Distinguish fn item types to allow reification from nothing to fn pointers.
The first commit is a rebase of #26284, except for files that have moved since.
This is a [breaking-change], due to:
* each FFI function has a distinct type, like all other functions currently do
* all generic parameters on functions are recorded in their item types, e.g.:
`size_of::<u8>` & `size_of::<i8>`'s types differ despite their identical signature.
* function items are zero-sized, which will stop transmutes from working on them
The first two cases are handled in most cases with the new coerce-unify logic,
which will combine incompatible function item types into function pointers,
at the outer-most level of if-else chains, match arms and array literals.
The last case is specially handled during type-checking such that transmutes
from a function item type to a pointer or integer type will continue to work for
another release cycle, but are being linted against. To get rid of warnings and
ensure your code will continue to compile, cast to a pointer before transmuting.
Optimize some functions in std::process
* Be sure that `read_to_end` gets directed towards `read_to_end_uninitialized` for all handles
* When spawning a child that guaranteed doesn't need a stdin, don't actually create a stdin pipe for that process, instead just redirect it to /dev/null
* When calling `wait_with_output`, don't spawn threads to read out the pipes of the child. Instead drain all pipes on the calling thread and *then* wait on the process.
Functionally, it is intended that nothing changes as part of this PR
---
Note that this was the same as #31613, and even after that it turned out that fixing Windows was easier than I thought! To copy a comment from over there:
> As some rationale for this as well, it's always bothered me that we've spawned threads in the standard library for this (seems a bit overkill), and I've also been curious lately as to our why our build times for Windows are so much higher than Unix (on the buildbots we have). I have done basically 0 investigation into why, but I figured it can't help to try to optimize Command::output which I believe is called quite a few times during the test suite.
collections: Fix broken doc example
PR #32135 was accidentally merged without tests passing, and unfortunately one
of the tests added was broken, so this fixes that.
Semantically there's actually no reason for us to spawn threads as part of the
call to `wait_with_output`, and that's generally an incredibly heavyweight
operation for just reading a few bytes (especially when stderr probably rarely
has bytes!). An equivalent operation in terms of what's implemented today would
be to just drain both pipes of all contents and then call `wait` on the child
process itself.
On Unix we can implement this through some convenient use of the `select`
function, whereas on Windows we can make use of overlapped I/O. Note that on
Windows this requires us to use named pipes instead of anonymous pipes, but
they're semantically the same under the hood.
Fix incorrect trait privacy error
This PR fixes#21670 by using the crate metadata instead of `ExternalExports` to determine if an external item is public.
r? @nikomatsakis
There's a lot of stuff wrong with the representation of these types:
TyFnDef doesn't actually uniquely identify a function, TyFnPtr is used to
represent method calls, TyFnDef in the sub-expression of a cast isn't
correctly reified, and probably some other stuff I haven't discovered yet.
Splitting them seems like the right first step, though.
Make errors for unnecessary visibility qualifiers consistent
This PR refactors away `syntax::parse::parser::ParsePub` so that unnecessary visibility qualifiers on variant fields are reported not by the parser but by `privacy::SanePrivacyVisitor` (thanks to @petrochenkov's drive-by improvements in #31919).
r? @nikomatsakis
tests: add test for empty <>
Rust allows to specify an empty list of type and lifetime parameters, but there are no tests for it:
```
user@UNIT-326 [12:53:45] [~/projects/rust] [diamonds-and-rust]
-> % grep "<>" -R src/test
src/test/compile-fail/generic-type-params-name-repr.rs: // And don't print <> at all when there's just defaults.
src/test/debuginfo/issue22656.rs:// when trying to handle a Vec<> or anything else that contains zero-sized
```
So let's add them! Besides it's such a wonderful opportunity to put a reference to Judas Priest band into the branch name ;)
For example if `Command::output` or `Command::status` is used then stdin is just
immediately closed. Add an option for this so as an optimization we can avoid
creating pipes entirely.
This should help reduce the number of active file descriptors when spawning
processes on Unix and the number of active handles on Windows.
This pushes the implementation detail of proxying `read_to_end` through to
`read_to_end_uninitialized` all the way down to the `FileDesc` and `Handle`
implementations on Unix/Windows. This way intermediate layers will also be able
to take advantage of this optimized implementation.
This commit also adds the optimized implementation for `ChildStdout` and
`ChildStderr`.
This defines `_mm256_broadcast_ps` and `_mm256_broadcast_pd`. The `_ss`
and `_sd` variants are not supported by LLVM. In Clang these intrinsics
are implemented as inline functions in C++.
Intel reference: https://software.intel.com/en-us/node/514144.
Note: the argument type should really be "0hPc" (a pointer to a vector
of half the width), but internally the LLVM intrinsic takes a pointer to
a signed integer, and for any other type LLVM will complain. This means
that a transmute is required to call these intrinsics.
The AVX2 broadcast intrinsics `_mm256_broadcastss_ps` and
`_mm256_broadcastsd_pd` are not available as LLVM intrinsics. In Clang
they are implemented using the shufflevector builtin.
