoiding `dso_local` causes LLVM to use the GOT instead, emitting a relocation
of type `X86_64_RELOC_GOT_LOAD`, which will properly handle the large offset
and cause the link to succeed.

As a side note, the static relocation model is effectively deprecated for
security reasons on macOS, as it prohibits PIE. It's also completely
unsupported on Apple Silicon, so I don't think it's worth going to the effort
of properly supporting this model on that platform.

2 years agoAuto merge of #88240 - GuillaumeGomez:rollup-wdom91m, r=GuillaumeGomez
bors [Sun, 22 Aug 2021 20:23:37 +0000 (20:23 +0000)]
Auto merge of #88240 - GuillaumeGomez:rollup-wdom91m, r=GuillaumeGomez

Rollup of 7 pull requests

Successful merges:

 - #86747 (Improve wording of the `drop_bounds` lint)
 - #87166 (Show discriminant before overflow in diagnostic for duplicate values.)
 - #88077 (Generate an iOS LLVM target with a specific version)
 - #88164 (PassWrapper: adapt for LLVM 14 changes)
 - #88211 (cleanup: `Span::new` -> `Span::with_lo`)
 - #88229 (Suggest importing the right kind of macro.)
 - #88238 (Stop tracking namespace in used_imports.)

Failed merges:

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

2 years agoRollup merge of #88238 - m-ou-se:used-imports-no-track-namespace, r=estebank
Guillaume Gomez [Sun, 22 Aug 2021 18:52:56 +0000 (20:52 +0200)]
Rollup merge of #88238 - m-ou-se:used-imports-no-track-namespace, r=estebank

Stop tracking namespace in used_imports.

This changes `used_imports` from a `FxHashSet<(NodeId, Namespace)>` to a `FxHashSet<NodeId>`, as the Namespace information isn't used.

The only point that uses it did three lookups, `|=`'ing them together.

r? `@estebank`

2 years agoRollup merge of #88229 - m-ou-se:macro-suggest-right-kind, r=estebank
Guillaume Gomez [Sun, 22 Aug 2021 18:52:55 +0000 (20:52 +0200)]
Rollup merge of #88229 - m-ou-se:macro-suggest-right-kind, r=estebank

Suggest importing the right kind of macro.

Fixes #88228.

r? `@estebank`

2 years agoRollup merge of #88211 - petrochenkov:withhilo, r=jyn514
Guillaume Gomez [Sun, 22 Aug 2021 18:52:54 +0000 (20:52 +0200)]
Rollup merge of #88211 - petrochenkov:withhilo, r=jyn514

cleanup: `Span::new` -> `Span::with_lo`

Extracted from https://github.com/rust-lang/rust/pull/84373 as suggested in https://github.com/rust-lang/rust/pull/84373#issuecomment-857773867.
It turned out less useful then I expected, but anyway.
r? `@cjgillot`
`@bors` rollup

2 years agoRollup merge of #88164 - durin42:llvm-14-san-opts, r=nikic
Guillaume Gomez [Sun, 22 Aug 2021 18:52:53 +0000 (20:52 +0200)]
Rollup merge of #88164 - durin42:llvm-14-san-opts, r=nikic

PassWrapper: adapt for LLVM 14 changes

These API changes appear to have all taken place in
https://reviews.llvm.org/D105007, which moved HWAddressSanitizerPass and
AddressSanitizerPass to only accept their options type as a ctor
argument instead of the sequence of bools etc. This required a couple of
parameter additions, which I made match the default prior to the
mentioned upstream LLVM change.

This patch restores rustc to building (though not quite passing all
tests, I've mailed other patches for those issues) against LLVM HEAD.

2 years agoRollup merge of #88077 - kit-981:feature/fix-minimum-os-version-in-header, r=petrochenkov
Guillaume Gomez [Sun, 22 Aug 2021 18:52:52 +0000 (20:52 +0200)]
Rollup merge of #88077 - kit-981:feature/fix-minimum-os-version-in-header, r=petrochenkov

Generate an iOS LLVM target with a specific version

This commit adds the `LC_VERSION_MIN_IPHONEOS` load command to the Mach-O header generated for `aarch64-apple-ios` binaries. The operating system will look for this load command to determine the minimum supported operating system version and will not allow the binary to run if it's absent. This logic already exists for the simulator toolchain.

I've been using `otool` from a [cctools](https://github.com/tpoechtrager/cctools-port) toolchain to parse the header and validate that this change adds the required load command.

This change appears to be enough to build Rust binaries that can run on a jailbroken iPhone.

2 years agoRollup merge of #87166 - de-vri-es:show-discriminant-before-overflow, r=jackh726
Guillaume Gomez [Sun, 22 Aug 2021 18:52:51 +0000 (20:52 +0200)]
Rollup merge of #87166 - de-vri-es:show-discriminant-before-overflow, r=jackh726

Show discriminant before overflow in diagnostic for duplicate values.

This PR adds the value before overflow for explicit discriminant values in the error for duplicate discriminant values.
I found it rather confusing to see only the overflowed value.

It only does this for literals, since overflows in const evaluated arithmetic are already a hard error.

This is my first PR to the compiler, so please let me know if the implementation can be improved :)

Before:
![image](https://user-images.githubusercontent.com/786213/125850097-bf5fb7e0-d800-4386-a738-c30f41822964.png)

After:
![image](https://user-images.githubusercontent.com/786213/125850120-e2bb765d-ad86-4888-a6cb-dec34fba3fea.png)

2 years agoRollup merge of #86747 - FabianWolff:issue-86653, r=GuillaumeGomez
Guillaume Gomez [Sun, 22 Aug 2021 18:52:50 +0000 (20:52 +0200)]
Rollup merge of #86747 - FabianWolff:issue-86653, r=GuillaumeGomez

Improve wording of the `drop_bounds` lint

This PR addresses #86653. The issue is sort of a false positive of the `drop_bounds` lint, but I would argue that the best solution for #86653 is simply a rewording of the warning message and lint description, because even if the lint is _technically_ wrong, it still forces the programmer to think about what they are doing, and they can always use `#[allow(drop_bounds)]` if they think that they really need the `Drop` bound.

