- Use the same compiler for stage0 and stage1. This should be fixed at
some point (so bootstrap isn't constantly rebuilt).
- Make sure `x.py build` and `x.py check` work.
- Use `git merge-base` to determine the most recent commit to download.
- Copy stage0 to the various sysroots in `Sysroot`, and delegate to
Sysroot in Assemble. Leave all other code unchanged.
- Rename date -> key
This can also be a commit hash, so 'date' is no longer a good name.
- Add the commented-out option to config.toml.example
- Disable all steps by default when `download-rustc` is enabled
Most steps don't make sense when downloading a compiler, because they'll
be pre-built in the sysroot. Only enable the ones that might be useful,
in particular Rustdoc and all `check` steps.
At some point, this should probably enable other tools, but rustdoc is
enough to test out `download-rustc`.
- Don't print 'Skipping' twice in a row
Bootstrap forcibly enables a dry run if it isn't already set, so
previously it would print the message twice:
```
Skipping bootstrap::compile::Std because it is not enabled for `download-rustc`
Skipping bootstrap::compile::Std because it is not enabled for `download-rustc`
```
Now it correctly only prints once.
## Future work
- Add FIXME about supporting beta commits
- Debug logging will never work. This should be fixed.
Support non-stage0 check
Seems to work locally - a full stage 1 check succeeds, building std (because we can't get away with checking it), and then checking the compiler and other tools. This ran into the problem that a unconditional x.py check in stage 1 *both* checks and builds stage 1 std, and then has to clean up because for some reason the rmeta and rlib artifacts conflict (though I'm not actually entirely sure why, but it doesn't seem worth digging in in too much detail).
Ideally we wouldn't be building and checking like that but it's a minor worry as checking std is pretty fast and you can avoid it if you're aiming for speed by passing the compiler (e.g., compiler/rustc) explicitly.
r? ```@jyn514```
- Don't ignore packaging `llvm/lib/` for `rust-dev` when LLVM is linked
statically
- Add `link-type.txt` so bootstrap knows whether llvm was linked
statically or dynamically
- Don't assume CI LLVM is linked dynamically in `bootstrap::config`
- Fall back to dynamic linking if `link-type.txt` doesn't exist
- Fix existing bug that split the output of `llvm-config` on lines, not spaces
- Enable building LLVM tests
This works around the following llvm bug:
```
llvm-config: error: component libraries and shared library
llvm-config: error: missing: /home/joshua/rustc2/build/x86_64-unknown-linux-gnu/llvm/build/lib/libgtest.a
llvm-config: error: missing: /home/joshua/rustc2/build/x86_64-unknown-linux-gnu/llvm/build/lib/libgtest_main.a
llvm-config: error: missing: /home/joshua/rustc2/build/x86_64-unknown-linux-gnu/llvm/build/lib/libLLVMTestingSupport.a
thread 'main' panicked at 'command did not execute successfully: "/home/joshua/rustc2/build/x86_64-unknown-linux-gnu/llvm/build/bin/llvm-config" "--libfiles"
```
I'm not sure why llvm-config thinks these are required, but to avoid
the error, this builds them anyway.
- Temporarily set windows as the try builder. This should be reverted
before merging.
- Bump version of `download-ci-llvm-stamp`
`src/llvm-project` hasn't changed, but the generated tarball has.
- Only special case MacOS when dynamic linking. Static linking works fine.
- Store `link-type.txt` to the top-level of the tarball
This allows writing the link type unconditionally. Previously, bootstrap
had to keep track of whether the file IO *would* succeed (it would fail
if `lib/` didn't exist), which was prone to bugs.
- Make `link-type.txt` required
Anyone downloading this from CI should be using a version of bootstrap
that matches the version of the uploaded artifacts. So a missing
link-type indicates a bug in x.py.
Utilize PGO for rustc linux dist builds
This implements support for applying PGO to the rustc compilation step (not
standard library or any tooling, including rustdoc). Expanding PGO to more tools
is not terribly difficult but will involve more work and greater CI time
commitment.
For the same reason of avoiding greater implementation time commitment,
implementing for platforms outside of x86_64-unknown-linux-gnu is skipped.
In practice it should be quite simple to extend over time to more platforms. The
initial implementation is intentionally minimal here to avoid too much work
investment before we start seeing wins for a subset of Rust users.
The choice of workloads to profile here is somewhat arbitrary, but the general
rationale was to aim for a small set that largely avoided time regressions on
perf.rust-lang.org's full suite of crates. The set chosen is libcore, cargo (and
its dependencies), and a few ad-hoc stress tests from perf.rlo. The stress tests
are arguably the most controversial, but they benefit those cases (avoiding
regressions) and do not really remove wins from other benchmarks.
The primary next step after this PR lands is to implement support for PGO in
LLVM. It is unclear whether we can afford a full LLVM rebuild in CI, though, so
the approach taken there may need to be more staggered. rustc-only PGO seems
well affordable on linux at least, giving us up to 20% wall time wins on some
crates for 15 minutes of extra CI time (1 hour with this PR, up from 45 minutes).
