| Age | Commit message (Collapse) | Author | Lines |
|---|
|
|
|
|
|
|
|
Rollup of 6 pull requests
Successful merges:
- #107043 (Support `true` and `false` as boolean flag params)
- #107831 (Query refactoring)
- #107841 (Handled snap curl issue inside Rust)
- #107852 (rustdoc: remove unused fn parameter `tab`)
- #107861 (Sync release notes for 1.67.1)
- #107863 (Allow multiple candidates with same response in new solver)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
|
|
r=jackh726
Allow multiple candidates with same response in new solver
Treat >1 candidates as *not* ambiguous if they return the same response.
|
|
Query refactoring
Just some cleanups I found when learning about the query system.
Best reviewed one commit at a time.
r? `@oli-obk`
|
|
Support `true` and `false` as boolean flag params
Implements [MCP 577](https://github.com/rust-lang/compiler-team/issues/577).
|
|
Fix implied outlives bounds logic for projections
The logic here is subtly wrong. I put a bit of an explanation in a767d7b5165cea8ee5cbe494a4a636c50ef67c9c.
TL;DR: we register outlives predicates to be proved, because wf code normalizes projections (from the unnormalized types) to type variables. This causes us to register those as constraints instead of implied. This was "fine", because we later added that implied bound in the normalized type, and delayed registering constraints. When I went to cleanup `free_region_relations` to *not* delay adding constraints, this bug was uncovered.
cc. `@aliemjay` because this caused your test failure in #99832 (I only realized as I was writing this)
r? `@nikomatsakis`
|
|
normalization
|
|
|
|
|
|
Add `-Z instrument-xray` flag
Implement MCP https://github.com/rust-lang/compiler-team/issues/561, adding `-Z instrument-xray` flag which enables XRay instrumentation in LLVM.
|
|
Disqualify `auto trait` built-in impl in new solver if explicit `impl` exists
|
|
eggyal:do_not_bring_trait_alias_supertraits_into_scope, r=compiler-errors
Do not bring trait alias supertraits into scope
Fixes #107747
cc #41517
|
|
Implement some tweaks in the new solver
I've been testing the new solver on some small codebases, and these are a few small changes I've needed to make.
The most "controversial" here is implementing `trait_candidate_should_be_dropped_in_favor_of`, which I just implemented to always return false. This surprisingly allows some code to compile, without us having to actually decide on any semantics yet.
r? `@rust-lang/initiative-trait-system-refactor`
|
|
test: snapshot for derive suggestion in diff files
fixed #107649
|
|
unused-lifetimes: don't warn about lifetimes originating from expanded code
previously, we would warn like this:
````
warning: lifetime parameter `'s` never used
--> /tmp/unusedlif/code.rs:6:62
|
5 | #[derive(Clone)]
| - help: elide the unused lifetime
6 | struct ShimMethod4<T: Trait2 + 'static>(pub &'static dyn for<'s> Fn(&'s mut T::As));
| ^^
|
= note: requested on the command line with `-W unused-lifetimes`
````
Fixes #104432
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Sync rustc_codegen_cranelift
* Couple of bugfixes
* A significant runtime perf improvement
* Implemented sym and const support for inline asm
* Improved self profile integration
r? `@ghost`
`@rustbot` label +A-codegen +A-cranelift +T-compiler
|
|
sync_cg_clif-2023-02-09
|
|
use idiomatic formatting
Also, remove needless `else`
|
|
Specify dlltool prefix when generating import libs
Ref: https://github.com/rust-lang/rust/pull/106610#issuecomment-1418221274
tl;dr: This PR adds an explicit dlltool temporary filename prefix. The prefix resolves a race condition by ensuring dlltool temporary files are siloed in an appropriate/unique Rust temporary directory.
---
GNU dlltool, as part of its import library generation logic, uses a bunch of temporary files on disk. In the interest of deterministic build runs, dlltool supports deterministic temporary filenames. The temporary filename prefix is automatically generated internally or can be explicitly specified via a `--temp-prefix` argument.
GNU dlltool **2.38** (that ships with `x86_64-12.2.0-release-posix-seh-rt_v10-rev0` [installed during CI](https://github.com/rust-lang/rust/blob/master/src/ci/scripts/install-mingw.sh)) generates a prefix based on the target library name ([source](https://sourceware.org/git/?p=binutils-gdb.git;a=blob;f=binutils/dlltool.c;h=d95bf3f5470b999fa3b30bc887791859f48d81d1;hb=20756b0fbe065a84710aa38f2457563b57546440#l3992)). The tool writes to files such as `target_dll_h.s` and `target_dll_s00203.o` in the current working directory.
This presents a problem when multiple instances of rustc_codegen_llvm are running to generate an import library (as part of the raw_dylib feature) for the same target library (e.g. kernel32) ([source](https://github.com/rust-lang/rust/blob/master/compiler/rustc_codegen_llvm/src/back/archive.rs#L185-L196)). That is, dlltool instances race and may overwrite or delete files belonging to each other.
