Rollup merge of #145310 - Kobzol:compiler-for-revamp, r=jieyouxu

Reduce usage of `compiler_for` in bootstrap

While working on refactoring/fixing `dist` steps, I realized that `build.full-bootstrap` does much more than it should, and that it its documentation is wrong. It seems that the main purpose of this option should be to enable/disable stdlib/compiler uplifting (https://rust-lang.zulipchat.com/#narrow/channel/326414-t-infra.2Fbootstrap/topic/Purpose.20of.20.60build.2Efull-bootstrap.60/with/533985624), but currently it also affects staging, or more precisely which compiler will be used to build selected steps, because this option is used in the cursed `compiler_for` function.

I would like to change the option it so that it *only* affects uplifting, and doesn't affect stage selection, which I (partially) did in this PR. I removed the usage of `compiler_for` from the `Std` and `Rustc` steps, and explicitly implemented uplifting, without going through `compiler_for`.

The only remaining usages of `compiler_for` are in dist steps (which I'm currently refactoring, will send a PR later) and test steps (which I will take a look at after dist). After that we can finally remove the function.

I tried to document the case when uplifting was happening during cross-compilation, which was very implicit before. I also did a slight change in the uplifting logic for rustc when cross-compiling. Before, we would attempt to uplift a stage1 rustc, but that is not really a thing when cross-compiling.

r? `@jieyouxu`
This commit is contained in:
Jakub Beránek 2025-08-15 16:03:56 +02:00 committed by GitHub
commit 87d677b7d0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 85 additions and 92 deletions

View file

@ -345,9 +345,9 @@
# want to use vendoring. See https://forge.rust-lang.org/infra/other-installation-methods.html#source-code.
#build.vendor = if "is a tarball source" && "vendor" dir exists && ".cargo/config.toml" file exists { true } else { false }
# Typically the build system will build the Rust compiler twice. The second
# compiler, however, will simply use its own libraries to link against. If you
# would rather to perform a full bootstrap, compiling the compiler three times,
# If you build the compiler more than twice (stage3+) or the standard library more than once
# (stage 2+), the third compiler and second library will get uplifted from stage2 and stage1,
# respectively. If you would like to disable this uplifting, and rather perform a full bootstrap,
# then you can set this option to true.
#
# This is only useful for verifying that rustc generates reproducible builds.

View file

@ -37,11 +37,12 @@ use crate::{
debug, trace,
};
/// Build a standard library for the given `target` using the given `compiler`.
/// Build a standard library for the given `target` using the given `build_compiler`.
#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct Std {
pub target: TargetSelection,
pub compiler: Compiler,
/// Compiler that builds the standard library.
pub build_compiler: Compiler,
/// Whether to build only a subset of crates in the standard library.
///
/// This shouldn't be used from other steps; see the comment on [`Rustc`].
@ -54,10 +55,10 @@ pub struct Std {
}
impl Std {
pub fn new(compiler: Compiler, target: TargetSelection) -> Self {
pub fn new(build_compiler: Compiler, target: TargetSelection) -> Self {
Self {
target,
compiler,
build_compiler,
crates: Default::default(),
force_recompile: false,
extra_rust_args: &[],
@ -120,7 +121,7 @@ impl Step for Std {
trace!(force_recompile);
run.builder.ensure(Std {
compiler: run.builder.compiler(run.builder.top_stage, run.build_triple()),
build_compiler: run.builder.compiler(run.builder.top_stage, run.build_triple()),
target: run.target,
crates,
force_recompile,
@ -138,8 +139,8 @@ impl Step for Std {
let target = self.target;
// We already have std ready to be used for stage 0.
if self.compiler.stage == 0 {
let compiler = self.compiler;
if self.build_compiler.stage == 0 {
let compiler = self.build_compiler;
builder.ensure(StdLink::from_std(self, compiler));
return;
@ -148,9 +149,10 @@ impl Step for Std {
let build_compiler = if builder.download_rustc() && self.force_recompile {
// When there are changes in the library tree with CI-rustc, we want to build
// the stageN library and that requires using stageN-1 compiler.
builder.compiler(self.compiler.stage.saturating_sub(1), builder.config.host_target)
builder
.compiler(self.build_compiler.stage.saturating_sub(1), builder.config.host_target)
} else {
self.compiler
self.build_compiler
};
// When using `download-rustc`, we already have artifacts for the host available. Don't
@ -187,51 +189,50 @@ impl Step for Std {
let mut target_deps = builder.ensure(StartupObjects { compiler: build_compiler, target });
let compiler_to_use =
builder.compiler_for(build_compiler.stage, build_compiler.host, target);
trace!(?compiler_to_use);
// Stage of the stdlib that we're building
let stage = build_compiler.stage;
if compiler_to_use != build_compiler
// Never uplift std unless we have compiled stage 1; if stage 1 is compiled,
// uplift it from there.
//
// FIXME: improve `fn compiler_for` to avoid adding stage condition here.
&& build_compiler.stage > 1
// If we're building a stage2+ libstd, full bootstrap is
// disabled and we have a stage1 libstd already compiled for the given target,
// then simply uplift a previously built stage1 library.
if build_compiler.stage > 1
&& !builder.config.full_bootstrap
// This estimates if a stage1 libstd exists for the given target. If we're not
// cross-compiling, it should definitely exist by the time we're building a stage2
// libstd.
// Or if we are cross-compiling, and we are building a cross-compiled rustc, then that
// rustc needs to link to a cross-compiled libstd, so again we should have a stage1
// libstd for the given target prepared.
// Even if we guess wrong in the cross-compiled case, the worst that should happen is
// that we build a fresh stage1 libstd below, and then we immediately uplift it, so we
// don't pay the libstd build cost twice.
&& (target == builder.host_target || builder.config.hosts.contains(&target))
{
trace!(
?compiler_to_use,
?build_compiler,
"build_compiler != compiler_to_use, uplifting library"
);
let build_compiler_for_std_to_uplift = builder.compiler(1, builder.host_target);
builder.std(build_compiler_for_std_to_uplift, target);
builder.std(compiler_to_use, target);
let msg = if compiler_to_use.host == target {
let msg = if build_compiler_for_std_to_uplift.host == target {
format!(
"Uplifting library (stage{} -> stage{})",
compiler_to_use.stage, build_compiler.stage
"Uplifting library (stage{} -> stage{stage})",
build_compiler_for_std_to_uplift.stage
)
} else {
format!(
"Uplifting library (stage{}:{} -> stage{}:{})",
compiler_to_use.stage, compiler_to_use.host, build_compiler.stage, target
"Uplifting library (stage{}:{} -> stage{stage}:{target})",
build_compiler_for_std_to_uplift.stage, build_compiler_for_std_to_uplift.host,
)
};
builder.info(&msg);
// Even if we're not building std this stage, the new sysroot must
// still contain the third party objects needed by various targets.
self.copy_extra_objects(builder, &build_compiler, target);
builder.ensure(StdLink::from_std(self, compiler_to_use));
builder.ensure(StdLink::from_std(self, build_compiler_for_std_to_uplift));
return;
}
trace!(
?compiler_to_use,
?build_compiler,
"compiler == compiler_to_use, handling not-cross-compile scenario"
);
target_deps.extend(self.copy_extra_objects(builder, &build_compiler, target));
// We build a sysroot for mir-opt tests using the same trick that Miri does: A check build
@ -299,7 +300,7 @@ impl Step for Std {
}
fn metadata(&self) -> Option<StepMetadata> {
Some(StepMetadata::build("std", self.target).built_by(self.compiler))
Some(StepMetadata::build("std", self.target).built_by(self.build_compiler))
}
}
@ -665,6 +666,14 @@ pub fn std_cargo(builder: &Builder<'_>, target: TargetSelection, cargo: &mut Car
cargo.rustdocflag("-Zcrate-attr=warn(rust_2018_idioms)");
}
/// Link all libstd rlibs/dylibs into a sysroot of `target_compiler`.
///
/// Links those artifacts generated by `compiler` to the `stage` compiler's
/// sysroot for the specified `host` and `target`.
///
/// Note that this assumes that `compiler` has already generated the libstd
/// libraries for `target`, and this method will find them in the relevant
/// output directory.
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct StdLink {
pub compiler: Compiler,
@ -680,7 +689,7 @@ impl StdLink {
pub fn from_std(std: Std, host_compiler: Compiler) -> Self {
