The compiler-rt build system has been a never ending cause of pain for Rust
unfortunately:
* The build system is very difficult to invoke and configure to only build
compiler-rt, especially across platforms.
* The standard build system doesn't actually do what we want, not working for
some of our platforms and requiring a significant number of patches on our end
which are difficult to apply when updating compiler-rt.
* Compiling compiler-rt requires LLVM to be compiled, which... is a big
dependency! This also means that over time compiler-rt is not guaranteed to
build against older versions of LLVM (or newer versions), and we often want to
work with multiple versions of LLVM simultaneously.
The makefiles and rustbuild already know how to compile C code, the code here is
far from the *only* C code we're compiling. This patch jettisons all logic to
work with compiler-rt's build system and just goes straight to the source. We
just list all files manually (copied from compiler-rt's
lib/builtins/CMakeLists.txt) and compile them into an archive.
It's likely that this means we'll fail to pick up new files when we upgrade
compiler-rt, but that seems like a much less significant cost to pay than what
we're currently paying.
cc #34400, first steps towards that
llvm, rt: build using the Ninja generator if available
The Ninja generator generally builds much faster than make. It may also
be used on Windows to have a vast speed improvement over the Visual
Studio generators.
Currently hidden behind an `--enable-ninja` flag because it does not
obey the top-level `-j` or `-l` flags given to `make`.
If local-rust is the same as the current version, then force a local-rebuild
In Debian, we would like the option to build/rebuild the current release from
*either* the current or previous stable release. So we use enable-local-rust
instead of enable-local-rebuild, and read the bootstrap key dynamically from
whatever is installed locally.
In general, it does not make much sense to allow enable-local-rust without also
setting the bootstrap key, since the build would fail otherwise.
(The way I detect "the bootstrap key of [the local] rustc installation" is a bit hacky, suggestions welcome.)
Soon the LLVM upgrade (#34743) will require an updated CMake installation, and
the easiest way to do this was to upgrade the Ubuntu version of the bots to
16.04. This in turn brings in a new MIPS compiler on the linux-cross builder,
which is now from the "official" ubuntu repositories. Unfortunately these
new compilers don't support compiling with the `-msoft-float` flag like we're
currently passing, causing compiles to fail.
This commit removes these flags as it's not clear *why* they're being passed, as
the mipsel targets also don't have it. At least if it's not supported by a
debian default compiler, perhaps it's not too relevant to support?
The Ninja generator generally builds much faster than make. It may also
be used on Windows to have a vast speed improvement over the Visual
Studio generators.
Currently hidden behind an `--enable-ninja` flag because it does not
obey the top-level `-j` or `-l` flags given to `make`.
mk: Request -march=i686 on i686 Linux
Apparently the gcc on our dist bot is so old and/or obscure that the default
`-m32` switch doesn't think it can generate i686 code (or something like that).
The compiler-rt build system probes for the `__i686__` define in GCC to compile
for an i686 (vs i386) target, so this was failing on the bots.
This tweaks instead to pass `-march=i686` on i686-unknown-linux-gnu to C code to
ensure that we're compiling for i686 instead of i386. This should hopefully not
actually have an impact other than maybe doing some random optimization it
wasn't able to do so before. In theory this isn't making the target less
compatible as all Rust code is already compiled for i686.
Hopefully closes#34572
Apparently the gcc on our dist bot is so old and/or obscure that the default
`-m32` switch doesn't think it can generate i686 code (or something like that).
The compiler-rt build system probes for the `__i686__` define in GCC to compile
for an i686 (vs i386) target, so this was failing on the bots.
This tweaks instead to pass `-march=i686` on i686-unknown-linux-gnu to C code to
ensure that we're compiling for i686 instead of i386. This should hopefully not
actually have an impact other than maybe doing some random optimization it
wasn't able to do so before. In theory this isn't making the target less
compatible as all Rust code is already compiled for i686.
Hopefully closes#34572
Currently if an LLVM build is interrupted *after* it creates the llvm-config
binary but before it's done it puts us in an inconsistent state where we think
LLVM is compiled but it's not actually. This tweaks our logic to only consider
LLVM done building once it's actually done building.
This should hopefully alleviate problems on the bots where if we interrupt at
the wrong time it doesn't corrupt the build directory.
Try to fix the nightlies
They look to be failing right after the CMake PR landed. I've diagnosed and confirmed the first issue fixed, the second is a bit of a shot in the dark to see if it fixes things.
* Implement the clean-llvm target for those cases where makefiles are being used
* Have all cross-compiled LLVMs depend on the **host** LLVM as they'll require
the llvm-tablegen executable from there
This PR refactors the 'errors' part of libsyntax into its own crate (librustc_errors). This is the first part of a few refactorings to simplify error reporting and potentially support more output formats (like a standardized JSON output and possibly an --explain mode that can work with the user's code), though this PR stands on its own and doesn't assume further changes.