GNU dlltool **2.39**+ (not used in Rust CI) generates a prefix based on the output library path ([source](https://sourceware.org/git/?p=binutils-gdb.git;a=blob;f=binutils/dlltool.c;h=e2af20847009945b4c61a6fef08268fbb4429715;hb=b51c2fec1da205ea3e7354cbb3e253018d64873c#l3992)). The tool, when invoked as part of rustc_codegen_llvm, writes to files at paths such as `C_Users_Foo_AppData_Local_Temp_rustcOFqhXZ_target_lib_h.s`. (The output library path is normalized and non-alphanumeric characters are replaced with underscores.)
|
|
Migrate some of `rustc_parse` to derive diagnostics
`@rustbot` label +A-translation
r? rust-lang/diagnostics
cc #100717
|
|
Also, remove needless else
|
|
Add missing normalization for union fields types
Overshadows https://github.com/rust-lang/rust/pull/106808.
From the experiment https://github.com/rust-lang/rust/pull/103985.
In short, it allows to use projections as a type for union's fields.
cc `@compiler-errors`
r? `@oli-obk`
|
|
There is a type `QueryCtxt`, which impls the trait `QueryContext`.
Confusingly, there is another type `QueryContext`. The latter is (like
`TyCtxt`) just a pointer to a `GlobalContext`. It's not used much, e.g.
its `impl` block has a single method.
This commit removes `QueryContext`, replacing its use with direct
`GlobalCtxt` use.
|
|
These are in places where a `tcx` is easily obtained.
|
|
|
|
|
|
r=estebank
Do not eagerly recover for bad `impl Trait` types in macros
Fixes #107796
cc #106712, ```@estebank``` and ```@Ezrashaw``` please make sure to use [`Parser::may_recover`](https://doc.rust-lang.org/nightly/nightly-rustc/rustc_parse/parser/struct.Parser.html#method.may_recover) for all eager-token-consuming parser recoveries.
This also fixes a separate regression from #99915, that was introduced before we added `may_recover` though.
|
|
correctly update goals in the cache
we may want to actually write the response for our goal into the provisional or global cache instead of simply using the result from the last iteration '^^
r? ```@rust-lang/initiative-trait-system-refactor```
|
|
Replace a command line flag with an env var to allow tools to initialize the tracing loggers at their own discretion
fixes https://github.com/rust-lang/miri/issues/2778
this was introduced in https://github.com/rust-lang/rust/pull/104645, so this PR reverts the flag-part and uses an env var instead.
|
|
nnethercote:extend-BYTE_SLICE_IN_PACKED_STRUCT_WITH_DERIVE, r=RalfJung
Extend `BYTE_SLICE_IN_PACKED_STRUCT_WITH_DERIVE`.
To temporarily allow a `str` field in a packed struct using `derive`, along with `[u8]`.
r? ``@RalfJung``
|
|
This is somewhat important because LLVM enables the pass based on
target architecture, but support by the target OS also matters.
For example, XRay attributes are processed by codegen for macOS
targets, but Apple linker fails to process relocations in XRay
data sections, so the feature as a whole is not supported there
for the time being.
|
|
Specify where XRay is supported. I only test ARM64 and x86_64, but hey
those others should work too, right? LLVM documentation says that MIPS
and PPC are also supported, but I don't have the hardware, so I won't
pretend. Naturally, more targets can be added later with more testing.
|
|
Add the attributes to functions according to the settings.
"xray-always" overrides "xray-never", and they both override
"xray-ignore-loops" and "xray-instruction-threshold", but we'll
let lints deal with warnings about silly attribute combinations.
|
|
Four because that's the new reasonable maximum for XRay instrumentation
attributes in the following commit.
|
|
Recognize all bells and whistles that LLVM's XRay pass is capable of.
The always/never settings are a bit dumb without attributes but they're
still there. The default instruction count is chosen by the compiler,
not LLVM pass. We'll do it later.
|
|
To temporarily allow a `str` field in a packed struct using `derive`,
along with `[u8]`.
|
|
Optimize `TyKind::eq`.
r? `@ghost`
|
|
Rollup of 8 pull requests
Successful merges:
- #105641 (Implement cursors for BTreeMap)
- #107271 (Treat Drop as a rmw operation)
- #107710 (Update strip-ansi-escapes and vte)
- #107758 (Change `arena_cache` to not alter the declared query result)
- #107777 (Make `derive_const` derive properly const-if-const impls)
- #107780 (Rename `replace_bound_vars_with_*` to `instantiate_binder_with_*`)
- #107793 (Add missing tracking issue for `RawOsError`)
- #107807 (Fix small debug typo)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
|
|
|
|
|
|
|
|
Rename `replace_bound_vars_with_*` to `instantiate_binder_with_*`
Mentioning "binder" rather than "bound vars", imo, makes it clearer that we're doing something to the binder as a whole.
Also, "instantiate" is the verb that I'm always reaching for when I'm looking for these functions, and the name that we use in the new solver anyways.
r? types
|