The PGO data is uploaded to allow others to reuse it if attempting to reproduce
the CI build or potentially, in the future, on other platforms where an
off-by-one strategy is used for dist builds at minimal performance cost.
r? `@michaelwoerister` (but tell me if you don't want to / don't feel comfortable approving and we can find others)
This implements support for applying PGO to the rustc compilation step (not
standard library or any tooling, including rustdoc). Expanding PGO to more tools
is not terribly difficult but will involve more work and greater CI time
commitment.
For the same reason of avoiding greater time commitment, this currently avoids
implementing for platforms outside of x86_64-unknown-linux-gnu, though in
practice it should be quite simple to extend over time to more platforms. The
initial implementation is intentionally minimal here to avoid too much work
investment before we start seeing wins for a subset of Rust users.
The choice of workloads to profile here is somewhat arbitrary, but the general
rationale was to aim for a small set that largely avoided time regressions on
perf.rust-lang.org's full suite of crates. The set chosen is libcore, cargo (and
its dependencies), and a few ad-hoc stress tests from perf.rlo. The stress tests
are arguably the most controversial, but they benefit those cases (avoiding
regressions) and do not really remove wins from other benchmarks.
The primary next step after this PR lands is to implement support for PGO in
LLVM. It is unclear whether we can afford a full LLVM rebuild in CI, though, so
the approach taken there may need to be more staggered. rustc-only PGO seems
well affordable on linux at least, giving us up to 20% wall time wins on some
crates for 15 minutes of extra CI time (1 hour up from 45 minutes).
The PGO data is uploaded to allow others to reuse it if attempting to reproduce
the CI build or potentially, in the future, on other platforms where an
off-by-one strategy is used for dist builds at minimal performance cost.
`dsymutil` adds time to builds on Apple platforms for no clear benefit, and also
makes it more difficult for debuggers to find debug info. The compiler currently
defaults to running `dsymutil` to preserve its historical default, but when
compiling the compiler itself, we skip it by default since we know it's safe to
do so in that case.
x.py: allow a custom string appended to the version
This adds `rust.description` to the config as a descriptive string to be
appended to `rustc --version` output, which is also used in places like
debuginfo `DW_AT_producer`. This may be useful for supplementary build
information, like distro-specific package versions.
For example, in Fedora 33, `gcc --version` outputs:
gcc (GCC) 10.2.1 20201016 (Red Hat 10.2.1-6)
With this change, we can add similar vendor info to `rustc --version`.
This adds `rust.description` to the config as a descriptive string to be
appended to `rustc --version` output, which is also used in places like
debuginfo `DW_AT_producer`. This may be useful for supplementary build
information, like distro-specific package versions.
For example, in Fedora 33, `gcc --version` outputs:
gcc (GCC) 10.2.1 20201016 (Red Hat 10.2.1-6)
With this change, we can add similar vendor info to `rustc --version`.
Support enable/disable sanitizers/profiler per target
This PR add options under `[target.*]` of `config.toml` which can enable or disable sanitizers/profiler runtime for corresponding target.
If these options are empty, the global options under `[build]` will take effect.
Fix#78329
Prior to this, setting the rustfmt configuration was ignored:
```
% mkdir example
% cd example
% ../configure --set build.rustfmt=/usr/bin/true
% ../x.py fmt
./x.py fmt is not supported on this channel
failed to run: /Users/shep/Projects/rust/example/build/bootstrap/debug/bootstrap fmt
Build completed unsuccessfully in 0:00:01
```
And after:
```
% ../x.py fmt
Build completed successfully in 0:00:11
```
Don't warn if the config file is somewhere other than `config.toml`
Previously, `config.config` was always hardcoded as `"config.toml"`.
I thought that it was being overridden with the actual value later, but
it turns out `flags.config` was being completely discarded. This keeps
`config.config` in sync with `flags.config`.
Fixes https://github.com/rust-lang/rust/issues/77293
r? `@Mark-Simulacrum`
cc `@davidtwco`
This changes the behavior from *not* building for host whenever an
explicit target is specified. I find this much less confusing.
You can still disable host steps by passing an explicit empty list for
host.
Fixes#76990.
Previously, `config.config` was always hardcoded as `"config.toml"`.
I thought that it was being overridden with the actual value later, but
it turns out `flags.config` was being completely discarded. This keeps
`config.config` in sync with `flags.config`.
Add `x.py setup`
Closes#76503.
- Suggest `x.py setup` if config.toml doesn't exist yet
- Prompt for a profile if not given on the command line
- Print the configuration that will be used
- Print helpful starting commands after setup
- Link to the dev-guide after finishing
- Suggest `x.py setup` if config.toml doesn't exist yet (twice, once
before and once after the build)
- Prompt for a profile if not given on the command line
- Print the configuration file that will be used
- Print helpful starting commands after setup
- Link to the dev-guide after finishing
- Note that distro maintainers will see the changelog warning
This keeps only the `std` artifacts compiled by the given stage, not the
compiler. This is useful when working on the latter stages of the
compiler in tandem with the standard library, since you don't have to
rebuild the *entire* compiler when the standard library changes.
