this improves typeck performance by 5% (LLVM times are still huge).
Basically fixes#25916 (still O(n^2), but the example takes <1s to
compile).
r? @nikomatsakis
This PR disallows non-inline modules without path annotations that are either in a block or in an inline module whose containing file is not a directory owner (fixes#29765).
This is a [breaking-change].
r? @nikomatsakis
The issue was that the const evaluator was returning an error because
the feature flag const_indexing wasn't turned on. The error was then
reported as a bug.
Fixes#29914
Because we no longer use `GetFileAttributesExW` FileAttr is never created directly from `WIN32_FILE_ATTRIBUTE_DATA` anymore. So we should no longer store FileAttr's attributes in that c struct.
r? @alexcrichton
Is this what you had in mind?
use CXX value found at configure time inside run-make tests.
it permits OpenBSD to pass llvm-module-pass test (which use CXX
variable).
r? @alexcrichton
This follows the pattern already used for stat functions from #31551. Now
`ftruncate`, `lseek`, and `readdir_r` use their explicit 64-bit variants for
LFS support, using wider `off_t` and `dirent` types. This also updates to
`open64`, which uses no different types but implies the `O_LARGEFILE` flag.
Non-Linux platforms just map their normal functions to the 64-bit names.
r? @alexcrichton
This enables `*` in all type positions in doc searches, which I often
want in order to find functions that create or convert specific
types (e.g. `* -> vec`) but I don't actually know what kinds of input
they expect.
I actually started working on this because of #31598, but I've wanted it
several times when exploring new crates.
This allows obtaining a textual MIR dump for individual items or all items in the crate.
I haven't added any tests since ~~I'm too lazy~~ this is an unstable debugging option, but I'll add one if required.
MIR for a single function can now be dumped using `rustc -Zunstable-options --unpretty mir=my_function` and no longer requires the use of in-source `#[rustc_mir]` attributes.
Blocks rust-lang/rust-playpen#154 (if MIR dump support from the playpen is even wanted).
Example output:
```rust
fn main() {
let x = Some(0);
x.unwrap_or_else(|| 1);
}
```
```
MIR for expr || 1 (id=16)
fn(arg0: [closure@test.rs:3:22: 3:26]) -> i32 {
let mut tmp0: ();
bb0: {
return = const 1;
goto -> bb1;
}
bb1: {
return;
}
}
MIR for fn main::main (id=4)
fn() -> () {
let var0: core::option::Option<i32>; // x
let mut tmp0: ();
let mut tmp1: i32;
let mut tmp2: core::option::Option<i32>;
let mut tmp3: [closure@test.rs:3:22: 3:26];
bb0: {
var0 = core::option::Option::Some(const 0);
tmp2 = var0;
tmp3 = [closure@test.rs:3:22: 3:26];
tmp1 = core::option::Option<T>::unwrap_or_else(tmp2, tmp3) -> bb2;
}
bb1: {
return;
}
bb2: {
drop(tmp1) -> bb3;
}
bb3: {
return = ();
goto -> bb1;
}
}
```
Because we no longer use `GetFileAttributesExW` FileAttr is never created
directly from `WIN32_FILE_ATTRIBUTE_DATA` anymore.
So we should no longer store FileAttr's attributes in that c struct.
Needs a correct review because I'm not too confident with how this works.
All tests related to the C ABI are now passing.
References:
- dbe68fecd0/lib/CodeGen/TargetInfo.cpp (L479-L489)
- dbe68fecd0/lib/CodeGen/TargetInfo.cpp (L466-L477)
The `classifyArgumentType` function has two different paths depending on `RAA == CGCXXABI::RAA_DirectInMemory`, but I don't really know what's the corresponding option in Rust.
cc @brson @eddyb
This PR fixes two unrelated diagnostics bugs in resolve.
First, it reports privacy errors for an import only after the import resolution is determined, fixing #31402.
Second, it expands the per-module map from block ids to anonymous modules so that it also maps module declarations ids to modules, and it uses this map to in `with_scope` to fix#31644.
r? @nrc
This is because the tool compiler passes the name of the tool
as a command line `--cfg`. The improved session config parser
is stricter and no longer permits invalid meta items (such as
"error-index-generator").
Some tests just add the extra errors, others I fix by doing some simple error recovery. I've tried to avoid doing too much in the hope of doing something more principled later.
In general error messages are getting worse at this stage, but I think in the long run they will get better.
Now that we properly only link in jemalloc when building executables, we have
far less to worry about in terms of polluting the global namespace with the
`free` and `malloc` symbols on Linux. This commit will primarily allow LLVM to
use jemalloc so the compiler will only be using one allocator overall.
Locally this took compile time for libsyntax from 95 seconds to 89 (a 6%
improvement).
Back in 9bc8e6d14 the linking of rlibs changed to using the `link_whole_rlib`
function. This change, however was only intended to affect dylibs, not
executables. For executables we don't actually want to link entire rlibs because
we want the linker to strip out as much as possible.
This commit adds a conditional to this logic to only link entire rlibs if we're
creating a dylib, and otherwise an executable just links an rlib as usual. A
test is included which will fail to link if this behavior is reverted.
Right now the primary hashing algorithm of the compiler isn't actually inlined
across crates, meaning that it may be missing out on some crucial optimizations
in a few places (perhaps unrolling smaller loops, etc).
This commit made the hashing function disappear from a profiled version of the
compiler, but that's likely because it was just inlined elsewhere. When
compiling winapi, however, this decreased compile time from 18.3 to 17.8 seconds
(a 3% improvement).