There are two issues with the current warning message and lint description:
- First, it says that `Drop` bounds are "useless", which is technically incorrect because they actually do have the effect of allowing you e.g. to call methods that also have a `Drop` bound on their generic arguments for some reason. I have changed the wording to emphasize not that the bound is "useless", but that it is most likely not what was intended.
- Second, it claims that `std::mem::needs_drop` detects whether a type has a destructor. But I think this is also technically wrong: The `Drop` bound says whether the type has a destructor or not, whereas `std::mem::needs_drop` also takes nested types with destructors into account, even if the top-level type does not itself have one (although I'm not 100% sure about the exact terminology here, i.e. whether the "drop glue" of the top-level type counts as a destructor or not).

cc `@jonhoo,` does this solve the issue for you?

r? `@GuillaumeGomez`

2 years agoAuto merge of #88166 - BoxyUwU:const-equate-canon, r=lcnr
bors [Sun, 22 Aug 2021 18:00:22 +0000 (18:00 +0000)]
Auto merge of #88166 - BoxyUwU:const-equate-canon, r=lcnr

canonicalize consts before calling try_unify_abstract_consts query

Fixes #88022
Fixes #86953
Fixes #77708
Fixes #82034
Fixes #85031

these ICEs were all caused by calling the `try_unify_abstract_consts` query with inference vars in substs

r? `@lcnr`

2 years agoStop tracking namespce in used_imports.
Mara Bos [Sun, 22 Aug 2021 14:50:59 +0000 (16:50 +0200)]
Stop tracking namespce in used_imports.

The information was tracked, but unused.

2 years agoTest TAITs different lifetimes in defining uses fail
Santiago Pastorino [Sat, 21 Aug 2021 15:26:25 +0000 (12:26 -0300)]
Test TAITs different lifetimes in defining uses fail

2 years agoAuto merge of #88163 - camsteffen:collapsible-match-fix, r=Manishearth
bors [Sun, 22 Aug 2021 13:26:32 +0000 (13:26 +0000)]
Auto merge of #88163 - camsteffen:collapsible-match-fix, r=Manishearth

Fix clippy::collapsible_match with let expressions

This fixes rust-lang/rust-clippy#7575 which is a regression from #80357. I am fixing the bug here instead of in the clippy repo (if that's okay) because a) the regression has not been synced yet and b) I would like to land the fix on nightly asap.

The fix is basically to re-generalize `match` and `if let` for the lint implementation (they were split because `if let` no longer desugars to `match` in the HIR).

Also fixes rust-lang/rust-clippy#7586 and fixes rust-lang/rust-clippy#7591
cc `@rust-lang/clippy`
`@xFrednet` do you want to review this?

2 years agoAdd doctests for 's into_values and into_keys methods
ccQpein [Thu, 29 Jul 2021 17:24:25 +0000 (13:24 -0400)]
Add doctests for 's into_values and into_keys methods

2 years agoSuggest importing the right kind of macro.
Mara Bos [Sun, 22 Aug 2021 12:06:45 +0000 (14:06 +0200)]
Suggest importing the right kind of macro.

2 years agoAuto merge of #88139 - lcnr:marker-trait-attr, r=nikomatsakis
bors [Sun, 22 Aug 2021 10:44:38 +0000 (10:44 +0000)]
Auto merge of #88139 - lcnr:marker-trait-attr, r=nikomatsakis

marker_traits: require `EvaluatedToOk` during winnowing

closes #84955, while it doesn't really fix it in a way that makes me happy it should prevent the issue for now and this
test can't be reproduced anyways, so it doesn't make much sense to keep it open.

fixes #84917 as only one of the impls depends on regions, so we now drop the ambiguous one instead of the correct one.

cc https://rust-lang.zulipchat.com/#narrow/stream/144729-wg-traits/topic/winnowing.20soundly/near/247899832

r? `@nikomatsakis`

2 years agoAuto merge of #88122 - Seppel3210:master, r=dtolnay
bors [Sun, 22 Aug 2021 08:03:47 +0000 (08:03 +0000)]
Auto merge of #88122 - Seppel3210:master, r=dtolnay

Fix example in `Extend<(A, B)>` impl

After looking over the examples in my last PR (#85835) on doc.rust-lang.org/nightly I realized that the example didn't actually show what I wanted it to show ðŸ˜…
So here's the better example

2 years agoAuto merge of #85166 - mbhall88:file-prefix, r=dtolnay
bors [Sun, 22 Aug 2021 05:19:48 +0000 (05:19 +0000)]
Auto merge of #85166 - mbhall88:file-prefix, r=dtolnay

add file_prefix method to std::path

This is an initial implementation of `std::path::Path::file_prefix`. It is effectively a "left" variant of the existing [`file_stem`](https://doc.rust-lang.org/std/path/struct.Path.html#method.file_stem) method. An illustration of the difference is

```rust
use std::path::Path;

let path = Path::new("foo.tar.gz");
assert_eq!(path.file_stem(), Some("foo.tar"));
assert_eq!(path.file_prefix(), Some("foo"));
```

In my own development, I generally find I almost always want the prefix, rather than the stem, so I thought it might be best to suggest it's addition to libstd.

Of course, as this is my first contribution, I expect there is probably more work that needs to be done. Additionally, if the libstd team feel this isn't appropriate then so be it.

There has been some [discussion about this on Zulip](https://rust-lang.zulipchat.com/#narrow/stream/219381-t-libs/topic/file_lstem/near/238076313) and a user there suggested I open a PR to see whether someone in the libstd team thinks it is worth pursuing.

2 years agoAuto merge of #88217 - jackh726:rollup-3k74o2m, r=jackh726
bors [Sun, 22 Aug 2021 02:16:35 +0000 (02:16 +0000)]
Auto merge of #88217 - jackh726:rollup-3k74o2m, r=jackh726

Rollup of 13 pull requests

Successful merges:

 - #87604 (CI: Verify commits in beta & stable are in upstream branches.)
 - #88057 (Update RELEASES to clarify attribute macro values.)
 - #88072 (Allow the iOS toolchain to be built on Linux)
 - #88170 (Update release note for 1.55.0.)
 - #88172 (Test that type alias impl trait happens in a submodule)
 - #88179 (Mailmap entry for myself)
 - #88182 (We meant to use a trait instead of lifetime here)
 - #88183 (test TAIT in different positions)
 - #88189 (Add TAIT struct test)
 - #88192 (Use of impl trait in an impl as the value for an associated type in a dyn)
 - #88194 (Test use of impl Trait in an impl as the value for an associated type in an impl trait)
 - #88197 (Test tait use in a fn type)
 - #88201 (Test that incomplete inference for TAITs fail)

Failed merges:

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

2 years agoRollup merge of #88201 - spastorino:tait-incomplete-inference-test, r=oli-obk
Jack Huey [Sun, 22 Aug 2021 00:56:39 +0000 (20:56 -0400)]
Rollup merge of #88201 - spastorino:tait-incomplete-inference-test, r=oli-obk

Test that incomplete inference for TAITs fail

r? `@oli-obk`

Related to #86727

2 years agoRollup merge of #88197 - spastorino:tait-test-fn-type, r=oli-obk
Jack Huey [Sun, 22 Aug 2021 00:56:38 +0000 (20:56 -0400)]
Rollup merge of #88197 - spastorino:tait-test-fn-type, r=oli-obk

Test tait use in a fn type

r? `@oli-obk`

I thought this was going to work but doesn't, quickly checked with Niko and he told me that we ruled this out for now. I'm not exactly sure why and how but here we have a test with a FIXME :)

Related to #86727

2 years agoRollup merge of #88194 - spastorino:test-tait-assoc-impl-trait, r=oli-obk
Jack Huey [Sun, 22 Aug 2021 00:56:37 +0000 (20:56 -0400)]
Rollup merge of #88194 - spastorino:test-tait-assoc-impl-trait, r=oli-obk

Test use of impl Trait in an impl as the value for an associated type in an impl trait

r? `@oli-obk`

Related to #86727

2 years agoRollup merge of #88192 - spastorino:add-tait-test-for-assoc-dyn, r=oli-obk
Jack Huey [Sun, 22 Aug 2021 00:56:36 +0000 (20:56 -0400)]
Rollup merge of #88192 - spastorino:add-tait-test-for-assoc-dyn, r=oli-obk

Use of impl trait in an impl as the value for an associated type in a dyn

r? `@oli-obk`

Related to #86727

2 years agoRollup merge of #88189 - spastorino:add-tait-struct-test, r=oli-obk
Jack Huey [Sun, 22 Aug 2021 00:56:35 +0000 (20:56 -0400)]
Rollup merge of #88189 - spastorino:add-tait-struct-test, r=oli-obk

Add TAIT struct test

r? `@oli-obk`

Related to #86727

2 years agoRollup merge of #88183 - spastorino:add-tait-in-different-tuple-position, r=oli-obk
Jack Huey [Sun, 22 Aug 2021 00:56:34 +0000 (20:56 -0400)]
Rollup merge of #88183 - spastorino:add-tait-in-different-tuple-position, r=oli-obk

test TAIT in different positions

r? `@oli-obk`

Related to #86727

2 years agoRollup merge of #88182 - spastorino:use-trait-in-tait-tests, r=oli-obk
Jack Huey [Sun, 22 Aug 2021 00:56:33 +0000 (20:56 -0400)]
Rollup merge of #88182 - spastorino:use-trait-in-tait-tests, r=oli-obk

We meant to use a trait instead of lifetime here

r? `@oli-obk`

Related to #86727

2 years agoRollup merge of #88179 - steffahn:mailmap, r=Mark-Simulacrum
Jack Huey [Sun, 22 Aug 2021 00:56:32 +0000 (20:56 -0400)]
Rollup merge of #88179 - steffahn:mailmap, r=Mark-Simulacrum

Mailmap entry for myself

2 years agoRollup merge of #88172 - spastorino:tait-defining-use-submodule-test, r=oli-obk
Jack Huey [Sun, 22 Aug 2021 00:56:31 +0000 (20:56 -0400)]
Rollup merge of #88172 - spastorino:tait-defining-use-submodule-test, r=oli-obk

Test that type alias impl trait happens in a submodule

r? `@oli-obk`

Related to #86727

2 years agoRollup merge of #88170 - nebkor:release-note-1.55, r=Mark-Simulacrum
Jack Huey [Sun, 22 Aug 2021 00:56:29 +0000 (20:56 -0400)]
Rollup merge of #88170 - nebkor:release-note-1.55, r=Mark-Simulacrum

Update release note for 1.55.0.

Added a line about new formatting option, `{lib}`, for `cargo
tree` (https://github.com/rust-lang/cargo/pull/9663).

2 years agoRollup merge of #88072 - kit-981:feature/build-ios-toolchain-on-linux, r=Mark-Simulacrum
Jack Huey [Sun, 22 Aug 2021 00:56:29 +0000 (20:56 -0400)]
Rollup merge of #88072 - kit-981:feature/build-ios-toolchain-on-linux, r=Mark-Simulacrum

Allow the iOS toolchain to be built on Linux

The iOS toolchain can be built on Linux with minor changes. The compilation will invoke `xcrun` to find the path to the iPhone SDK but a fake `xcrun` executable can be used.

```
#!/bin/sh
echo "/path/to/sdk"
```

The iOS toolchain can then be built and linked with rustup.

```
$ ./x.py build --stage 2 --host x86_64-unknown-linux-gnu \
    --target aarch64-apple-ios
$ rustup toolchain link stage1 build/x86_64-unknown-linux-gnu/stage1
```

It's possible to take this toolchain and compile an iOS executable with it. This requires the ld64 linker and an iOS SDK. The ld64 linker can be taken from [cctools](https://github.com/tpoechtrager/cctools-port). A project's .cargo/config can then be edited to use the linker for this target.

```
[target.aarch64-apple-ios]
linker = "/path/to/cctools/bin/arm-apple-darwin-ld"
rustflags = [
    "-C",
    """
link-args=
    -F/path/to/sdk/System/Library/Frameworks
    -L/path/to/sdk/usr/lib
    -L/path/to/sdk/usr/lib/system/
    -adhoc_codesign
    """,
]
```

2 years agoRollup merge of #88057 - ehuss:releases-doc-macros, r=Mark-Simulacrum
Jack Huey [Sun, 22 Aug 2021 00:56:28 +0000 (20:56 -0400)]
Rollup merge of #88057 - ehuss:releases-doc-macros, r=Mark-Simulacrum

Update RELEASES to clarify attribute macro values.

As noted in #87681, macros do not work with the `#[path]` attribute.  Since the places where macros *can* be used is very limited, I have changed this to just focus on `#[doc]` which is the only attribute where this is really useful.

2 years agoRollup merge of #87604 - yaymukund:verify-backported-commits, r=Mark-Simulacrum
Jack Huey [Sun, 22 Aug 2021 00:56:27 +0000 (20:56 -0400)]
Rollup merge of #87604 - yaymukund:verify-backported-commits, r=Mark-Simulacrum

CI: Verify commits in beta & stable are in upstream branches.

