Respect verify-llvm-ir option in the backend
We are currently unconditionally verifying the LLVM IR in the backend (twice), ignoring the value of the verify-llvm-ir option. This has substantial compile-time impact for debug builds.
Bump boostrap compiler to new beta
Currently failing due to something about the const stability checks and `panic!`. I'm not sure why though since I wasn't able to see any PRs merged in the past few days that would result in a `cfg(bootstrap)` that shouldn't be removed. cc `@RalfJung` #131349
Remove `HybridBitSet`
`HybridBitSet` was introduced under the name `HybridIdxSetBuf` way back in #53383 where it was a big win for NLL borrow checker performance. In #93984 the more flexible `ChunkedBitSet` was added. Uses of `HybridBitSet` have gradually disappeared (e.g. #116152) and there are now few enough that they can be replaced with `BitSet` or `ChunkedBitSet`, and `HybridBitSet` can be removed, cutting more than 700 lines of code.
r? `@Mark-Simulacrum`
ensure JSON-defined targets are consistent
We have a `check_consistency` check that ensures some invariants which (presumably) the rest of the compiler relies on. However, JSON targets can easily be written in a way that violates those invariants. So this PR applies the same consistency check to JSON targets that we already enforce for built-in targets.
I have converted many of the assertions in that function to new macros that show a nice error instead of a panic; if people are okay with the general approach here, I can do that for the rest of the checks as well.
Enable -Zshare-generics for inline(never) functions
This avoids inlining cross-crate generic items when possible that are
already marked inline(never), implying that the author is not intending
for the function to be inlined by callers. As such, having a local copy
may make it easier for LLVM to optimize but mostly just adds to binary
bloat and codegen time. In practice our benchmarks indicate this is
indeed a win for larger compilations, where the extra cost in dynamic
linking to these symbols is diminished compared to the advantages in
fewer copies that need optimizing in each binary.
It might also make sense it expand this with other heuristics (e.g.,
`#[cold]`) in the future, but this seems like a good starting point.
FWIW, I expect that doing cleanup in where we make the decision
what should/shouldn't be shared is also a good idea. Way too
much code needed to be tweaked to check this. But I'm hoping
to leave that for a follow-up PR rather than blocking this on it.
Compiletest: add proc-macro header
This adds a `proc-macro` header to simplify using proc-macros, and to reduce boilerplate. This header works similar to the `aux-build` header where you pass a path for a proc-macro to be built.
This allows the `force-host`, `no-prefer-dynamic` headers, and `crate_type` attribute to be removed. Additionally it uses `--extern` like `aux_crate` (allows implicit `extern crate` in 2018) and `--extern proc_macro` (to place in the prelude in 2018).
~~This also includes a secondary change which defaults the edition of proc-macros to 2024. This further reduces boilerplate (removing `extern crate proc_macro;`), and allows using modern Rust syntax. I was a little on the fence including this. I personally prefer it, but I can imagine it might be confusing to others.~~ EDIT: Removed
Some tests were changed so that when there is a chain of dependencies A→B→C, that the `@ proc-macro` is placed in `B` instead of `A` so that the `--extern` flag works correctly (previously it depended on `-L` to find `C`). I think this is better to make the dependencies more explicit. None of these tests looked like the were actually testing this behavior.
There is one test that had an unexplained output change: `tests/ui/macros/same-sequence-span.rs`. I do not know why it changed, but it didn't look like it was particularly important. Perhaps there was a normalization issue?
This is currently not compatible with the rustdoc `build-aux-docs` header. It can probably be fixed, I'm just not feeling motivated to do that right now.
### Implementation steps
- [x] Document this new behavior in rustc-dev-guide once we figure out the specifics. https://github.com/rust-lang/rustc-dev-guide/pull/2149
Rollup of 7 pull requests
Successful merges:
- #133358 (Don't type error if we fail to coerce `Pin<T>` because it doesnt contain a ref)
- #133422 (Fix clobber_abi in RV32E and RV64E inline assembly)
- #133452 (Support predicate registers (clobber-only) in Hexagon inline assembly)
- #133463 (Fix handling of x18 in AArch64 inline assembly on ohos/trusty or with -Zfixed-x18)
- #133487 (fix confusing diagnostic for reserved `##`)
- #133557 (Small doc fixes in `rustc_codegen_ssa`)
- #133560 (Trim extra space in 'repeated `mut`' diagnostic)
r? `@ghost`
`@rustbot` modify labels: rollup
Use edition of `macro_rules` when compiling the macro
This changes the edition assigned to a macro_rules macro when it is compiled to use the edition of where the macro came from instead of the local crate's edition.
