modifed y.sh to allow for running cargo tests. #669
Conversation
build_system/src/test.rs
Outdated
| // That would force `cg_gcc` to *rebuild itself* and only then run tests, which is undesirable. | ||
| let mut env = test_env.clone(); | ||
| env.remove("RUSTFLAGS"); | ||
| // FIXME: maybe we could allow for only a select test to be run? |
There was a problem hiding this comment.
Yes, that would be nice to forward arguments to cargo test which would allow to filter the tests.
There was a problem hiding this comment.
Args can now be forwarded by using the -- symbol.
Eg.
./y.sh test --cargo-tests --use-system-gcc -- all flags after this get passed to cargo testThere was a problem hiding this comment.
Nice.
I believe only this is missing for this PR:
Could you please update the doc and the CI to use this new command instead of cargo test?
| @@ -203,6 +206,21 @@ fn clean(_env: &Env, args: &TestArg) -> Result<(), String> { | |||
| create_dir(&path) | |||
| } | |||
|
|
|||
| fn cargo_tests(test_env: &Env, test_args: &TestArg) -> Result<(), String> { | |||
There was a problem hiding this comment.
The following is not necessary before merging this PR, but would be a nice to have:
currently, this seems to always recompile rustc_codegen_gcc and the tests/lang_tests_*.rs stuff, even when there are no changes.
Do you know what causes this?
Would that be easy to fix?
Maybe that's because we have some environment variables set that affect the build, which makes it different than the normal build? (It could be worth just extract LD_LIBRARY_PATH and LIBRARY_PATH from test_env and only sending those to run_command_with_output_and_env to see if this is the case and fixes this issue).
There was a problem hiding this comment.
Sadly, that is unlikely to fix this, at least from my (rather amateur) analysis of the cargo logs.
(One of) the rebuild reasons is this:
fingerprint dirty for rustc_codegen_gcc v0.1.0
cargo::core::compiler::fingerprint: dirty: ProfileConfigurationChanged
Now, if it was a flag issue, I'd expect this to be the recompilation reason:
if self.rustflags != old.rustflags {
return DirtyReason::RustflagsChanged {
old: old.rustflags.clone(),
new: self.rustflags.clone(),
};
}I guess I will have to ask about what ProfileConfigurationChanged means - it is compared as a 64 bit hash in the cargo code, I will have to see what it is a hash of.
There was a problem hiding this comment.
BTW, all the other rebuild reasons in the logs only tell me that the tests are getting rebuilt, because cg_gcc got rebuilt.
target="lang_tests_common"}: cargo::core::compiler::fingerprint: dirty: FsStatusOutdated(StaleDependency { name: "rustc_codegen_gcc",
There was a problem hiding this comment.
I tested my idea and it does seem to help.
With it, running --cargo-tests a second time gives:
[BUILD] build system
Finished `release` profile [optimized] target(s) in 0.01s
GCC path retrieved from `config.toml`. Using `/opt/gcc/lib` as path for libgccjit
Finished `dev` profile [optimized + debuginfo] target(s) in 0.01s
[…]
Finished `test` profile [optimized + debuginfo] target(s) in 0.03s
Running unittests src/lib.rs (target/debug/deps/rustc_codegen_gcc-e4ba054637ab4819)
while without it:
[BUILD] build system
Finished `release` profile [optimized] target(s) in 0.03s
GCC path retrieved from `config.toml`. Using `/opt/gcc/lib` as path for libgccjit
Compiling rustc_codegen_gcc v0.1.0 (/home/bouanto/Ordinateur/Programmation/Rust/Projets/rustc_codegen_gcc)
Finished `dev` profile [optimized + debuginfo] target(s) in 1.37s
[…]
Compiling rustc_codegen_gcc v0.1.0 (/home/bouanto/Ordinateur/Programmation/Rust/Projets/rustc_codegen_gcc)
Finished `test` profile [optimized + debuginfo] target(s) in 8.51s
Running unittests src/lib.rs (target/debug/deps/rustc_codegen_gcc-4d02d698ebbfa187)
Added a new option to
./y.sh test:./y.sh test --cargo-tests.When this option is selected,
y.shwill build & test minicore(it is a dependency). After that, it will simply callcargo testwith the selected environment flags.For now
./y.sh test --cargo-testsignoresRUSTFLAGS.Normally,
RUSTFLAGScontain the-Zcodegen-backendoption, which would makecg_gccrebuild itself. This is undesirable, and ignoringRUSTFLAGSaltogether was simply the easiest option.In the future, we maybe could filter out
-Zcodegen-backend, and keep the rest of the flags intact.