Ninja substantially improves LLVM build time. On a 96-way system, using
Make took 248s, and using Ninja took 161s, a 35% improvement.
We already require a variety of tools to build Rust. If someone wants to
build without Ninja (for instance, to minimize the set of packages
required to bootstrap a new target), they can easily set `ninja=false`
in `config.toml`. Our defaults should help people build Rust (and LLVM)
faster, to speed up development.
Improve defaults in x.py
- Make the default stage dependent on the subcommand
- Don't build stage1 rustc artifacts with x.py build --stage 1. If this is what you want, use x.py build --stage 2 instead, which gives you a working libstd.
- Change default debuginfo when debug = true from 2 to 1
I tried to fix CI to use `--stage 2` everywhere it currently has no stage, but I might have missed a spot.
This does not update much of the documentation - most of it is in https://github.com/rust-lang/rustc-dev-guide/ or https://github.com/rust-lang/rust-forge and will need a separate PR.
See individual commits for a detailed rationale of each change.
See also the MCP: https://github.com/rust-lang/compiler-team/issues/326
r? @Mark-Simulacrum , but anyone is free to give an opinion.
From [a conversation in discord](https://discordapp.com/channels/442252698964721669/443151243398086667/719200989269327882):
> Linking seems to consume all available RAM, leading to the OS to swap memory to disk and slowing down everything in the process
Compiling itself doesn't seem to take up as much RAM, and I'm only looking to check whether a minimal testcase can be compiled by rustc, where the runtime performance isn't much of an issue
> do you have debug = true or debuginfo-level = 2 in config.toml?
> if so I think that results in over 2GB of debuginfo nowadays and is likely the culprit
> which might mean we're giving out bad advice :(
Anecdotally, this sped up my stage 1 build from 15 to 10 minutes.
This still adds line numbers, it only removes variable and type information.
- Improve wording for debuginfo description
Co-authored-by: Teymour Aldridge <42674621+teymour-aldridge@users.noreply.github.com>
bootstrap: Configurable musl libdir
Make it possible to customize the location of musl libdir using
musl-libdir in config.toml, e.g., to use lib64 instead of lib.
If config.toml `profiler = false`, the test/mir-opt/instrument_coverage
test is ignored. Otherwise, this patch ensures the profiler_runtime is
loaded when -Zinstrument-coverage is enabled. Confirmed that this works
for MacOS.
Provide separate option for std debug asserts
On local one-off benchmarking of libcore metadata-only, debug asserts in std are a significant hit (15s to 20s). Provide an option for compiler developers to disable them. A build with a nightly compiler is around 10s, for reference.
This allows configuring the directory for build artifacts, instead of having it always be ./build. This means you can set it to a constant location, letting you reuse the same cache while working in several different directories.
The configuration lives in config.toml under build.build-dir. By default, it keeps the existing default of ./build, but it can be configured to any relative or absolute path. Additionally, it allows making outputs relative to the root of the git repository using $ROOT.
Make LLVM version suffix independent of rustc version on dev channel
Remove rustc version from LLVM version suffix on dev channel,
avoiding the need for full rebuilds when switching between
branches with different LLVM submodule & rustc version.
Note: To avoid full rebuild, on subsequent LLVM submodule update, copy the
current value of `LLVM_VERSION_SUFFIX` from `build/*/llvm/build/CMakeCache.txt`,
to `version-suffix` in `config.toml`.
Remove rustc version from LLVM version suffix on dev channel, avoiding
the need for full rebuilds when moving between commits with different
LLVM submodule & rustc version.
Enable Control Flow Guard in rustbuild
Now that Rust supports Control Flow Guard (#68180), add a config.toml option to build the standard library with CFG enabled.
r? @nagisa
allow rustfmt key in [build] section
Permit using `rustfmt` in `config.toml`. It will allow to not download `rustfmt` binary, which is not possible for at least some tiers-3 platforms.
Fixes: #67624
r? @Mark-Simulacrum
This commit builds on #65501 continue to simplify the build system and
compiler now that we no longer have multiple LLVM backends to ship by
default. Here this switches the compiler back to what it once was long
long ago, which is linking LLVM directly to the compiler rather than
dynamically loading it at runtime. The `codegen-backends` directory of
the sysroot no longer exists and all relevant support in the build
system is removed. Note that `rustc` still supports a dynamically loaded
codegen backend as it did previously, it just no longer supports
dynamically loaded codegen backends in its own sysroot.
Additionally as part of this the `librustc_codegen_llvm` crate now once
again explicitly depends on all of its crates instead of implicitly
loading them through the sysroot. This involved filling out its
`Cargo.toml` and deleting all the now-unnecessary `extern crate`
annotations in the header of the crate. (this in turn required adding a
number of imports for names of macros too).