Self {
compiler: host_compiler,
target_compiler: std.compiler,
target_compiler: std.build_compiler,
target: std.target,
crates: std.crates,
force_recompile: std.force_recompile,
@ -951,14 +960,8 @@ impl Rustc {
}
impl Step for Rustc {
/// We return the stage of the "actual" compiler (not the uplifted one).
///
/// By "actual" we refer to the uplifting logic where we may not compile the requested stage;
/// instead, we uplift it from the previous stages. Which can lead to bootstrap failures in
/// specific situations where we request stage X from other steps. However we may end up
/// uplifting it from stage Y, causing the other stage to fail when attempting to link with
/// stage X which was never actually built.
type Output = u32;
type Output = ();
const IS_HOST: bool = true;
const DEFAULT: bool = false;
@ -997,7 +1000,7 @@ impl Step for Rustc {
/// This will build the compiler for a particular stage of the build using
/// the `build_compiler` targeting the `target` architecture. The artifacts
/// created will also be linked into the sysroot directory.
fn run(self, builder: &Builder<'_>) -> u32 {
fn run(self, builder: &Builder<'_>) {
let build_compiler = self.build_compiler;
let target = self.target;
@ -1013,7 +1016,7 @@ impl Step for Rustc {
&sysroot,
builder.config.ci_rustc_dev_contents(),
);
return build_compiler.stage;
return;
}
// Build a standard library for `target` using the `build_compiler`.
@ -1027,31 +1030,33 @@ impl Step for Rustc {
builder.info("WARNING: Use `--keep-stage-std` if you want to rebuild the compiler when it changes");
builder.ensure(RustcLink::from_rustc(self, build_compiler));
return build_compiler.stage;
return;
}
let compiler_to_use =
builder.compiler_for(build_compiler.stage, build_compiler.host, target);
if compiler_to_use != build_compiler {
builder.ensure(Rustc::new(compiler_to_use, target));
let msg = if compiler_to_use.host == target {
format!(
"Uplifting rustc (stage{} -> stage{})",
compiler_to_use.stage,
build_compiler.stage + 1
)
// The stage of the compiler that we're building
let stage = build_compiler.stage + 1;
// If we are building a stage3+ compiler, and full bootstrap is disabled, and we have a
// previous rustc available, we will uplift a compiler from a previous stage.
if build_compiler.stage >= 2
&& !builder.config.full_bootstrap
&& (target == builder.host_target || builder.hosts.contains(&target))
{
// If we're cross-compiling, the earliest rustc that we could have is stage 2.
// If we're not cross-compiling, then we should have rustc stage 1.
let stage_to_uplift = if target == builder.host_target { 1 } else { 2 };
let rustc_to_uplift = builder.compiler(stage_to_uplift, target);
let msg = if rustc_to_uplift.host == target {
format!("Uplifting rustc (stage{} -> stage{stage})", rustc_to_uplift.stage,)
} else {
format!(
"Uplifting rustc (stage{}:{} -> stage{}:{})",
compiler_to_use.stage,
compiler_to_use.host,
build_compiler.stage + 1,
target
"Uplifting rustc (stage{}:{} -> stage{stage}:{target})",
rustc_to_uplift.stage, rustc_to_uplift.host,
)
};
builder.info(&msg);
builder.ensure(RustcLink::from_rustc(self, compiler_to_use));
return compiler_to_use.stage;
builder.ensure(RustcLink::from_rustc(self, rustc_to_uplift));
return;
}
// Build a standard library for the current host target using the `build_compiler`.
@ -1128,8 +1133,6 @@ impl Step for Rustc {
self,
builder.compiler(build_compiler.stage, builder.config.host_target),
));
build_compiler.stage
}
fn metadata(&self) -> Option<StepMetadata> {
@ -1909,12 +1912,18 @@ impl Step for Sysroot {
}
}
/// Prepare a compiler sysroot.
///
/// The sysroot may contain various things useful for running the compiler, like linkers and
/// linker wrappers (LLD, LLVM bitcode linker, etc.).
///
/// This will assemble a compiler in `build/$target/stage$stage`.
#[derive(Debug, PartialOrd, Ord, Clone, PartialEq, Eq, Hash)]
pub struct Assemble {
/// The compiler which we will produce in this step. Assemble itself will
/// take care of ensuring that the necessary prerequisites to do so exist,
/// that is, this target can be a stage2 compiler and Assemble will build
/// previous stages for you.
/// that is, this can be e.g. a stage2 compiler and Assemble will build
/// the previous stages for you.
pub target_compiler: Compiler,
}
@ -1932,11 +1941,6 @@ impl Step for Assemble {
});
}
/// Prepare a new compiler from the artifacts in `stage`
///
/// This will assemble a compiler in `build/$host/stage$stage`. The compiler
/// must have been previously produced by the `stage - 1` builder.build
/// compiler.
fn run(self, builder: &Builder<'_>) -> Compiler {
let target_compiler = self.target_compiler;
@ -2065,7 +2069,7 @@ impl Step for Assemble {
target_compiler.stage - 1,
builder.config.host_target,
);
let mut build_compiler =
let build_compiler =
builder.compiler(target_compiler.stage - 1, builder.config.host_target);
// Build enzyme
@ -2089,24 +2093,13 @@ impl Step for Assemble {
}
// Build the libraries for this compiler to link to (i.e., the libraries
// it uses at runtime). NOTE: Crates the target compiler compiles don't
// link to these. (FIXME: Is that correct? It seems to be correct most
// of the time but I think we do link to these for stage2/bin compilers
// when not performing a full bootstrap).
// it uses at runtime).
debug!(
?build_compiler,
"target_compiler.host" = ?target_compiler.host,
"building compiler libraries to link to"
);
let actual_stage = builder.ensure(Rustc::new(build_compiler, target_compiler.host));
// Current build_compiler.stage might be uplifted instead of being built; so update it
// to not fail while linking the artifacts.
debug!(
"(old) build_compiler.stage" = build_compiler.stage,
"(adjusted) build_compiler.stage" = actual_stage,
"temporarily adjusting `build_compiler.stage` to account for uplifted libraries"
);
build_compiler.stage = actual_stage;
builder.ensure(Rustc::new(build_compiler, target_compiler.host));
let stage = target_compiler.stage;
let host = target_compiler.host;