Fix sysroot option not being honored across rustc

[rcvalle]

Change link_sanitizer_runtime() to check if the sanitizer library exists in the specified/session sysroot, and if it doesn't exist, use the default sysroot. (See #79253.)

[rust-highfive]

r? @oli-obk

(rust-highfive has picked a reviewer for you, use r? to override)

[rcvalle]

r? @Mark-Simulacrum

[jyn514]

Is there a way to test this?

[rcvalle]

It can be tested it by moving the libraries that come with the Rust distribution (including the sanitizer libraries) to a different location, and compiling a program using the -Zsanitizer option (e.g., -Zsanitizer=address). Before the fix, the build fails trying to link the sanitizer libraries from the same location regardless whether sysroot was specified; after the fix, the build succeeds if sysroot is specified to the (sysroot equivalent relative to the) location of the previously moved libraries that come with the Rust distribution and sanitizer libraries.

[nagisa]

@bors r+ Thanks!

[bors]

📌 Commit 8c16ad5de78883c2600072ec8d74abace5dd9fe4 has been approved by nagisa

[nagisa]

(Approved this even without the test because it seems super convoluted to test for, unlikely to regress, and the change looks correct without much doubt anyway)

[bjorn3]

@bors r-

The reason this uses the default sysroot instead of --sysroot is probably that when using a sanitizer the sysroot passed using --sysroot is often build using xargo or cargo-xbuild to ensure that the sysroot is also build with the sanitizer enabled. Neither of which build the sanitizer runtime. Instead the sanitizer runtime from the default sysroot is used.

[nagisa]

Uh, using the default sysroot regardless of --sysroot specified sounds like a wrong way to solve that problem, to be honest. But I guess it can still be accommodated for by having rustc look in multiple locations in some order?

[tmiasko]

For the motivation behind the current implementation see #65241 (comment). Starting the search in the provided sysroot and then falling back to the default one sound fine to me.

[rust-log-analyzer]

The job mingw-check failed! Check out the build log: (web)(plain)

Click to see the possible cause of the failure (guessed by this bot)

configure: rust.channel := nightly
configure: rust.debug-assertions := True
configure: llvm.assertions := True
configure: dist.missing-tools := True
configure: build.configure-args := ['--enable-sccache', '--disable-manage-submodu ...
configure: writing `config.toml` in current directory
configure: configure: run `python /checkout/x.py --help`
configure: ---
skip untracked path cpu-usage.csv during rustfmt invocations
skip untracked path src/doc/book/ during rustfmt invocations
skip untracked path src/doc/rust-by-example/ during rustfmt invocations
skip untracked path src/llvm-project/ during rustfmt invocations
Diff in /checkout/compiler/rustc_codegen_ssa/src/back/link.rs at line 889:
fn link_sanitizer_runtime(sess: &Session, linker: &mut dyn Linker, name: &str) {
fn find_sanitizer_runtime(sess: &Session, filename: &String) -> PathBuf {
let session_tlib =
- filesearch::make_target_lib_path(&sess.sysroot,
- sess.opts.target_triple.triple());
+ filesearch::make_target_lib_path(&sess.sysroot, sess.opts.target_triple.triple());
let path = session_tlib.join(&filename);
if path.exists() {
return session_tlib;
Diff in /checkout/compiler/rustc_codegen_ssa/src/back/link.rs at line 897:
} else {
let default_sysroot = filesearch::get_or_default_sysroot();
- let default_tlib =
- filesearch::make_target_lib_path(&default_sysroot,
- sess.opts.target_triple.triple());
+ let default_tlib = filesearch::make_target_lib_path(
+ &default_sysroot,
+ sess.opts.target_triple.triple(),
+ );
return default_tlib;
}
}
Running `"/checkout/obj/build/x86_64-unknown-linux-gnu/stage0/bin/rustfmt" "--config-path" "/checkout" "--edition" "2018" "--unstable-features" "--skip-children" "--check" "/checkout/compiler/rustc_codegen_ssa/src/back/link.rs"` failed.
If you're running `tidy`, try again with `--bless`. Or, if you just want to format code, run `./x.py fmt` instead.
Build completed unsuccessfully in 0:00:14

[rcvalle]

I've amended my commit to check if the sanitizer library exists in the specified/session sysroot, and if it doesn't exist, use the default sysroot. What do you think?

[rcvalle]

@rustbot modify labels: +S-waiting-on-review -S-waiting-on-author

[nagisa]

@bors r+

Now that it falls back to the old implementation if the prerequisite libraries are not found in the proper sysroot, there should be no regression for xargo/build-std use-cases.

[bors]

📌 Commit 7378676 has been approved by nagisa

[bors]

⌛ Testing commit 7378676 with merge 0d7abce4a48a9ef8907159ca96c1be64e762342e...

[bors]

💥 Test timed out

[rust-log-analyzer]

A job failed! Check out the build log: (web)(plain)

Click to see the possible cause of the failure (guessed by this bot)

[jyn514]

@bors retry