The end results of this change are:
* Rustbuild's build process for the compiler as all the "oh don't forget
the codegen backend" checks can be easily removed.
* Building `rustc_codegen_llvm` is much simpler since it's simply
another compiler crate.
* Managing the dependencies of `rustc_codegen_llvm` is much simpler since
it's "just another `Cargo.toml` to edit"
* The build process should be a smidge faster because there's more
parallelism in the main rustc build step rather than splitting
`librustc_codegen_llvm` out to its own step.
* The compiler is expected to be slightly faster by default because the
codegen backend does not need to be dynamically loaded.
* Disabling LLVM as part of rustbuild is still supported, supporting
multiple codegen backends is still supported, and dynamic loading of a
codegen backend is still supported.
With #65251 landed there's no need to build two LLVM backends and ship
them with rustc, every target we have now uses the same LLVM backend!
This removes the `src/llvm-emscripten` submodule and additionally
removes all support from rustbuild for building the emscripten LLVM
backend. Multiple codegen backend support is left in place for now, and
this is intended to be an easy 10-15 minute win on CI times by avoiding
having to build LLVM twice.
- Compatible with Emscripten 1.38.46-upstream or later upstream.
- Refactors the Emscripten target spec to share code with other wasm
targets.
- Replaces the old incorrect wasm32 C call ABI with the correct one,
preserving the old one as wasm32_bindgen_compat for wasm-bindgen
compatibility.
- Updates the varargs ABI used by Emscripten and deletes the old one.
- Removes the obsolete wasm32-experimental-emscripten target.
- Uses EMCC_CFLAGS on CI to avoid the timeout problems with #63649.
- Refactors the Emscripten target spec to share code with other wasm
targets.
- Replaces the incorrect wasm32 C call ABI with the old asmjs
version, which is correct for both wasm32 and JS.
- Updates the varargs ABI used by Emscripten and deletes the old one.
- Removes the obsolete wasm32-experimental-emscripten target.
- Temporarily makes Emscripten targets use panic=abort by default
because supporting unwinding will require an LLVM patch.
This commit removes the `wasm_syscall` feature from the
wasm32-unknown-unknown build of the standard library. This feature was
originally intended to allow an opt-in way to interact with the
operating system in a posix-like way but it was never stabilized.
Nowadays with the advent of the `wasm32-wasi` target that should
entirely replace the intentions of the `wasm_syscall` feature.
This commit moves RISC-V from the experimental LLVM targets to the
regular LLVM targets. RISC-V was made non-experimental in
https://reviews.llvm.org/rL366399
I have also sorted the list of LLVM targets, and changed the code
around setting llvm_exp_targets (and its default) to match the code
setting llvm_targets (and its default), ensuring future changes to
the defaults, as LLVM targets become stable, affect as few places as
possible.
This renames wasm32-unknown-wasi to wasm32-wasi, omitting the vendor
component. This follows aarch64-linux-android, x86_64-fuchsia, and others in
omitting the vendor field, which has the advantage of aligning with the
[multiarch tuple](https://wiki.debian.org/Multiarch/Tuples), and of being
less noisy.
This commit adds a new wasm32-based target distributed through rustup,
supported in the standard library, and implemented in the compiler. The
`wasm32-unknown-wasi` target is intended to be a WebAssembly target
which matches the [WASI proposal recently announced.][LINK]. In summary
the WASI target is an effort to define a standard set of syscalls for
WebAssembly modules, allowing WebAssembly modules to not only be
portable across architectures but also be portable across environments
implementing this standard set of system calls.
The wasi target in libstd is still somewhat bare bones. This PR does not
fill out the filesystem, networking, threads, etc. Instead it only
provides the most basic of integration with the wasi syscalls, enabling
features like:
* `Instant::now` and `SystemTime::now` work
* `env::args` is hooked up
* `env::vars` will look up environment variables
* `println!` will print to standard out
* `process::{exit, abort}` should be hooked up appropriately
None of these APIs can work natively on the `wasm32-unknown-unknown`
target, but with the assumption of the WASI set of syscalls we're able
to provide implementations of these syscalls that engines can implement.
Currently the primary engine implementing wasi is [wasmtime], but more
will surely emerge!
In terms of future development of libstd, I think this is something
we'll probably want to discuss. The purpose of the WASI target is to
provide a standardized set of syscalls, but it's *also* to provide a
standard C sysroot for compiling C/C++ programs. This means it's
intended that functions like `read` and `write` are implemented for this
target with a relatively standard definition and implementation. It's
unclear, therefore, how we want to expose file descriptors and how we'll
want to implement system primitives. For example should `std::fs::File`
have a libc-based file descriptor underneath it? The raw wasi file
descriptor? We'll see! Currently these details are all intentionally
hidden and things we can change over time.