This fixes a problem when a macro_rules macro is created by a proc-macro. Previously that macro would be tagged with the local edition, which would cause problems with using the correct edition behavior inside the macro. For example, the check for unsafe attributes would cause errors in 2024 when using proc-macros from older editions.
This is partially related to https://github.com/rust-lang/rust/issues/132906. Unfortunately this is only a half fix for that issue. It fixes the error that happens in 2024, but does not fix the lint firing in 2021. I'm still trying to think of some way to fix that, but I'm running low on ideas.
Reduce integer `Display` implementation size
I was thinking about #128204 and how we could reduce the size of the code and just realized that we didn't need the `_fmt` method to be implemented on signed integers, which in turns allow to simplify greatly the macro call.
r? `@workingjubilee`
rustc: Fail fast when compiling a source file larger than 4 GiB
Currently if you try to compile a file that is larger than 4 GiB, `rustc` will first read the whole into memory before failing.
If we can read the metadata of the file, we can fail before reading the file.
Add `AsyncFn*` to the prelude in all editions
The general vibe is that we will most likely stabilize the `feature(async_closure)` *without* the `async Fn()` trait bound modifier.
Without `async Fn()` bound syntax, this necessitates users to spell the bound like `AsyncFn()`. Since `core::ops::AsyncFn` is not in the prelude, users will need to import these any time they actually want to use the trait. This seems annoying, so let's add these traits to the prelude unstably.
We're trying to work on the general vision of `async` trait bound modifier in general in: https://github.com/rust-lang/rfcs/pull/3710, however that RFC still needs more time for consensus to converge, and we've decided that the value that users get from calling the bound `async Fn()` is *not really* worth blocking landing async closures in general.
btree: don't leak value if destructor of key panics
This PR fixes a regression from https://github.com/rust-lang/rust/pull/84904.
The `BTreeMap` already attempts to handle panicking destructors of the key-value pairs by continuing to execute the remaining destructors after one destructor panicked. However, after #84904 the destructor of a value in a key-value pair gets skipped if the destructor of the key panics, only continuing with the next key-value pair. This PR reverts to the behavior before #84904 to also drop the corresponding value if the destructor of a key panics.
This avoids potential memory leaks and can fix the soundness of programs that rely on the destructors being executed (even though this should not be relied upon, because the std collections currently do not guarantee that the remaining elements are dropped after a panic in a destructor).
cc `@Amanieu` because you had opinions on panicking destructors
Stabilize `Ipv6Addr::is_unique_local` and `Ipv6Addr::is_unicast_link_local`
Make `Ipv6Addr::is_unique_local` and `Ipv6Addr::is_unicast_link_local` stable (+const).
Newly stable API:
```rust
impl Ipv6Addr {
// Newly stable under `ipv6_is_unique_local`
const fn is_unique_local(&self) -> bool;
// Newly stable under `ipv6_is_unique_local`
const fn is_unicast_link_local(&self) -> bool;
}
```
These stabilise a subset of the following tracking issue:
- #27709
I have looked and could not find any issues with `is_unique_local` and `is_unicast_link_local`. There is a well received comment calling for stabilisation of the latter function.
