Bootstrap regression in 1.71.0 #114338
Assignees
Labels
C-bug
Category: This is a bug.
regression-from-stable-to-stable
Performance or correctness regression from one stable version to another.
T-bootstrap
Relevant to the bootstrap subteam: Rust's build system (x.py and src/bootstrap)
Comments
lovesegfault
added
C-bug
rustbot
added
needs-triage
|
This is a stripped down version of the build script in use, which was unchanged between 1.70.0 and 1.71.0: #!/bin/bash
# Fail when commands exit unsuccesfully.
set -o errexit
# Fail when using an undefined variable.
set -o nounset
# Fail if commands fail as part of a pipeline.
set -o pipefail
echo "Discovering build and source directory..."
RUST_CONFIGURE_ARGS=(
# https://github.com/rust-lang/rust/blob/934624fe5f66ce3fb8abf0597a6deb079783335f/src/ci/run.sh#L50
--disable-manage-submodules
--enable-locked-deps
--enable-cargo-native-static
--set rust.codegen-units-std=1
--set rust.remap-debuginfo
--debuginfo-level-std=1
# don't spend time _also_ building .gz files
--dist-compression-formats=xz
# avoid taking a runtime dependency on libstdc++
--enable-llvm-static-stdcpp
# disallow nightly features
--release-channel=stable
# don't access the internet to get dependencies
--enable-vendor
--local-rust-root "$BOOTSTRAP_INSTALL_DIR"
# we're building with the same version of the compiler, so speed up a bit
--enable-local-rebuild
# https://github.com/rust-lang/rust/blob/673d0db5e393e9c64897005b470bfeb6d5aec61b/src/ci/docker/host-x86_64/dist-x86_64-linux/Dockerfile#L94
# note that full-tools implies --extended
--enable-full-tools
--enable-sanitizers
--enable-profiler
# don't spend time generating docs, since those won't be patched
--disable-docs
--set llvm.thin-lto=true
--set rust.jemalloc
# limit the number of concurrent link jobs to avoid build host OOMs
--set llvm.link-jobs=3
# install to build/
--prefix "$INSTALL_DIR"
# and stick configuration under there too (note the relative path)
--sysconfdir "etc"
# we also need to make sure `--remap-path-prefix` is passed to rustc and
# llvm builds to avoid the absolute builder path from ending up in the
# final artifact. If we don't do this, re-builds of the exact same Rust
# version generates different binaries entirely unnecessarily (i.e., when
# there are no source-level differences). We also do it so that the paths
# in the debug symbols in our binaries match those in the upstream release
# artifacts, namely /rustc/<git commit of release>, which other tooling
# might depend on.
--set rust.remap-debuginfo=true
--tools "cargo,clippy,rustfmt,llvm-tools,src,rustdoc,rust-analyzer-proc-macro-srv"
--build "$targetTriple"
--target "$targetTriple"
)
./configure "${RUST_CONFIGURE_ARGS[@]}"
echo "Patching out unneeded build targets..."
sed -i.bkp \
-e '/dist::RustDev,/d' \
-e '/dist::Extended,/d' \
src/bootstrap/builder.rs
echo "Make build scripts respect LDFLAGS during bootstrap..."
mkdir -p .cargo
cat > .cargo/config <<EOF
target-applies-to-host = false
[host]
rustflags=["-Clink-args=$LDFLAGS"]
# include the vendor source replacement that x.py would normally generate with
# ./configure --enable-vendor:
[source.crates-io]
replace-with = 'vendored-sources'
registry = 'https://example.com'
[source.vendored-sources]
directory = '$PWD/vendor'
EOF
echo "Running rust build..."
python x.py dist
echo "Installing build artifacts..."
python x.py install
echo "Exporting rustup artifacts..."
ls -l build/dist/
mkdir -p "$BUILD_DIR"/rustup
cp -r build/dist/* "$BUILD_DIR"/rustup |
What is bug? We are not blindly copying the entire sysroot. I'm not sure if I understand the problem. Are we supposed to find these binaries in tarballs too, or is it something else? |
oli-obk
added
regression-from-stable-to-stable
|
Did some investigating. Just running Now running with |
apiraino
removed
the
I-prioritize
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
If I look at the bins that are part of
rustc's tarball in 1.70.0 I see this:But when I built Rust again in 1.71.0, and inspected the produced artifacts, I see:
The difference being that
cargo-fmt,rustfmt,cargo-clippy, andclippy-driverare now included.I tracked this change down to this commit: 6d99d6a, which was a part of #110365 by @ozkanonur.
I validated my claim by applying this patch to Rust's source:
And upon re-building, inspecting the tarball again:
I spoke with @jyn514, who said:
One thing that isn't clear to me is why do the officially produced artifacts served through
static.rust-lang.orgnot suffer from this:The text was updated successfully, but these errors were encountered: