The primary reason for this is that make can result in a substantial
under utilization of parallelism, mostly due to the submake structure
preventing good dependency tracking and scheduling.
In f758c7b2a7 (Debian 6 doesn't have ninja, so use make for the dist builds)
llvm.ninja was disabled due to lack of distro package. This is no longer the
case with the CentOS 7 base, so bring ninja back for a performance boost.
Let expressions on RHS shouldn't be terminating scopes
Fixes#100276.
Before this PR, we were unconditionally marking the RHS of short-circuiting binary expressions as a terminating scope.
In the case of a let chain where the `let` expression was on the RHS, this meant that temporaries within the `let` expr would only live until the end of the expression. Since this only affected the RHS, this led to surprising behavior ([example](https://play.rust-lang.org/?version=nightly&mode=debug&edition=2021&gist=d1b0a5d1f01882f9c89c2194a75eb19f)).
After this PR, we only mark the RHS as a terminating scope if it is not a `let` expression.
Standardize "use parentheses to call" suggestions between typeck and trait selection
1. Suggest calling constructors, since they're basically `FnDef`s but they have a different def kind and hir representation, so we were leaving them out.
2. Standardize the call suggestions between trait fulfillment errors and type mismatch. In the type mismatch suggestion, we suggest `/* Ty */` as the placeholder for an arg, and not the parameter's name, which is less helpful.
3. Use `predicate_must_hold_modulo_regions` instead of matching on `EvaluationResult` -- this might cause some suggestions to be filtered out, but we really shouldn't be suggesting a call if it "may" hold, only when it "must" hold.
4. Borrow some logic from `extract_callable_info` to generalize this suggestion to fn pointers, type parameters, and opaque types.
Fixes#102852
Handle core dumps output in QEMU user mode
In addition to the whole-system emulation/virtualization, QEMU also supports user-mode emulation, where the emulation happens as a normal process inside the parent system. This allows running most tests by simply spawning remote-test-server inside user-mode emulation.
Unfortunately, QEMU always writes its own message in addition to the system one when a core dump happens, which breaks a few tests which match on the exact output of the system.
This PR changes those tests to strip the (possible) QEMU output before checking if the output is expected.
Partially fix `src/test/run-make/coverage-reports` when cross-compiling
The test does not work on cross-compiled targets because the --target flag was not passed to rustc inside the test. This commit fixes that by adding the flag to the invocations.
Note that the test still fails on cross-compiled targets using remote-test, as remote-test is not capable (yet) of sending back to the host system the `*.profraw` file generated by the instrumentation.
Because of that, this is only a partial fix, and the test has been ignored on cross-compilation.
Rollup of 6 pull requests
Successful merges:
- #101889 (doc: rewrite doc for uint::{carrying_add,borrowing_sub})
- #102507 (More slice::partition_point examples)
- #103164 (rustdoc: remove CSS ``@media` (min-width: 701px)`)
- #103189 (Clean up code-color and headers-color rustdoc GUI tests)
- #103203 (Retrieve LLVM version from llvm-filecheck binary if it is not set yet)
- #103204 (Add some more autolabels)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
By moving `RenderOptions` out of `Option`, because the two structs' uses
are almost entirely separate.
The only complication is that `unstable_features` is needed in both
structs, but it's a tiny `Copy` type so its duplication seems fine.
It turns out `markdown::render` is more complex than it first appears,
because it can invoke `doctest::make_test`, which requires session
globals and a thread pool.
So this commit changes it to use `interface::run_compiler`. Three of the
four paths in `main_args` now use `interface::run_compiler`.
rustc's startup has several layers, including:
- `interface::run_compiler` passes a closure, `f`, to
`run_in_thread_pool_with_globals`, which creates a thread pool, sets
up session globals, and passes `f` to `create_compiler_and_run`.
- `create_compiler_and_run` creates a `Session`, a `Compiler`, sets the
source map, and calls `f`.
rustdoc is a bit different.
- `main_args` calls `main_options` via
`run_in_thread_pool_with_globals`, which (again) creates a thread pool
(hardcoded to a single thread!) and sets up session globals.
- `main_options` has four different paths.
- The second one calls `interface::run_compiler`, which redoes the
`run_in_thread_pool_with_globals`! This is bad.
- The fourth one calls `interface::create_compiler_and_run`, which is
reasonable.
- The first and third ones don't do anything of note involving the
above functions, except for some symbol interning which requires
session globals.
In other words, rustdoc calls into `rustc_interface` at three different
levels. It's a bit confused, and feels like code where functionality has
been added by different people at different times without fully
understanding how the globally accessible stuff is set up.
This commit tidies things up. It removes the
`run_in_thread_pool_with_globals` call in `main_args`, and adjust the
four paths in `main_options` as follows.
- `markdown::test` calls `test::test_main`, which provides its own
parallelism and so doesn't need a thread pool. It had one small use of
symbol interning, which required session globals, but the commit
removes this.
- `doctest::run` already calls `interface::run_compiler`, so it doesn't
need further adjustment.
- `markdown::render` is simple but needs session globals for interning
(which can't easily be removed), so it's now wrapped in
`create_session_globals_then`.
- The fourth path now uses `interface::run_compiler`, which is
equivalent to the old `run_in_thread_pool_with_globals` +
`create_compiler_and_run` pairing.
Retrieve LLVM version from llvm-filecheck binary if it is not set yet
In `rustc_codegen_gcc`, we run the `ASM` test suite. The problem is that, if a too recent version of the `llvm-filecheck` binary is provided, an extra argument needs to be passed and the to detect this version, it currently only expects a `--llvm-version` argument. With this, the version can be determined directly from the `llvm-filecheck` binary.
r? ``@Amanieu``
Make diagnostic for unsatisfied `Termination` bounds more precise
Don't blindly emit a diagnostic claiming that “*`main` has an invalid return type*” if we encounter a type that should but doesn't implement `std::process::Termination` and isn't actually the return type of the program entry `main`.
Fixes#103052.
``@rustbot`` label A-diagnostics T-compiler T-libs
r? diagnostics
In addition to the whole-system emulation/virtualization, QEMU also
supports user-mode emulation, where the emulation happens as a normal
process inside the parent system. This allows running most tests by
simply spawning remote-test-server inside user-mode emulation.
Unfortunately, QEMU always writes its own message in addition to the
system one when a core dump happens, which breaks a few tests which
match on the exact output of the system.
This PR changes those tests to strip the (possible) QEMU output before
checking if the output is expected.
The test does not work on cross-compiled targets because the --target
flag was not passed to rustc inside the test. This commit fixes that by
adding the flag to the invocations.
Note that the test still fails on cross-compiled targets using
remote-test, as remote-test is not capable (yet) of sending back to the
host system the *.profraw file generated by the instrumentation.
Because of that, this is only a partial fix, and the test has been
ignored on cross-compilation.
PR #98758 introduced code to avoid redundant assertions in derived code
like this:
```
let _: ::core::clone::AssertParamIsClone<u32>;
let _: ::core::clone::AssertParamIsClone<u32>;
```
But the predicate `is_simple_path` introduced as part of this failed to
account for generic arguments. Therefore the deriving code erroneously
considers types like `Option<bool>` and `Option<f32>` to be the same.
This commit fixes `is_simple_path`.
Fixes#103157.
The two rules within it can and should be done without the separate
media query:
* There ain't no rule saying a viewport can't be `700.5px` wide, since
hardware pixels can be finer than CSS pixels.
* The rule for the first example-wrap child should probably apply
on mobile.
* The rule for the source sidebar is overriden by the mobile rule
setting `max-width: 100vw`, so it can be merged with the rest
of the styles.