Both functions are well defined and consistent with implementations in other languages:
- [Go](https://cs.opensource.google/go/go/+/refs/tags/go1.23.0:src/net/netip/netip.go;l=518)
- [Python](e9d1bf353c/Lib/ipaddress.py (L2319-L2321))
- [Ruby (unique local)](https://ruby-doc.org/stdlib-2.5.1/libdoc/ipaddr/rdoc/IPAddr.html#private-3F-source)
- [Ruby (unicast link local)](https://ruby-doc.org/stdlib-2.5.1/libdoc/ipaddr/rdoc/IPAddr.html#link_local-3F-source)
cc implementor `@little-dude`
(I can't find the original PR for `is_unqiue_local`)
r? libs-api
`@rustbot` label +T-libs-api +needs-fcp
Update cargo
5 commits in 69e595908e2c420e7f0d1be34e6c5b984c8cfb84..66221abdeca2002d318fde6efff516aab091df0e
2024-11-16 01:26:11 +0000 to 2024-11-19 21:30:02 +0000
- Docs for optional registry JSON fields (rust-lang/cargo#14839)
- Allow registries to omit empty/default fields in JSON (rust-lang/cargo#14838)
- docs(unstable): Link to -Zwarnings issue, tracking issue (rust-lang/cargo#14836)
- fix(): error context for git_fetch refspec not found (rust-lang/cargo#14806)
- you we distinction (rust-lang/cargo#14829)
Drop debug info instead of panicking if we exceed LLVM's capability to represent it
Recapping a bit of history here:
In #128861 I made debug info correctly represent parameters to inline functions by removing a fake lexical block that had been inserted to suppress LLVM assertions and by deduplicating those parameters.
LLVM, however, expects to see a single parameter _with distinct locations_, particularly distinct inlinedAt values on the DILocations. This generally worked because no matter how deep the chain of inlines it takes two different call sites in the original function to result in the same function being present multiple times, and a function call requires a non-zero number of characters, but macros threw a wrench in that in #131944. At the time I thought the issue there was limited to proc-macros, where an arbitrary amount of code can be generated at a single point in the source text.
In #132613 I added discriminators to DILocations that would otherwise be the same to repair #131944[^1]. This works, but LLVM's capacity for discriminators is not infinite (LLVM actually only allocates 12 bits for this internally). At the time I thought it would be very rare for anyone to hit the limit, but #132900 proved me wrong. In the relatively-minimized test case it also became clear to me that the issue affects regular macros too, because the call to the inlined function will (without collapse_debuginfo on the macro) be attributed to the (repeated, if the macro is used more than once) textual callsite in the macro definition.
This PR fixes the panic by dropping debug info when we exceed LLVM's maximum discriminator value. There's also a preceding commit for a related but distinct issue: macros that use collapse_debuginfo should in fact have their inlinedAts collapsed to the macro callsite and thus not need discriminators at all (and not panic/warn accordingly when the discriminator limit is exhausted).
Fixes#132900
r? `@jieyouxu`
[^1]: Editor's note: `fix` is a magic keyword in PR description that apparently will close the linked issue (it's closed already in this case, but still).
Default-enable `llvm_tools_enabled` when no `config.toml` is present
Fixes#133195. cc `@wesleywiser` could you double check if with this patch and no `config.toml` that you can run `./x test tests/ui --stage 1`?
`llvm-objcopy` is usually required by cg_ssa on macOS to workaround bad `strip`s.
cc `@bjorn3` I hope this doesn't break cg_clif...
r? bootstrap
Add `visit` methods to ast nodes that already have `walk`s on ast visitors
Some `walk` functions are called directly, because there were no correspondent visit functions.
related to #128974 & #127615
r? `@petrochenkov`
Add vec_deque::Iter::as_slices and friends
Add the following methods, that work similarly to VecDeque::as_slices:
- alloc::collections::vec_deque::Iter::as_slices
- alloc::collections::vec_deque::IterMut::into_slices
- alloc::collections::vec_deque::IterMut::as_slices
- alloc::collections::vec_deque::IterMut::as_mut_slices
Obtaining slices from a VecDeque iterator was not previously possible.
It was added in #123752 to handle some cases involving emoji, but it
isn't necessary because it's always treated the same as
`TokenKind::InvalidIdent`. This commit removes it, which makes things a
little simpler.
Improve VecCache under parallel frontend
This replaces the single Vec allocation with a series of progressively larger buckets. With the cfg for parallel enabled but with -Zthreads=1, this looks like a slight regression in i-count and cycle counts (~1%).
With the parallel frontend at -Zthreads=4, this is an improvement (-5% wall-time from 5.788 to 5.4688 on libcore) than our current Lock-based approach, likely due to reducing the bouncing of the cache line holding the lock. At -Zthreads=32 it's a huge improvement (-46%: 8.829 -> 4.7319 seconds).
try-job: i686-gnu-nopt
try-job: dist-x86_64-linux
