`TokenStream` used to be a complex type, but it is now just a newtype
around a `Lrc<Vec<TreeAndJoint>>`. Currently it uses custom encoding
that discards the `IsJoint` and custom decoding that adds `NonJoint`
back in for every token tree. This requires building intermediate
`Vec<TokenTree>`s.
This commit makes `TokenStream` derive `Rustc{En,De}codable`. This
simplifies the code, and avoids the creation of the intermediate
vectors, saving up to 3% on various benchmarks. It also changes the AST
JSON output in one test.
Remove unreachable unit tuple compare binop codegen
Closes#63906Fixes#63551
This is based on #63906 by @luigishat, who had trouble with rebasing his PR.
Avoid unnecessary arena allocations in `expand_pattern()`.
`expand_pattern()` has two callsites. One of them needs arena
allocation, but the other does not.
This commit moves the arena allocation out of the function. This avoids
the allocation of many 4 KiB page arena chunks that only hold a single
small allocation. It reduces the number of bytes allocated by up to 2%
for various benchmarks, albeit without only a very small improvement in
runtime.
This attribute causes DOT files to be generated in the top-level
directory. It is intended to be used only temporarily and should never
appear on master.
This will prevent #65071 from occurring again.
Remove unneeded `ref` from docs
Will reduce confusion like in https://users.rust-lang.org/t/help-understanding-the-ref-t-syntax/33779 since match ergonomics means you (almost) never have to say `ref` anymore!
There might be more like this, but I don't have a checkout on my computer right this second and I'm on slow wifi and GitHub search isn't powerful enough and that's my story.
add test for calling non-const fn
The good news is that there is an error. But I expected to see [this error](9578272d68/src/librustc_mir/const_eval.rs (L346)) surface. @oli-obk any idea why that message is not shown anywhere?
r? @oli-obk
Avoid realloc in `CString::new`
If `&[u8]` or `&str` is given, `CString::new` allocates a new `Vec` with the exact capacity, and then `CString::from_vec_unchecked` calls `.reserve_exact(1)` for nul byte. This PR avoids the reallocation by allocationg `len + 1` bytes beforehand. In microbenchmark this PR speeds up `CString::new(&[u8])` by 30%.
Inline `ptr::null(_mut)` even in debug builds
I think we should treat `ptr::null(_mut)` as a constant. As It may help reduce code size
in debug build.
See godbolt link: https://godbolt.org/z/b9YMtD
Avoid injecting sanitizer runtimes into staticlibs (#64629).
This fixes the remaining issue in `creader.rs` and also fixes the expected test failure. I have explicitly turned the `$(CC)` call into a negative check with the `!` to ensure that this command is really failing (if it is not, then either the runtime is attached to the lib or the lib has not been instrumented and both would be an error).
I've also borrowed `program.rs` and the additional `rustc` invocation from @tmiasko 's PR since he pointed out that using `-fsanitize=address` with `$(CC)` for linking could fail if the sanitizer runtimes on the system are incompatible.
With this toolchain I was able to compile Firefox locally without any linker errors. I am still seeing races with Rust in TSan but I assume that is because I did not build with `-Z build-std`.
Upgrade GCC to 8.3.0, glibc to 1.17.0 and crosstool-ng to 1.24.0 for dist-armv7-linux
#62896 was caused by the usage of the GCC 5.2.0 toolchain, which was released back in 2015 and may have bugs affecting LLVM 9.
This PR upgrade GCC to 8.3.0 from 5.2.0, glibc from 1.16.0 to 1.17.0 and crosstool-ng to 1.24.0 only for dist-armv7-linux.
Fixes#62896
r? @alexcrichton
Prefer statx on linux if available
This PR make `metadata`-related functions try to invoke `statx` first on Linux if available,
making `std::fs::Metadata::created` work on Linux with `statx` supported.
It follows the discussion in #61386 , and will fix#59743
The implementation of this PR is simply converting `struct statx` into `struct stat64` with
extra fields for `btime` if `statx` succeeds, since other fields are not currently used.
---
I also did a separated benchmark for `fs::metadata`, `stat64`, `statx`, and `statx` with conversion to `stat64`.
It shows that `statx` with conversion is even more faster than pure `statx`.
I think it's due to `sizeof stat64 == 114` but `sizeof statx == 256`.
Anyway, the bare implementation of `statx` with conversion is only about 0.2% slower than the original impl (`stat64`-family).
With heap-allocation counted (~8.5% of total cost), the difference between `stat` and `statx` (with or without conversion) is just nothing.
Therefore, I think it is not urgent to use bare `struct statx` as underlying representation now.
There is no need to break `std::os::linux::fs::MetadataExt::as_raw_stat` (https://github.com/rust-lang/rust/pull/61386#issuecomment-538074018)
[Separated bare benchmarks](https://gist.github.com/oxalica/c4073ecb202c599fe41b7f15f86dc79c):
```
metadata_ok time: [529.41 ns 529.77 ns 530.19 ns]
metadata_err time: [538.71 ns 539.39 ns 540.35 ns]
stat64_ok time: [484.32 ns 484.53 ns 484.75 ns]
stat64_err time: [481.77 ns 482.00 ns 482.24 ns]
statx_ok time: [488.07 ns 488.35 ns 488.62 ns]
statx_err time: [487.74 ns 488.00 ns 488.27 ns]
statx_cvt_ok time: [485.05 ns 485.28 ns 485.53 ns]
statx_cvt_err time: [485.23 ns 485.45 ns 485.67 ns]
```
r? @alexcrichton
Rc: value -> allocation
See https://github.com/rust-lang/rust/issues/64484. This does not yet edit `Arc` as I first wanted to be sure we agree on the terminology the way it actually ends up. "value" as a term appears a lot in this file, and sometimes it refers to the value stored inside the `RcBox` while sometimes it refers to the `RcBox` itself. I tried to properly tease these apart but may have made some mistakes. The former should now always be called "inner value" and the latter "allocation".
One area where I was very unsure of which terminology is dropping: the `value` field of the `RcBox` will get dropped *earlier* than the `RcBox` itself if there are weak references. I decided that "dropping the value stored in the allocation" refers to dropping the value field, while "destroying the allocation" refers to actually freeing its backing memory.
r? @Centril
BTreeSet symmetric_difference & union optimized
No scalability changes, but:
- Grew the cmp_opt function (shared by symmetric_difference & union) into a MergeIter, with less memory overhead than the pairs of Peekable iterators now, speeding up ~20% on my machine (not so clear on Travis though, I actually switched it off there because it wasn't consistent about identical code). Mainly meant to improve readability by sharing code, though it does end up using more lines of code. Extending and reusing the MergeIter in btree_map might be better, but I'm not sure that's possible or desirable. This MergeIter probably pretends to be more generic than it is, yet doesn't declare to be an iterator because there's no need to, it's only there to help construct genuine iterators SymmetricDifference & Union.
- Compact the code of #64820 by moving if/else into match guards.
r? @bluss
Use structured suggestion for restricting bounds
When a trait bound is not met and restricting a type parameter would
make the restriction hold, use a structured suggestion pointing at an
appropriate place (type param in param list or `where` clause).
Account for opaque parameters where instead of suggesting extending
the `where` clause, we suggest appending the new restriction:
`fn foo(impl Trait + UnmetTrait)`. Fix#64565, fix#41817, fix#24354,
cc #26026, cc #37808, cc #24159, fix#37138, fix#24354, cc #20671.