As part of separating out the errors crate, I have also refactored the code position portion of codemap into its own crate (libsyntax_pos). While it's helpful to have the common code positions in a separate crate for the new errors crate, this may also enable further simplifications in the future.
mk: Fix bootstrapping cross-hosts on beta
The beta builds are currently failing, unfortunately, due to what is presumably
some odd behavior with our makefiles. The wrong bootstrap key is being used to
generate the stage1 cross-compiled libraries, which fails the build.
Interestingly enough if the targets are directly specified as part of the build
then it works just fine! Just a bare `make` fails...
Instead of trying to understand what's happening in the makefiles instead just
tweak how we configure the bootstrap key in a way that's more likely to work.
The beta builds are currently failing, unfortunately, due to what is presumably
some odd behavior with our makefiles. The wrong bootstrap key is being used to
generate the stage1 cross-compiled libraries, which fails the build.
Interestingly enough if the targets are directly specified as part of the build
then it works just fine! Just a bare `make` fails...
Instead of trying to understand what's happening in the makefiles instead just
tweak how we configure the bootstrap key in a way that's more likely to work.
In Linux distributions, it is often necessary to rebuild packages for
cases like applying new patches or linking against new system libraries.
In this scenario, the rustc in the distro build environment may already
match the current release that we're trying to rebuild. Thus we don't
want to use the prior release's bootstrap key, nor `--cfg stage0` for
the prior unstable features.
The new `configure --enable-local-rebuild` option specifies that we are
rebuilding from the current release. The current bootstrap key is used
for the local rustc, and current stage1 features are also assumed.
This commit is an implementation of [RFC 1513] which allows applications to
alter the behavior of panics at compile time. A new compiler flag, `-C panic`,
is added and accepts the values `unwind` or `panic`, with the default being
`unwind`. This model affects how code is generated for the local crate, skipping
generation of landing pads with `-C panic=abort`.
[RFC 1513]: https://github.com/rust-lang/rfcs/blob/master/text/1513-less-unwinding.md
Panic implementations are then provided by crates tagged with
`#![panic_runtime]` and lazily required by crates with
`#![needs_panic_runtime]`. The panic strategy (`-C panic` value) of the panic
runtime must match the final product, and if the panic strategy is not `abort`
then the entire DAG must have the same panic strategy.
With the `-C panic=abort` strategy, users can expect a stable method to disable
generation of landing pads, improving optimization in niche scenarios,
decreasing compile time, and decreasing output binary size. With the `-C
panic=unwind` strategy users can expect the existing ability to isolate failure
in Rust code from the outside world.
Organizationally, this commit dismantles the `sys_common::unwind` module in
favor of some bits moving part of it to `libpanic_unwind` and the rest into the
`panicking` module in libstd. The custom panic runtime support is pretty similar
to the custom allocator support with the only major difference being how the
panic runtime is injected (takes the `-C panic` flag into account).
Add armv7-linux-androideabi target
This PR adds `armv7-linux-androideabi` target that matches `armeabi-v7a` Android ABI, ~~downscales `arm-linux-androideabi` target to match `armeabi` Android ABI~~ (TBD later if needed).
This should allow us to get the best performance from every [Android ABI level](http://developer.android.com/ndk/guides/abis.html).
Currently existing target `arm-linux-androideabi` started gaining features out of the supported range of [android `armeabi`](http://developer.android.com/ndk/guides/abis.html). While android compiler does not use a different target for later supported `armv7` architecture, it has distinct ABI name `armeabi-v7a`. We decided to add rust target `armv7-linux-androideabi` to match it.
Note that `NEON`, `VFPv3-D32`, and `ThumbEE` instruction sets are not added, because not all android devices are guaranteed to support all or some of these, and [their availability should be checked at runtime](http://developer.android.com/ndk/guides/abis.html#v7a).
~~This reduces performance of existing `arm-linux-androideabi` and may make it _much_ slower (we are talking more than order of magnitude in some random ad-hoc fp benchmark that I did).~~
Part of #33278.
make dist: specify the archive file as stdout
If the `-f` option isn't given, GNU tar will use environment variable
`TAPE` first, and next use the compiled-in default, which isn't
necessary `stdout` (it is the tape device `/dev/rst0` under OpenBSD for
example).
Implement constant support in MIR.
All of the intended features in `trans::consts` are now supported by `mir::constant`.
The implementation is considered a temporary measure until `miri` replaces it.
A `-Z orbit` bootstrap build will only translate LLVM IR from AST for `#[rustc_no_mir]` functions.
Furthermore, almost all checks of constant expressions have been moved to MIR.