Closes #74721

I think this does the trick. https://github.com/rust-lang/rust/pull/87597 is an example of it failing as it should.

2 years agoAuto merge of #88075 - Xuanwo:vec_deque_retain, r=dtolnay
bors [Sat, 21 Aug 2021 23:35:54 +0000 (23:35 +0000)]
Auto merge of #88075 - Xuanwo:vec_deque_retain, r=dtolnay

Optimize unnecessary check in VecDeque::retain

This pr is highly inspired by https://github.com/rust-lang/rust/pull/88060 which shared the same idea: we can split the `for` loop into stages so that we can remove unnecessary checks like `del > 0`.

## Benchmarks

Before

```rust
test collections::vec_deque::tests::bench_retain_half_10000  ... bench:     290,125 ns/iter (+/- 8,717)
test collections::vec_deque::tests::bench_retain_odd_10000   ... bench:     291,588 ns/iter (+/- 9,621)
test collections::vec_deque::tests::bench_retain_whole_10000 ... bench:     287,426 ns/iter (+/- 9,009)
```

After

```rust
test collections::vec_deque::tests::bench_retain_half_10000  ... bench:     243,940 ns/iter (+/- 8,563)
test collections::vec_deque::tests::bench_retain_odd_10000   ... bench:     242,768 ns/iter (+/- 3,903)
test collections::vec_deque::tests::bench_retain_whole_10000 ... bench:     202,926 ns/iter (+/- 6,332)
```

Based on the current benchmark, this PR will improve the perf of `VecDeque::retain` by around 16%. For special cases, the improvement will be up to 30%.

Signed-off-by: Xuanwo <github@xuanwo.io>
2 years agoAuto merge of #88135 - crlf0710:trait_upcasting_part_3, r=nikomatsakis
bors [Sat, 21 Aug 2021 21:14:07 +0000 (21:14 +0000)]
Auto merge of #88135 - crlf0710:trait_upcasting_part_3, r=nikomatsakis

Trait upcasting coercion (part 3)

By using separate candidates for each possible choice, this fixes type-checking issues in previous commits.

r? `@nikomatsakis`

2 years agoAuto merge of #82776 - jyn514:extern-url-fallback, r=GuillaumeGomez
bors [Sat, 21 Aug 2021 18:32:37 +0000 (18:32 +0000)]
Auto merge of #82776 - jyn514:extern-url-fallback, r=GuillaumeGomez

Give precedence to `html_root_url` over `--extern-html-root-url` by default, but add a way to opt-in to the previous behavior

## What is an HTML root url?

It tells rustdoc where it should link when documentation for a crate is
not available locally; for example, when a crate is a dependency of a
crate documented with `cargo doc --no-deps`.

 ## What is the difference between `html_root_url` and `--extern-html-root-url`?

Both of these tell rustdoc what the HTML root should be set to.
`doc(html_root_url)` is set by the crate author, while
`--extern-html-root-url` is set by the person documenting the crate.
These are often different. For example, docs.rs uses
`--extern-html-root-url https://docs.rs/crate-name/version` to ensure
all crates have documentation, even if `html_root_url` is not set.
Conversely, crates such as Rocket set `doc(html_root_url =
"https://api.rocket.rs")`, because they prefer users to view the
documentation on their own site.

Crates also set `html_root_url` to ensure they have
documentation when building locally when offline. This is unfortunate to
require, because it's more work from the library author. It also makes
it impossible to distinguish between crates that want to be viewed on a
different site (e.g. Rocket) and crates that just want documentation to
be visible offline at all (e.g. Tokio). I have authored a separate
change to the API guidelines to no longer recommend doing this:
rust-lang/api-guidelines#230.

 ## Why change the default?

In the past, docs.rs has been the main user of `--extern-html-root-url`.
However, it's useful for other projects as well. In particular, Cargo
wants to pass it by default when running `--no-deps`
(rust-lang/cargo#8296).

Unfortunately, for these other use cases, the priority order is
inverted. They want to give *precedence* to the URL the crate picks, and
only fall back to the `--extern-html-root` if no `html_root_url` is
present. That allows passing `--extern-html-root` unconditionally,
without having to parse the source code to see what attributes are
present.

For docs.rs, however, we still want to keep the old behavior, so that
all links on docs.rs stay on the site.

2 years agoAuto merge of #88134 - rylev:force-warn-improvements, r=nikomatsakis
bors [Sat, 21 Aug 2021 15:51:50 +0000 (15:51 +0000)]
Auto merge of #88134 - rylev:force-warn-improvements, r=nikomatsakis

Force warn improvements

As part of stablization of the `--force-warn` option (#86516) I've made the following changes:
* Error when the `warnings` lint group is based to the `--force-warn` option
* Tests have been updated to make it easier to understand the semantics of `--force-warn`

r? `@nikomatsakis`

2 years agocleanup: `Span::new` -> `Span::with_lo`
Vadim Petrochenkov [Sat, 21 Aug 2021 15:07:21 +0000 (18:07 +0300)]
cleanup: `Span::new` -> `Span::with_lo`

2 years agoAuto merge of #88128 - cuviper:needs-asm-support, r=Mark-Simulacrum
bors [Sat, 21 Aug 2021 12:57:00 +0000 (12:57 +0000)]
Auto merge of #88128 - cuviper:needs-asm-support, r=Mark-Simulacrum

Add needs-asm-support to more tests

These were found as test failures on s390x for RHEL and Fedora.

2 years agoAuto merge of #87570 - nikic:llvm-13, r=nagisa
bors [Sat, 21 Aug 2021 09:25:28 +0000 (09:25 +0000)]
Auto merge of #87570 - nikic:llvm-13, r=nagisa

Upgrade to LLVM 13

Work in progress update to LLVM 13. Main changes:

 * InlineAsm diagnostics reported using SrcMgr diagnostic kind are now handled. Previously these used a separate diag handler.
 * Codegen tests are updated for additional attributes.
 * Some data layouts have changed.
 * Switch `#[used]` attribute from `llvm.used` to `llvm.compiler.used` to avoid SHF_GNU_RETAIN flag introduced in https://reviews.llvm.org/D97448, which appears to trigger a bug in older versions of gold.
 * Set `LLVM_INCLUDE_TESTS=OFF` to avoid Python 3.6 requirement.

