The definition of this struct changes in LLVM 12 due to the addition
of branch coverage support. To avoid future mismatches, declare our
own struct and then convert between them.
Whether for Rust's own `target_os`, LLVM's triples, or GNU config's, the
OS-related have fields have been for code running *on* that OS, not code
that is *part* of the OS.
The difference is huge, as syscall interfaces are nothing like
freestanding interfaces. Kernels are (hypervisors and other more exotic
situations aside) freestanding programs that use the interfaces provided
by the hardware. It's *those* interfaces, the ones external to the
program being built and its software dependencies, that are the content
of the target.
For the Linux Kernel in particular, `target_env: "gnu"` is removed for
the same reason: that `-gnu` refers to glibc or GNU/linux, neither of
which applies to the kernel itself.
Relates to #74247
Thanks @ojeda for catching some things.
Apply lint restrictions from renamed lints
Previously, if you denied the old name of a renamed lint, it would warn
about using the new name, but otherwise do nothing. Now, it will behave
the same as if you'd used the new name.
Fixes https://github.com/rust-lang/rust/issues/82615.
r? `@ehuss`
make x86_64-pc-solaris the default target for x86-64 Solaris
This change makes `x86_64-pc-solaris` the default compilation target for x86-64 Solaris/Illumos (based on [this exchange](https://github.com/rust-lang/rust/issues/68214#issuecomment-748042054) with `@varkor).`
I tried several ways of doing this (leveraging the alias support added with #61761 and improved/fixed with #80073) and found out that cross-compilation to the new one is by far the simplest way of doing this. It can be achieved by adding the following arguments: `--build x86_64-sun-solaris --host x86_64-pc-solaris --target x86_64-pc-solaris` and enabling the cross compilation with `PKG_CONFIG_ALLOW_CROSS=1` environment variable.
I also removed alias support altogether - `x86_64-pc-solaris` and `x86_64-sun-solaris` are now two separate targets. The problem with aliases is that even if rust internally knows that two are the same, other tools building with rust don't know that, resulting in build issues like the one with firefox mentioned [here](https://github.com/rust-lang/rust/issues/68214#issuecomment-746144229). I think that once the dust settles and `x86_64-pc-solaris` becomes the default, `x86_64-sun-solaris` can be removed.
If you agree with the above, I have two subsequent questions:
1. Is there a preferred way to display deprecation warnings when `x86_64-sun-solaris` is passed into the compiler as an argument? I am not sure whether target deprecation was done before.
2. Where would be the best way to document this change for those using rust on Solaris? Without the cross-compilation arguments (used once to build a new version), the build won't work. Should I add it into [RELEASES.md](https://github.com/rust-lang/rust/blob/master/RELEASES.md)?
Thanks!
Remove storage markers if they won't be used during code generation
The storage markers constitute a substantial portion of all MIR
statements. At the same time, for builds without any optimizations,
the storage markers have no further use during and after MIR
optimization phase.
If storage markers are not necessary for code generation, remove them.
Fixed support for macOS Catalyst on ARM64
When I initially added Arm64 Catalyst support in https://github.com/rust-lang/rust/pull/77484 I had access to a DTK. However, while waiting to merge the PR some other changes were merged which caused conflicts in the branch. When fixing those conflicts I had no access to the DTK anymore and didn't try out if the resulting binaries did indeed work on Apple Silicon. I finally have a M1 and I realized that some small changes were necessary to support Apple Silicon. This PR adds the required changes. I've been running binaries generated with this branch for some time now and they work without issues.
Apparently #35870 caused a problem in this code (which originally
returned an impl trait) and `#[inline]` was added as a workaround, in
ade79d7609.
The issue is now fixed and the comment and `#[inline]` can now be
removed.
The first argument to an x86-interrupt ABI function was implicitly
treated as byval prior to LLVM 12. Since LLVM 12, it has to be
marked as such explicitly: 2e0e03c6a0
Previously, if you denied the old name of a renamed lint, it would warn
about using the new name, but otherwise do nothing. Now, it will behave
the same as if you'd used the new name.
Remove the x86_64-rumprun-netbsd target
Herein we remove the target from the compiler and the code from libstd intended to support the now-defunct rumprun project.
Closes#81514
The storage markers constitute a substantial portion of all MIR
statements. At the same time, for builds without any optimizations,
the storage markers have no further use during and after MIR
optimization phase.
If storage markers are not necessary for code generation, remove them.
Update measureme dependency to the latest version
This version adds the ability to use `rdpmc` hardware-based performance
counters instead of wall-clock time for measuring duration. This also
introduces a dependency on the `perf-event-open-sys` crate on Linux
which is used when using hardware counters.
r? ```@oli-obk```
Link crtbegin/crtend on musl to terminate .eh_frame
For some targets, rustc uses a "CRT fallback", where it links CRT
object files it ships instead of letting the host compiler link
them.
On musl, rustc currently links crt1, crti and crtn (provided by
libc), but does not link crtbegin and crtend (provided by libgcc).
In particular, crtend is responsible for terminating the .eh_frame
section. Lack of terminator may result in segfaults during
unwinding, as reported in #47551 and encountered by the LLVM 12
update in #81451.
This patch links crtbegin and crtend for musl as well, following
the table at the top of crt_objects.rs.
r? ``@nagisa``
Suggest character encoding is incorrect when encountering random null bytes
This adds a note whenever null bytes are seen at the start of a token unexpectedly, since those tend to come from UTF-16 encoded files without a [BOM](https://en.wikipedia.org/wiki/Byte_order_mark) (if a UTF-16 BOM appears it won't be valid UTF-8, but if there is no BOM it be both valid UTF-16 and valid but garbled UTF-8). This approach was suggested in https://github.com/rust-lang/rust/issues/73979#issuecomment-653976451.
Closes#73979.
Skip Ty w/o infer ty/const in trait select
Remove some allocations & also add `skip_current_subtree` to skip subtrees with no inferred items.
r? `@eddyb` since marked in the FIXME
This adds recovery when in array type syntax user writes
[X; Y<Z, ...>]
instead of
[X; Y::<Z, ...>]
Fixes#82566
Note that whenever we parse an expression and know that the next token
cannot be `,`, we should be calling
check_mistyped_turbofish_with_multiple_type_params for this recovery.
Previously we only did this for statement parsing (e.g. `let x = f<a,
b>;`). We now also do it when parsing the length field in array type
syntax.
check_mistyped_turbofish_with_multiple_type_params was previously
expecting type arguments between angle brackets, which is not right, as
we can also see const expressions. We now use generic argument parser
instead of type parser.
Test with one, two, and three generic arguments added to check
consistentcy between
1. check_no_chained_comparison: Called after parsing a nested binop
application like `x < A > ...` where angle brackets are interpreted as
binary operators and `A` is an expression.
2. check_mistyped_turbofish_with_multiple_type_params: called by
`parse_full_stmt` when we expect to see a semicolon after parsing an
expression but don't see it.
(In `T2<1, 2>::C;`, the expression is `T2 < 1`)