In non-`const` functions, trees of temporaries are promoted, as per RFC 1414 (rvalue promotion).
Promotion before MIR borrowck would allow reasoning about promoted values' lifetimes.
The improved checking comes at the cost of four `[breaking-change]`s:
* repeat counts must contain a constant expression, e.g.:
`let arr = [0; { println!("foo"); 5 }];` used to be allowed (it behaved like `let arr = [0; 5];`)
* dereference of a reference to a `static` cannot be used in another `static`, e.g.:
`static X: [u8; 1] = [1]; static Y: u8 = (&X)[0];` was unintentionally allowed before
* the type of a `static` *must* be `Sync`, irrespective of the initializer, e.g.
`static FOO: *const T = &BAR;` worked as `&T` is `Sync`, but it shouldn't because `*const T` isn't
* a `static` cannot wrap `UnsafeCell` around a type that *may* need drop, e.g.
`static X: MakeSync<UnsafeCell<Option<String>>> = MakeSync(UnsafeCell::new(None));`
was previously allowed based on the fact `None` alone doesn't need drop, but in `UnsafeCell`
it can be later changed to `Some(String)` which *does* need dropping
The drop restrictions are relaxed by RFC 1440 (#33156), which is implemented, but feature-gated.
However, creating `UnsafeCell` from constants is unstable, so users can just enable the feature gate.
mk: Fix building with --enable-ccache
We will no longer use `ccache` in the makefiles for our local dependencies like
miniz, but they're so small anyway it doesn't really matter.
Closes#33285
Move auxiliary directories to live with the tests
This is a step for enabling testing of cross-crate incremental compilation. The idea is that instead of having a central auxiliary directory, when you have a `// aux-build:foo.rs` annotation in the test `run-pass/bar.rs`, it will look in (e.g.) `run-pass/aux/foo.rs`. In general, it looks for an `aux` directory in the same directory as the test. We also ignore the `aux` directories when enumerating the set of tests.
As part of this PR, also refactor `runtest.rs` to use methods on a context, which means we can stop passing around context everywhere.
r? @alexcrichton
Looks like the real bug on nightlies is that the `llvm-pass` run-make test is
not actually getting the value of `LLVM_CXXFLAGS` correct. Namely, it's blank!
Now the only change #33093 which actually affected this is that the argument
`$(LLVM_CXXFLAGS_$(2))` was moved up from a makefile rule into the definition of
a variable. Sounds innocuous?
Turns out the variable this was moved into is defined with `:=`, which means
that it's not recursively expanded, which basically means that it's expanded
immediately. Unfortunately part of this expansion involves running
`llvm-config`, which doesn't exist at the start of distcheck build!
This didn't show up on the bots because they run `make` *then* `make check`, and
the first step builds llvm-config so the next time `make` is loaded everything
is available. The distcheck bots, however, run just a plain `distcheck` so
`make` doesn't exist ahead of time. You can see this in action where the
distcheck bots start out with a bunch of "llvm-config not found" error messages.
This commit just changes a few variables to be defined with `=` which
essentially means they're lazily expanded. I did not run a full distcheck
locally, but this makes the initial "llvm-config not found" error messages go
away so I suspect that this is the fix.
Closes#33379
We will no longer use `ccache` in the makefiles for our local dependencies like
miniz, but they're so small anyway it doesn't really matter.
Closes#33285
If the `-f` option isn't given, GNU tar will use environment variable
`TAPE` first, and next use the compiled-in default, which isn't
necessary `stdout` (it is the tape device `/dev/rst0` under OpenBSD for
example).
This changes the CFLAGS and related variables passed to compiletest to be passed
for the target, not the host, so we can correctly test 32-bit cross compiles on
64-bit host machines.
Hopefuly fixes#33379
test: Move run-make tests into compiletest
Forcing them to be embedded in makefiles precludes being able to run them in
rustbuild, and adding them to compiletest gives us a great way to leverage
future enhancements to our "all encompassing test suite runner" as well as just
moving more things into Rust.
All tests are still Makefile-based in the sense that they rely on `make` being
available to run them, but there's no longer any Makefile-trickery to run them
and rustbuild can now run them out of the box as well.
Forcing them to be embedded in makefiles precludes being able to run them in
rustbuild, and adding them to compiletest gives us a great way to leverage
future enhancements to our "all encompassing test suite runner" as well as just
moving more things into Rust.
All tests are still Makefile-based in the sense that they rely on `make` being
available to run them, but there's no longer any Makefile-trickery to run them
and rustbuild can now run them out of the box as well.
The `--android-cross-path` has been deprecated for some time now, we should use
`CFG_ARM_LINUX_ANDROIDEABI_NDK` instead.
Ideally this would use the right triple, but we're only testing ARM for now.