Upstream issues:

 * ~~https://bugs.llvm.org/show_bug.cgi?id=51210 (InlineAsm diagnostic reporting for module asm)~~ Fixed by https://github.com/llvm/llvm-project/commit/1558bb80c01b695ce12642527cbfccf16cf54ece.
 * ~~https://bugs.llvm.org/show_bug.cgi?id=51476 (Miscompile on AArch64 due to incorrect comparison elimination)~~ Fixed by https://github.com/llvm/llvm-project/commit/81b106584f2baf33e09be2362c35c1bf2f6bfe94.
 * https://bugs.llvm.org/show_bug.cgi?id=51207 (Can't set custom section flags anymore). Problematic change reverted in our fork, https://reviews.llvm.org/D107216 posted for upstream revert.
 * https://bugs.llvm.org/show_bug.cgi?id=51211 (Regression in codegen for #83623). This is an optimization regression that we may likely have to eat for this release. The fix for #83623 was based on an incorrect premise, and this needs to be properly addressed in the MergeICmps pass.

The [compile-time impact](https://perf.rust-lang.org/compare.html?start=ef9549b6c0efb7525c9b012148689c8d070f9bc0&end=0983094463497eec22d550dad25576a894687002) is mixed, but quite positive as LLVM upgrades go.

The LLVM 13 final release is scheduled for Sep 21st. The current nightly is scheduled for stable release on Oct 21st.

r? `@ghost`

2 years agoAlways use llvm.used for coverage symbols
Nikita Popov [Fri, 20 Aug 2021 19:13:18 +0000 (21:13 +0200)]
Always use llvm.used for coverage symbols

This follows what clang does in CoverageMappingGen. Using just
llvm.compiler.used is insufficient at least for MSVC targets.

2 years agoAuto merge of #88073 - lnicola:rust-analyzer-2021-08-16, r=lnicola
bors [Sat, 21 Aug 2021 04:20:33 +0000 (04:20 +0000)]
Auto merge of #88073 - lnicola:rust-analyzer-2021-08-16, r=lnicola

:arrow_up: rust-analyzer

2 years agoTest that incomplete inference for TAITs fail
Santiago Pastorino [Sat, 21 Aug 2021 01:50:13 +0000 (22:50 -0300)]
Test that incomplete inference for TAITs fail

2 years agoAuto merge of #88149 - Mark-Simulacrum:prep-never-type, r=jackh726
bors [Sat, 21 Aug 2021 01:29:12 +0000 (01:29 +0000)]
Auto merge of #88149 - Mark-Simulacrum:prep-never-type, r=jackh726

Refactor fallback code to prepare for never type

This PR contains cherry-picks of some of `@nikomatsakis's` work from #79366, and shouldn't (AFAICT) represent any change in behavior. However, the refactoring is good regardless of the never type work being landed, and will reduce the size of those eventual PR(s) (and rebase pain).

I am not personally an expert on this code, and the commits are essentially 100% `@nikomatsakis's,` but they do seem reasonable to me by my understanding. Happy to edit with review, of course. Commits are best reviewed in sequence rather than all together.

r? `@jackh726` perhaps?

2 years agoTest use of impl Trait in an impl as the value for an associated type in an impl...
Santiago Pastorino [Fri, 20 Aug 2021 21:06:31 +0000 (18:06 -0300)]
Test use of impl Trait in an impl as the value for an associated type in an impl trait

2 years agoStop emitting the `dso_local` LLVM attribute for external symbols under the static...
Patrick Walton [Fri, 20 Aug 2021 23:42:45 +0000 (16:42 -0700)]
Stop emitting the `dso_local` LLVM attribute for external symbols under the static relocation model on macOS.

This matches Clang's behavior:

https://github.com/llvm/llvm-project/blob/973cb2c326be9f256da0897c4d2ef117dc22761d/clang/lib/CodeGen/CodeGenModule.cpp#L1038-L1040

Even if `dso_local` were properly supported in this way on macOS, it seems
incorrect to add this annotation as liberally as we did. The `dso_local`
annotation is for symbols that ultimately end up in the same linkage unit, but
we were adding this annotation even for `static` values inside `extern` blocks
marked with `#[link(type="framework")]`, which should be considered dynamically
linked.  Note that Clang likewise avoids emitting `dso_local` for `dllimport`
symbols:

https://github.com/llvm/llvm-project/blob/973cb2c326be9f256da0897c4d2ef117dc22761d/clang/lib/CodeGen/CodeGenModule.cpp#L1005-L1007

This issue caused breakage in the `ring` crate, which links to a symbol defined
in `Security.framework` that ultimately resolves to address `0x0`:

https://github.com/briansmith/ring/blob/b94d61e044b42827fefd71d5f61e8c58a7659870/src/rand.rs#L390

For this symbol, the use of `dso_local` causes LLVM to emit a relocation of
type `X86_64_RELOC_SIGNED`, which is a 32-bit signed PC-relative offset. If the
binary is large enough, `0x0` might be out of range, and the link will fail.
Avoiding `dso_local` causes LLVM to use the GOT instead, emitting a relocation
of type `X86_64_RELOC_GOT_LOAD`, which will properly handle the large offset
and cause the link to succeed.

As a side note, the static relocation model is effectively deprecated for
security reasons on macOS, as it prohibits PIE. It's also completely
unsupported on Apple Silicon, so I don't think it's worth going to the effort
of properly supporting this model on that platform.

2 years agoAuto merge of #88087 - jesyspa:issue-87935-box, r=jackh726
bors [Fri, 20 Aug 2021 23:04:57 +0000 (23:04 +0000)]
Auto merge of #88087 - jesyspa:issue-87935-box, r=jackh726

Check that a box expression's type is Sized

This resolves [issue 87935](https://github.com/rust-lang/rust/issues/87935).

This makes E0161 (move from an unsized rvalue) much less common.  I've replaced the test to use [this case](https://github.com/rust-lang/rust/blob/master/src/test/ui/object-safety/object-safety-by-value-self-use.rs), when a boxed `dyn` trait is passed by value, but that isn't an error when `unsized_locals` is enabled.  I think it may be possible to get rid of E0161 entirely by checking that case earlier, but I'm not sure if that's desirable?

2 years agoTest tait use in a fn type
Santiago Pastorino [Fri, 20 Aug 2021 21:32:57 +0000 (18:32 -0300)]
Test tait use in a fn type

2 years agomove `fallback_if_possible` and friends to fallback.rs
Niko Matsakis [Sun, 22 Nov 2020 11:10:50 +0000 (06:10 -0500)]
move `fallback_if_possible` and friends to fallback.rs

Along the way, simplify and document the logic more clearly.

