test: fix eval and file-close - #43
Conversation
| let workspace = Path::new(env!("CARGO_MANIFEST_DIR")) | ||
| .parent() | ||
| .expect("vmm crate is inside the workspace"); | ||
|
|
||
| // Cargo supplies both binaries; VMM expects the guest ELF before guest argv | ||
| let child = Command::new(env!("CARGO_BIN_EXE_vmm")) | ||
| .arg(env!("CARGO_BIN_FILE_FIX_GUEST_fix")) | ||
| .arg("eval") | ||
| .arg(program.path()) | ||
| .current_dir(workspace) | ||
| .stdout(Stdio::piped()) | ||
| .stderr(Stdio::piped()) | ||
| .spawn() | ||
| .expect("launch Fix-on-Arca under the VMM"); |
There was a problem hiding this comment.
Can you have Rust embed the guest binary into this one instead or reading it? This currently seems brittle.
There was a problem hiding this comment.
In the Cargo.toml I added
[dev-dependencies]
fix-guest = { package = "fix", path = "../fix", artifact = "bin:fix", target = "x86_64-unknown-none" }
so calling cargo test should build the Fix guest artifact first. I tried embedding it but couldn't really see a clean way of doing so. Also, I'm not quite sure of the benefit of that approach, is there something that could fail with how it is right now?
There was a problem hiding this comment.
I think the main thing that concerns me is the path possibly being incorrect depending on the user's config; like their editor might invoke cargo test in the current file's directory rather than the project root. Right now the path is a compile-time literal but it's being evaluated at runtime which is odd. Also philosophically I'm hesitant to have a hard dependency that isn't visible to the compiler/runtime/OS...
There was a problem hiding this comment.
FYI, I moved the existing WAT build pipeline into a fix_wasm crate. It builds all the .wat files and exports their ELF's literal bytes. fix/build.rs uses those outputs, and the VMM test embeds fix_wasm::ADDBLOB directly.
a672b33 to
c53cb3d
Compare
c53cb3d to
f475175
Compare
|
Now that I'm reading this more closely, it seems like a bit of a roundabout way to accomplish the original goal of testing "fix eval" in CI and making sure it closes properly. E.g. having the test re-exec itself as a subprocess (to be able to capture the error message in stdout). I suspect there's probably an easier route where, e.g., maybe we add an assert_return "command" to the fix program (or the fix scripting language?) so we can directly write tests and run them from the command line without needing much extra in Rust. |
3 participants
Adds test for entire fix eval pipeline