- Add a changelog and instructions for updating it
- Use `changelog-seen` in `config.toml` and `VERSION` in bootstrap to determine whether the changelog has been read
- Nag people if they haven't read the x.py changelog
+ Print message twice to make sure it's seen
- Give different error messages depending on whether the version needs to be updated or added
Add sample defaults for config.toml
- Allow including defaults in `src/bootstrap/defaults` using `profile = "..."`.
- Add default config files, with a README noting they're experimental and asking you to open an issue if you run into trouble. The config files have comments explaining why the defaults are set.
- Combine config files using the `merge` dependency.
This introduces a new dependency on `merge` that hasn't yet been vetted.
I want to improve the output when `include = "x"` isn't found:
```
thread 'main' panicked at 'fs::read_to_string(&file) failed with No such file or directory (os error 2) ("configuration file did not exist")', src/bootstrap/config.rs:522:28
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
failed to run: /home/joshua/rustc/build/bootstrap/debug/bootstrap test tidy
Build completed unsuccessfully in 0:00:00
```
However that seems like it could be fixed in a follow-up.
Closes#76619
- Allow including defaults in `src/bootstrap/defaults` using `profile = "..."`
- Add default config files
- Combine config files using the merge dependency.
- Add comments to default config files
- Add a README asking to open an issue if the defaults are bad
- Give a loud error if trying to merge `.target`, since it's not
currently supported
- Use an exhaustive match
- Use `<none>` in config.toml.example to avoid confusion
- Fix bugs in `Merge` derives
Previously, it would completely ignore the profile defaults if there
were any settings in `config.toml`. I sent an email to the `merge` maintainer
asking them to make the behavior in this commit the default.
This introduces a new dependency on `merge` that hasn't yet been vetted.
I want to improve the output when `include = "x"` isn't found:
```
thread 'main' panicked at 'fs::read_to_string(&file) failed with No such file or directory (os error 2) ("configuration file did not exist")', src/bootstrap/config.rs:522:28
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
failed to run: /home/joshua/rustc/build/bootstrap/debug/bootstrap test tidy
Build completed unsuccessfully in 0:00:00
```
However that seems like it could be fixed in a follow-up.
This requires that bootstrap is run from the same worktree as the sources it'll
build, but this is basically required for the build to work anyway. You can
still run it from a different directory, just that the files it builds must be
beside it.
This moves build triple discovery for rustbuild from bootstrap.py into a build
script, meaning it will "just work" if building rustbuild via Cargo rather than
Python.
The performance difference is negligible, but it makes me feel better.
Note that this does not remove some clones in `config`, because it would
require changing the logic around (and performance doesn't matter
for bootstrap).
Make the default stage for x.py configurable
This also allows configuring each sub-command individually.
Possibly #76617 should land before this? I don't feel strongly either way, I don't mind waiting.
Closes https://github.com/rust-lang/rust/issues/76165.
r? `@Mark-Simulacrum`
This allows configuring the default stage for each sub-command individually.
- Normalize the stage as early as possible, so there's no confusion
about which to use.
- Don't add an explicit `stage` option in config.toml
This offers no more flexibility than `*_stage` and makes it confusing
which takes precedence.
- Always give `--stage N` precedence over config.toml
- Fix bootstrap tests
This changes the tests to go through `Config::parse` so that they test
the actual defaults, not the dummy ones provided by `default_opts`. To
make this workable (and independent of the environment), it does not
read `config.toml` for tests.
Add a dedicated debug-logging option to config.toml
`@Mark-Simulacrum` and I were talking in zulip and we found that turning on debug/trace logging in rustc is fairly confusing, as it effectively depends on debug-assertions and is not documented as such. `@Mark-Simulacrum` mentioned that we should probably have a separate option for logging anyways.
this diff adds that, having the option follow debug-assertions (so everyone's existing config.toml should be fine) and if the option is false
to test I ran ./x.py test <something> twice, once with `debug-logging = false` and once with `debug-logging = true` and made sure i only saw trace's when it was true
This avoids missing a shared build when uplifting LLVM artifacts into the
sysroot. We were already producing a shared link anyway, though, so this is not
a visible change from the end user's perspective.
Previously, the CLI --target/--host definitions and configured options differed
in their effect: when setting these on the CLI, only the passed triples would be
compiled for, while in config.toml we would also compile for the build triple
and any host triples. This is needlessly confusing; users expect --target and
--host to be identical to editing the configuration file.
The new behavior is to respect --host and --target when passed as the *only*
configured triples (no triples are implicitly added). The default for --host is
the build triple, and the default for --target is the host triple(s), either
configured or the default build triple.