A `WasiFd` sample struct was added to the standard library as part of
this commit, but it's not currently used. It shows how all the wasi
syscalls could be ergonomically bound in Rust, and they offer a possible
implementation of primitives like `std::fs::File` if we bind wasi file
descriptors exactly.
Apart from the standard library, there's also the matter of how this
target is integrated with respect to its C standard library. The
reference sysroot, for example, provides managment of standard unix file
descriptors and also standard APIs like `open` (as opposed to the
relative `openat` inspiration for the wasi ssycalls). Currently the
standard library relies on the C sysroot symbols for operations such as
environment management, process exit, and `read`/`write` of stdio fds.
We want these operations in Rust to be interoperable with C if they're
used in the same process. Put another way, if Rust and C are linked into
the same WebAssembly binary they should work together, but that requires
that the same C standard library is used.
We also, however, want the `wasm32-unknown-wasi` target to be
usable-by-default with the Rust compiler without requiring a separate
toolchain to get downloaded and configured. With that in mind, there's
two modes of operation for the `wasm32-unknown-wasi` target:
1. By default the C standard library is statically provided inside of
`liblibc.rlib` distributed as part of the sysroot. This means that
you can `rustc foo.wasm --target wasm32-unknown-unknown` and you're
good to go, a fully workable wasi binary pops out. This is
incompatible with linking in C code, however, which may be compiled
against a different sysroot than the Rust code was previously
compiled against. In this mode the default of `rust-lld` is used to
link binaries.
2. For linking with C code, the `-C target-feature=-crt-static` flag
needs to be passed. This takes inspiration from the musl target for
this flag, but the idea is that you're no longer using the provided
static C runtime, but rather one will be provided externally. This
flag is intended to also get coupled with an external `clang`
compiler configured with its own sysroot. Therefore you'll typically
use this flag with `-C linker=/path/to/clang-script-wrapper`. Using
this mode the Rust code will continue to reference standard C
symbols, but the definition will be pulled in by the linker configured.
Alright so that's all the current state of this PR. I suspect we'll
definitely want to discuss this before landing of course! This PR is
coupled with libc changes as well which I'll be posting shortly.
[LINK]:
[wasmtime]:
look for python2 symlinks before bootstrap python
Before this commit, if you're running x.py directly on a system where
`python` is symlinked to Python 3, then the `python` config option will
default to a Python 3 interpreter. This causes debuginfo tests to fail
with an opaque error message, since they have a hard requirement on
Python 2.
This commit modifies the Python probe behavior to look for python2.7 and
python2 *before* using the interpreter used to execute `x.py`.
JSBackend is implied when building the emscripten backend, and not available for the standard llvm backend. This commit also puts the example config in sync with the defaults in src/bootstrap/native.rs
Before this commit, if you're running x.py directly on a system where
`python` is symlinked to Python 3, then the `python` config option will
default to a Python 3 interpreter. This causes debuginfo tests to fail
with an opaque error message, since they have a hard requirement on
Python 2.
This commit modifies the Python probe behavior to look for python2.7 and
python2 *before* using the interpreter used to execute `x.py`.
We'll use this as a temporary measure to get an LLVM update landed, but
we'll have to go through and update images later to make sure they've
got the right toolchains.
compiletest: Support opt-in Clang-based run-make tests and use them for testing xLTO.
Some cross-language run-make tests need a Clang compiler that matches the LLVM version of `rustc`. Since such a compiler usually isn't available these tests (marked with the `needs-matching-clang`
directive) are ignored by default.
For some CI jobs we do need these tests to run unconditionally though. In order to support this a `--force-clang-based-tests` flag is added to compiletest. If this flag is specified, `compiletest` will fail if it can't detect an appropriate version of Clang.
@rust-lang/infra The PR doesn't yet enable the tests yet. Do you have any recommendation for which jobs to enable them?
cc #57438
r? @alexcrichton
Provide the option to use libc++ even on all platforms
This is the default on platforms which use libc++ as the default C++
library but this option allows using libc++ on others as well.
The run-pass test suite currently takes 30 minutes on Windows, and
that appears to be roughly split between two 15 minute runs of the test
suite: one without NLL and one with NLL. In discussion on Discord the
platform coverage of the NLL compare mode may not necessarily be worth
it, so this commit removes the NLL compare mode from tests by default,
and then reenables it on only one builder.
This commit adds opt-in support to the compiler to link to `jemalloc` in
the compiler. When activated the compiler will depend on `jemalloc-sys`,
instruct jemalloc to unprefix its symbols, and then link to it. The
feature is activated by default on Linux/OSX compilers for x86_64/i686
platforms, and it's not enabled anywhere else for now. We may be able to
opt-in other platforms in the future! Also note that the opt-in only
happens on CI, it's otherwise unconditionally turned off by default.