2 years agocreate `Coercion` obligations given 2 unbound type variables
Niko Matsakis [Sat, 21 Nov 2020 20:47:14 +0000 (15:47 -0500)]
create `Coercion` obligations given 2 unbound type variables

Motivation: in upcoming commits, we are going to create a graph of the
coercion relationships between variables. We want to
distinguish *coercion* specifically from other sorts of subtyping, as
it indicates values flowing from one place to another via assignment.

2 years agomove the `sub-unify` check and extend the documentation a bit
Niko Matsakis [Sat, 21 Nov 2020 20:33:36 +0000 (15:33 -0500)]
move the `sub-unify` check and extend the documentation a bit

I didn't like the sub-unify code executing when a predicate was
ENQUEUED, that felt fragile. I would have preferred to move the
sub-unify code so that it only occurred during generalization, but
that impacted diagnostics, so having it also occur when we process
subtype predicates felt pretty reasonable. (I guess we only need one
or the other, but I kind of prefer both, since the generalizer
ultimately feels like the *right* place to guarantee the properties we
want.)

2 years agoUse of impl trait in an impl as the valoe for an associated type in a dyn
Santiago Pastorino [Fri, 20 Aug 2021 20:33:21 +0000 (17:33 -0300)]
Use of impl trait in an impl as the valoe for an associated type in a dyn

2 years agoAdd TAIT struct test
Santiago Pastorino [Fri, 20 Aug 2021 19:24:05 +0000 (16:24 -0300)]
Add TAIT struct test

2 years agoAuto merge of #83093 - the8472:smaller-instant-hammer, r=Amanieu
bors [Fri, 20 Aug 2021 19:06:46 +0000 (19:06 +0000)]
Auto merge of #83093 - the8472:smaller-instant-hammer, r=Amanieu

where available use AtomicU{64,128} instead of mutex for Instant backsliding protection

This decreases the overhead of backsliding protection on x86 systems with unreliable TSC, e.g. windows. And on aarch64 systems where 128bit atomics are available.

The following benchmarks were taken on x86_64 linux though by overriding `actually_monotonic()`, the numbers may look different on other platforms

```
# actually_monotonic() == true
test time::tests::instant_contention_01_threads                   ... bench:          44 ns/iter (+/- 0)
test time::tests::instant_contention_02_threads                   ... bench:          44 ns/iter (+/- 0)
test time::tests::instant_contention_04_threads                   ... bench:          44 ns/iter (+/- 0)
test time::tests::instant_contention_08_threads                   ... bench:          44 ns/iter (+/- 0)
test time::tests::instant_contention_16_threads                   ... bench:          44 ns/iter (+/- 0)

# 1x AtomicU64
test time::tests::instant_contention_01_threads                   ... bench:          65 ns/iter (+/- 0)
test time::tests::instant_contention_02_threads                   ... bench:         157 ns/iter (+/- 20)
test time::tests::instant_contention_04_threads                   ... bench:         281 ns/iter (+/- 53)
test time::tests::instant_contention_08_threads                   ... bench:         555 ns/iter (+/- 77)
test time::tests::instant_contention_16_threads                   ... bench:         883 ns/iter (+/- 107)

# mutex
test time::tests::instant_contention_01_threads                   ... bench:          60 ns/iter (+/- 2)
test time::tests::instant_contention_02_threads                   ... bench:         770 ns/iter (+/- 231)
test time::tests::instant_contention_04_threads                   ... bench:       1,347 ns/iter (+/- 45)
test time::tests::instant_contention_08_threads                   ... bench:       2,693 ns/iter (+/- 114)
test time::tests::instant_contention_16_threads                   ... bench:       5,244 ns/iter (+/- 487)
```

Since I don't have an arm machine with 128bit atomics I wasn't able to benchmark the AtomicU128 implementation.

2 years agofix tests on wasm targets that have 32bit time_t and don't have threads
The8472 [Wed, 18 Aug 2021 20:44:10 +0000 (22:44 +0200)]
fix tests on wasm targets that have 32bit time_t and don't have threads

2 years agoAuto merge of #88176 - erikdesjardins:rezst, r=oli-obk
bors [Fri, 20 Aug 2021 16:25:34 +0000 (16:25 +0000)]
Auto merge of #88176 - erikdesjardins:rezst, r=oli-obk

Reenable RemoveZsts

Now that the underlying issue has been fixed by #88124, we can reland #83417.

r? `@oli-obk`

2 years agoUpdate old box expression tests and add a new one.
Anton Golov [Fri, 20 Aug 2021 14:26:24 +0000 (16:26 +0200)]
Update old box expression tests and add a new one.

New tests also check that we're not triggering this error
over-zealously.

2 years agoRequire a box expression's type to be Sized
Anton Golov [Fri, 20 Aug 2021 14:25:52 +0000 (16:25 +0200)]
Require a box expression's type to be Sized

2 years agotest TAIT in different positions
Santiago Pastorino [Fri, 20 Aug 2021 14:15:48 +0000 (11:15 -0300)]
test TAIT in different positions

2 years agoChange example and tests for E0161.
Anton Golov [Fri, 20 Aug 2021 13:59:42 +0000 (15:59 +0200)]
Change example and tests for E0161.

The code will not emit this warning once box expressions require a sized
type (since that error is emitted earlier in the flow).

2 years agoAuto merge of #87686 - matthiaskrgr:clippy_august_21_perf, r=jackh726
bors [Fri, 20 Aug 2021 13:41:42 +0000 (13:41 +0000)]
Auto merge of #87686 - matthiaskrgr:clippy_august_21_perf, r=jackh726

clippy::perf fixes

2 years agoWe meant to use a trait instead of lifetime here
Santiago Pastorino [Fri, 20 Aug 2021 13:36:24 +0000 (10:36 -0300)]
We meant to use a trait instead of lifetime here

2 years agoAuto merge of #87329 - sunfishcode:sunfishcode/io-safety, r=joshtriplett
bors [Fri, 20 Aug 2021 11:00:55 +0000 (11:00 +0000)]
Auto merge of #87329 - sunfishcode:sunfishcode/io-safety, r=joshtriplett

I/O safety.

Introduce `OwnedFd` and `BorrowedFd`, and the `AsFd` trait, and
implementations of `AsFd`, `From<OwnedFd>` and `From<T> for OwnedFd`
for relevant types, along with Windows counterparts for handles and
sockets.

