`cargo test` without `cargo build` won't create `target/debug/libfoo.rlib` while `cargo build` does

[LukasKalbertodt]

I always expect that cargo test basically runs a cargo build before doing anything test specific. But that's apparently not the case.

---

Minimal example:

Cargo.toml

[package]
name = "foo"
version = "0.1.0"
authors = ["me"]
edition = "2018"

src/lib.rs

pub fn foo() {}

tests/main.rs

use foo::foo;

#[test]
fn main() {
    foo();
}

Executing cargo test will compile the foo crate and execute tests. Fine. BUT: cargo clean && cargo test won't create the file target/debug/libfoo.rlib while cargo clean && cargo build does create that file. This is very unexpected.

Tests on rustc 1.31.0-nightly (5597ee8a6 2018-10-03)

[dwijnand]

I wonder: why is this a problem?

My thinking is, if cargo test correctly runs the tests, but in doing so doesn't create libfoo.rlib, isn't that kind of cargo's implementation details?

[LukasKalbertodt]

Mhhh maybe?

The reason I noticed this is because I am writing my own test runners which call rustc with an --extern parameter. There I pass the path to my library (the .rlib file). So for my test runners it is important that this file exists.

Of course, maybe I shouldn't use this file as it's implementation detail. That would be fine. So far I've been using the build-plan feature to get the correct path. But I still think it's strange that cargo test doesn't run a normal cargo build beforehand. I think it's surprising and not what the user would expect.

[matthiaskrgr]

Perhaps you need something like

[lib]
name = "foo"
path = "src/lib.rs"
test = true #  <-

[alexcrichton]

Interesting! I don't think it's necessarily problematic that we don't do this but I don't see a great reason to not create these artifacts. Seems good to me to fix!

[ehuss]

I think this is behaving as expected. This is a result of #5460. Because the lib is not an explicit target in this case, it is not uplifted out of deps. Previously @alexcrichton you said that you didn't think rlib should ever be uplifted, so changing that would be the opposite of the direction you previously indicated. I think in this case using the true path in deps is the correct thing to do.

[alexcrichton]

Heh I do flip flop a bit. If it can be made to work without uplifting then we shouldn't implement this, but if uplifting is required then I don't think it should be too much of a problem to do so.

[epage]

As far as I can tell, this is working as expected. cargo test is focused on building and preparing what is needed for running tests. It is not a superset of cargo build. I'm going to propose to the cargo team that we close this.

[weihanglo]

Agree with closing. Thank you for the triage.