Closes#36963
This commit removes all jemalloc related submodules, configuration, etc,
from the bootstrap, from the standard library, and from the compiler.
This will be followed up with a change to use jemalloc specifically as
part of rustc on blessed platforms.
In addition to to updating Cargo's submodule and Cargo's dependencies,
this also updates Cargo's build to build OpenSSL statically into Cargo
as well as libcurl unconditionally. This removes OpenSSL build logic
from the bootstrap code, and otherwise requests that even on OSX we
build curl statically.
This patch adds a few improvements to how the build system finds
LLVM's FileCheck program.
* On Fedora, the system LLVM installs FileCheck in the "llvm"
subdirectory of the LLVM libdir. This patch teaches the build
system to look there.
* This adds a configure option to specify which llvm-config executable
to use. This is handy on systems that can parallel install multiple
versions of LLVM; for example I can now:
./configure --llvm-config=/bin/llvm-config-5.0-64
... to build against LLVM 5, rather than whatever the default
llvm-config might be.
* Finally, this adds a configure- and config.toml- option to set the
path to FileCheck. This is handy when building against an LLVM
where FileCheck was not installed. This happens on compatibility
installs of LLVM on Fedora.
Add rustc SHA to released DWARF debuginfo
This commit updates the debuginfo that is encoded in all of our released
artifacts by default. Currently it has paths like `/checkout/src/...` but these
are a little inconsistent and have changed over time. This commit instead
attempts to actually define the file paths in our debuginfo to be consistent
between releases.
All debuginfo paths are now intended to be `/rustc/$sha` where `$sha` is the git
sha of the released compiler. Sub-paths are all paths into the git repo at that
`$sha`.
This commit updates the debuginfo that is encoded in all of our released
artifacts by default. Currently it has paths like `/checkout/src/...` but these
are a little inconsistent and have changed over time. This commit instead
attempts to actually define the file paths in our debuginfo to be consistent
between releases.
All debuginfo paths are now intended to be `/rustc/$sha` where `$sha` is the git
sha of the released compiler. Sub-paths are all paths into the git repo at that
`$sha`.
This optionally adds lldb (and clang, which it needs) to the build.
Because rust uses LLVM 7, and because clang 7 is not yet released, a
recent git master version of clang is used.
The lldb that is used includes the Rust plugin.
lldb is only built when asked for, or when doing a nightly build on
macOS. Only macOS is done for now due to difficulties with the Python
dependency.
Disable LLVM verification by default
Currently -Z no-verify only controls IR verification prior to LLVM codegen, while verification is performed unconditionally both before and after linking with (Thin)LTO.
Also wondering what the sentiment is on disabling verification by default (and e.g. only enabling it on ALT builds with assertions). This does not seem terribly useful outside of rustc development and it does seem to show up in profiles (at something like 3%).
**EDIT:** A table showing the various configurations and what is enabled when.
| Configuration | Dynamic verification performed | LLVM static assertions compiled in |
| --- | --- | --- |
| alt builds | | yes |
| nightly builds | | no |
| stable builds | | no |
| CI builds | | |
| dev builds in a checkout | | |
ship LLVM tools with the toolchain
this PR adds llvm-{nm,objcopy,objdump,size} to the rustc sysroot (right next to LLD)
this slightly increases the size of the rustc component. I measured these numbers on x86_64 Linux:
- rustc-1.27.0-dev-x86_64-unknown-linux-gnu.tar.gz 180M -> 193M (+7%)
- rustc-1.27.0-dev-x86_64-unknown-linux-gnu.tar.xz 129M -> 137M (+6%)
r? @alexcrichton
cc #49584
Currently on CI we predominately compile LLVM with the default system compiler
which means gcc on Linux, some version of Clang on OSX, MSVC on Windows, and
gcc on MinGW. This commit switches Linux, OSX, and Windows to all use Clang
6.0.0 to build LLVM (aka the C/C++ compiler as part of the bootstrap). This
looks to generate faster code according to #49879 which translates to a faster
rustc (as LLVM internally is faster)
The major changes here were to the containers that build Linux releases,
basically adding a new step that uses the previous gcc 4.8 compiler to compile
the next Clang 6.0.0 compiler. Otherwise the OSX and Windows scripts have been
updated to download precompiled versions of Clang 6 and configure the build to
use them.
Note that `cc` was updated here to fix using `clang-cl` with `cc-rs` on MSVC, as
well as an update to `sccache` on Windows which was needed to correctly work
with `clang-cl`. Finally the MinGW compiler is entirely left out here
intentionally as it's currently thought that Clang can't generate C++ code for
MinGW and we need to use gcc, but this should be verified eventually.