Tracking issue: <https://github.com/rust-lang/rust/issues/87074>

RFC: <https://github.com/rust-lang/rfcs/blob/master/text/3128-io-safety.md>

Highlights:
 - The doc comments at the top of library/std/src/os/unix/io/mod.rs and library/std/src/os/windows/io/mod.rs
 - The new types and traits in library/std/src/os/unix/io/fd.rs and library/std/src/os/windows/io/handle.rs
 - The removal of the `RawHandle` struct the Windows impl, which had the same name as the `RawHandle` type alias, and its functionality is now folded into `Handle`.

Managing five levels of wrapping (File wraps sys::fs::File wraps sys::fs::FileDesc wraps OwnedFd wraps RawFd, etc.) made for a fair amount of churn and verbose as/into/from sequences in some places. I've managed to simplify some of them, but I'm open to ideas here.

r? `@joshtriplett`

2 years agoAdd support for including non-backport commits.
Mukund Lakshman [Fri, 20 Aug 2021 00:15:35 +0000 (01:15 +0100)]
Add support for including non-backport commits.

2 years agoAdd cherry-pick.sh convenience script.
Mukund Lakshman [Thu, 19 Aug 2021 20:58:17 +0000 (21:58 +0100)]
Add cherry-pick.sh convenience script.

2 years agoMailmap entry for myself
Frank Steffahn [Fri, 20 Aug 2021 10:11:45 +0000 (12:11 +0200)]
Mailmap entry for myself

2 years agoAuto merge of #88000 - bjorn3:fix_cg_llvm_clif_compile, r=Mark-Simulacrum
bors [Fri, 20 Aug 2021 07:30:28 +0000 (07:30 +0000)]
Auto merge of #88000 - bjorn3:fix_cg_llvm_clif_compile, r=Mark-Simulacrum

Fix compiling other codegen backends when llvm is enabled

Extracted from #81746

Without this change rustbuild will not pass the required linker argument to find libllvm. While other backends likely don't use libllvm, it is necessary to be able to link against rustc_driver as the llvm backend is linked into it.

2 years agoAuto merge of #86898 - the8472:path-cmp, r=dtolnay
bors [Fri, 20 Aug 2021 05:00:45 +0000 (05:00 +0000)]
Auto merge of #86898 - the8472:path-cmp, r=dtolnay

Add fast path for Path::cmp that skips over long shared prefixes

```
# before
test path::tests::bench_path_cmp_fast_path_buf_sort               ... bench:      60,811 ns/iter (+/- 865)
test path::tests::bench_path_cmp_fast_path_long                   ... bench:       6,459 ns/iter (+/- 275)
test path::tests::bench_path_cmp_fast_path_short                  ... bench:       1,777 ns/iter (+/- 34)

# after
test path::tests::bench_path_cmp_fast_path_buf_sort               ... bench:      38,140 ns/iter (+/- 211)
test path::tests::bench_path_cmp_fast_path_long                   ... bench:       1,471 ns/iter (+/- 24)
test path::tests::bench_path_cmp_fast_path_short                  ... bench:       1,106 ns/iter (+/- 9)
```

2 years agoRevert "Revert "Auto merge of #83417 - erikdesjardins:enableremovezsts, r=oli-obk""
Erik Desjardins [Fri, 20 Aug 2021 02:28:32 +0000 (22:28 -0400)]
Revert "Revert "Auto merge of #83417 - erikdesjardins:enableremovezsts, r=oli-obk""

This reverts commit 8e11199a153218c13a419df37a9bb675181cccb7.

2 years agoAuto merge of #88039 - sexxi-goose:fix-87987, r=nikomatsakis
bors [Fri, 20 Aug 2021 02:19:58 +0000 (02:19 +0000)]
Auto merge of #88039 - sexxi-goose:fix-87987, r=nikomatsakis

RFC2229 Only compute place if upvars can be resolved

Closes https://github.com/rust-lang/rust/issues/87987

This PR fixes an ICE when trying to unwrap an Err. This error appears when trying to convert a PlaceBuilder into Place when upvars can't yet be resolved. We should only try to convert a PlaceBuilder into Place if upvars can be resolved.

r? `@nikomatsakis`

2 years agoAuto merge of #87996 - sexxi-goose:fix-87988, r=nikomatsakis
bors [Thu, 19 Aug 2021 23:33:22 +0000 (23:33 +0000)]
Auto merge of #87996 - sexxi-goose:fix-87988, r=nikomatsakis

RFC2229 Add missing edge case

Closes https://github.com/rust-lang/rust/issues/87988

This PR fixes an ICE where a match discriminant is not being read when expected. This ICE was the result of a missing edge case which assumed that if a pattern is of type `PatKind::TupleStruct(..) | PatKind::Path(..) | PatKind::Struct(..) | PatKind::Tuple(..)` then a place could only be a multi variant if the place is of type kind Adt.

2 years agoFix doc test failures on Windows.
Dan Gohman [Thu, 19 Aug 2021 23:15:29 +0000 (16:15 -0700)]
Fix doc test failures on Windows.

2 years agoCI: Verify commits in beta & stable are in upstream branches.
Mukund Lakshman [Thu, 29 Jul 2021 17:00:13 +0000 (18:00 +0100)]
CI: Verify commits in beta & stable are in upstream branches.

Closes #74721

2 years agoTest that type alias impl trait happens in a submodule
Santiago Pastorino [Thu, 19 Aug 2021 21:18:34 +0000 (18:18 -0300)]
Test that type alias impl trait happens in a submodule

2 years agointroduce a Coerce predicate
Niko Matsakis [Sat, 21 Nov 2020 12:06:16 +0000 (07:06 -0500)]
introduce a Coerce predicate

2 years agocleanup: extract a helper for coercion from inference variables
Niko Matsakis [Sat, 21 Nov 2020 10:45:13 +0000 (05:45 -0500)]
cleanup: extract a helper for coercion from inference variables

2 years agoshallow resolve target type in coercion
Niko Matsakis [Fri, 20 Nov 2020 10:52:27 +0000 (05:52 -0500)]
shallow resolve target type in coercion

We used to avoid doing this because we didn't want to make coercion depend on
the state of inference. For better or worse, we have moved away from this
position over time. Therefore, I am going to go ahead and resolve the `b`
target type early on so that it is done uniformly.

