This was originally copied over from Cargo and Cargo has since [been
updated][update] so let's pull in the fixes here too!
[update]: https://github.com/rust-lang/cargo/pull/5030
This commit updates our Fat LTO logic to tweak our custom wrapper around LLVM's
"link modules" functionality. Previously whenever the
`LLVMRustLinkInExternalBitcode` function was called it would call LLVM's
`Linker::linkModules` wrapper. Internally this would crate an instance of a
`Linker` which internally creates an instance of an `IRMover`. Unfortunately for
us the creation of `IRMover` is somewhat O(n) with the input module. This means
that every time we linked a module it was O(n) with respect to the entire module
we had built up!
Now the modules we build up during LTO are quite large, so this quickly started
creating an O(n^2) problem for us! Discovered in #48025 it turns out this has
always been a problem and we just haven't noticed it. It became particularly
worse recently though due to most libraries having 16x more object files than
they previously did (1 -> 16).
This commit fixes this performance issue by preserving the `Linker` instance
across all links into the main LLVM module. This means we only create one
`IRMover` and allows LTO to progress much speedier.
From the `cargo-cache` project in #48025 a **full build** locally when from
5m15s to 2m24s. Looking at the timing logs each object file was linked in in
single-digit millisecond rather than hundreds, clearly being a nice improvement!
Closes#48025
This commit adds the ability for rustc to not run `dsymutil` by default
on OSX. A new codegen option, `-Z run-dsymutil=no`, was added to specify
that `dsymutil` should *not* run and instead the compiler should
unconditionally keep the object files around in a compilation if
necessary for debug information.
cc #47240
When encountering a variadic argument in a function definition that
doesn't accept it, if immediately after there's a closing paren,
continue parsing as normal. Otherwise keep current behavior of emitting
error and stopping.
This isn't a perfect heuristic, but since the amount of run-fail tests
is far lower than run-pass tests for now, it should be sufficient to
ensure that we don't run into CI limits. This makes it possible to run
the test binary manually (e.g., under gdb/lldb) if it failed to attempt
to find out why.
Previously, a Step would be able to tell on its own when it was invoked
"by-default" (that is, `./x.py test` was called instead of `./x.py test
some/path`). This commit replaces that functionality, invoking each Step
with each of the paths it has specified as "should be invoked by."
For example, if a step calls `path("src/tools/cargo")` and
`path("src/doc/cargo")` then it's make_run will be called twice, with
"src/tools/cargo" and "src/doc/cargo." This makes it so that default
handling logic is in builder, instead of spread across various Steps.
However, this meant that some Step specifications needed to be updated,
since for example `rustdoc` can be built by `./x.py build
src/librustdoc` or `./x.py build src/tools/rustdoc`. A `PathSet`
abstraction is added that handles this: now, each Step can not only list
`path(...)` but also `paths(&[a, b, ...])` which will make it so that we
don't invoke it with each of the individual paths, instead invoking it
with the first path in the list (though this shouldn't be depended on).
Future work likely consists of implementing a better/easier way for a
given Step to work with "any" crate in-tree, especially those that want
to run tests, build, or check crates in the std, test, or rustc crate
trees. Currently this is rather painful to do as most of the logic is
duplicated across should_run and make_run. It seems likely this can be
abstracted away into builder somehow.
Fix oversized loads on x86_64 SysV FFI calls
The x86_64 SysV ABI should use exact sizes for small structs passed in
registers, i.e. a struct that occupies 3 bytes should use an i24,
instead of the i32 it currently uses.
Refs #45543
Emit data::Impl in save-analysis
As discussed on [internals.rust-lang](https://internals.rust-lang.org/t/rustdoc2-rls-analysis-and-the-compiler-help-wanted/6592/5), this PR emits `rls-data::Impl` in the save-analysis.
A number of questions are outstanding:
- [x] A few `???` around row 356. We need to discuss what goes here, if anything.
- [ ] ~~Deriving `id` for impl using hashing. Is this going to clash with rustc defids?~~
- [ ] ~~Deriving `id` for impl using hashing. Is the conversion from 64 bit -> 32 bit problematic?~~
- [x] Need a new rls-data with an `id` field in `Impl` struct.
- [ ] ~~Need a new rls-data which `derive` `Hash` for `ImplKind` enum.~~
rustc_mir: insert a dummy access to places being matched on, when building MIR.
Fixes#47412 by adding a `_dummy = Discriminant(place)` before each `match place {...}`.
r? @nikomatsakis
Document the behaviour of infinite iterators on potentially-computable methods
It’s not entirely clear from the current documentation what behaviour
calling a method such as `min` on an infinite iterator like `RangeFrom`
is. One might expect this to terminate, but in fact, for infinite
iterators, `min` is always nonterminating (at least in the standard
library). This adds a quick note about this behaviour for clarification.
Explain unusual debugging code in librustc
Introduced in #47828 to help track down some bugs, it landed a bit hastily so
this is intended on cleaning it up a bit.
Update the dlmalloc submodule
A bug was recently fixed in dlmalloc which meant that released memory to the
system accidentally wasn't getting reused, causing programs to be far slower
than they should be!