(The older technique for managing this was always something of a hack
regardless; if we really wanted to avoid integrating coercion and inference we
needed to be more disciplined about it.)

2 years agointroduce a `Diverging` enum instead of a bool
Niko Matsakis [Fri, 20 Nov 2020 10:40:33 +0000 (05:40 -0500)]
introduce a `Diverging` enum instead of a bool

2 years agofactor fallback code into its own module
Niko Matsakis [Thu, 19 Nov 2020 23:24:03 +0000 (18:24 -0500)]
factor fallback code into its own module

2 years agoUpdate release note for 1.55.0.
Joe Ardent [Thu, 19 Aug 2021 20:54:26 +0000 (13:54 -0700)]
Update release note for 1.55.0.

Added a line about new formatting option, `{lib}`, for `cargo
tree` (https://github.com/rust-lang/cargo/pull/9663).

2 years agoAuto merge of #88165 - GuillaumeGomez:rollup-4o0v2ps, r=GuillaumeGomez
bors [Thu, 19 Aug 2021 20:31:05 +0000 (20:31 +0000)]
Auto merge of #88165 - GuillaumeGomez:rollup-4o0v2ps, r=GuillaumeGomez

Rollup of 8 pull requests

Successful merges:

 - #86123 (Preserve more spans in internal `rustc_queries!` macro)
 - #87874 (Add TcpStream type to TcpListener::incoming docs)
 - #88034 (rustc_privacy: Replace `HirId`s and `DefId`s with `LocalDefId`s where possible)
 - #88050 (Remove `HashStable` impls for `FileName` and `RealFileName`)
 - #88093 ([rustdoc] Wrap code blocks in `<code>` tag)
 - #88146 (Add tests for some `feature(const_evaluatable_checked)` incr comp issues)
 - #88153 (Update .mailmap)
 - #88159 (Use a trait instead of the now disallowed missing trait there)

Failed merges:

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

2 years agoFactor out a common `RawFd`/`AsRawFd`/etc for Unix and WASI.
Dan Gohman [Thu, 19 Aug 2021 19:24:25 +0000 (12:24 -0700)]
Factor out a common `RawFd`/`AsRawFd`/etc for Unix and WASI.

2 years agomove code to `InferCtxt` method
Ellen [Thu, 19 Aug 2021 20:04:04 +0000 (21:04 +0100)]
move code to `InferCtxt` method

2 years agoFix syntax for non-doc comments, and use `crate::` instead of `std::`.
Dan Gohman [Thu, 19 Aug 2021 19:23:04 +0000 (12:23 -0700)]
Fix syntax for non-doc comments, and use `crate::` instead of `std::`.

2 years agoFix clippy let expressions fallout
Cameron Steffen [Thu, 19 Aug 2021 18:31:25 +0000 (13:31 -0500)]
Fix clippy let expressions fallout

2 years agoAdd I/O safety trait impls for process::Stdio and process::Child.
Dan Gohman [Tue, 17 Aug 2021 23:28:00 +0000 (16:28 -0700)]
Add I/O safety trait impls for process::Stdio and process::Child.

2 years agoUse the correct `into_*` on Windows to avoid dropping a stdio handle.
Dan Gohman [Fri, 13 Aug 2021 05:00:06 +0000 (22:00 -0700)]
Use the correct `into_*` on Windows to avoid dropping a stdio handle.

Use `into_raw_handle()` rather than `into_inner()` to completely consume a
`Handle` without dropping its contained handle.

2 years agoRemove the `#![feature(io_safety)]` from lib.rs.
Dan Gohman [Fri, 6 Aug 2021 13:11:22 +0000 (06:11 -0700)]
Remove the `#![feature(io_safety)]` from lib.rs.

2 years agoFix an unused import warning.
Dan Gohman [Wed, 4 Aug 2021 12:58:06 +0000 (07:58 -0500)]
Fix an unused import warning.

2 years agoUpdate PidFd for the new I/O safety APIs.
Dan Gohman [Tue, 3 Aug 2021 12:31:59 +0000 (07:31 -0500)]
Update PidFd for the new I/O safety APIs.

2 years agoRename OptionFileHandle to HandleOrInvalid and make it just wrap an Option<OwnedHandle>
Dan Gohman [Thu, 29 Jul 2021 20:36:44 +0000 (13:36 -0700)]
Rename OptionFileHandle to HandleOrInvalid and make it just wrap an Option<OwnedHandle>

The name (and updated documentation) make the FFI-only usage clearer, and wrapping Option<OwnedHandle> avoids the need to write a separate Drop or Debug impl.

Co-authored-by: Josh Triplett <josh@joshtriplett.org>
2 years agoDon't encourage migration until io_safety is stablized.
Dan Gohman [Wed, 28 Jul 2021 16:00:18 +0000 (09:00 -0700)]
Don't encourage migration until io_safety is stablized.

2 years agoFactor out Unix and WASI fd code into a common module.
Dan Gohman [Wed, 28 Jul 2021 14:28:58 +0000 (07:28 -0700)]
Factor out Unix and WASI fd code into a common module.

2 years agoSynchronize minor differences between Unix and WASI implementations.
Dan Gohman [Wed, 28 Jul 2021 14:02:51 +0000 (07:02 -0700)]
Synchronize minor differences between Unix and WASI implementations.

2 years agoAdd more comments about the `INVALID_HANDLE_VALUE` situation.
Dan Gohman [Wed, 28 Jul 2021 13:52:45 +0000 (06:52 -0700)]
Add more comments about the `INVALID_HANDLE_VALUE` situation.

2 years agoAdd comments about impls for File, TcpStream, ChildStdin, etc.
Dan Gohman [Wed, 28 Jul 2021 13:28:06 +0000 (06:28 -0700)]
Add comments about impls for File, TcpStream, ChildStdin, etc.

2 years agoFix copypasta of "Unix" within the WASI directory.
Dan Gohman [Wed, 28 Jul 2021 00:33:55 +0000 (17:33 -0700)]
Fix copypasta of "Unix" within the WASI directory.

2 years agoReword the description of dup2/dup3.
Dan Gohman [Wed, 28 Jul 2021 00:32:04 +0000 (17:32 -0700)]
Reword the description of dup2/dup3.

2 years agoAdd Safety comments to the `As*` for `Owned*` implementations.
Dan Gohman [Wed, 28 Jul 2021 00:15:40 +0000 (17:15 -0700)]
Add Safety comments to the `As*` for `Owned*` implementations.