[BUG] Guest build.rs logs output to stdout overwrites tty display using Helix editor with Rust analyzer LSP

[nuke-web3]

Bug Report

When using rust-analyzer as the LSP with Helix as terminal-based editor, risc0 guest crate build output will be displayed overriding the display of the code loaded in the editor. See a screen recording here: helix-editor/helix#10038

I assumed this was an issue with the editor only at first (and it should not allow this to happen IMHO) but this would not be an issue if risc0 didn't write out to the stdout/err buffer in such a way that causes this. Note that no other crates I have used with helix have had such an issue for me, but all risc0 ones with methods/build.rs do.

Steps to Reproduce

1. Setup helix with rust analyzer
2. Open any risc0 project (includes the standard methods/build.rs tooling that writes logs to some buffer)
3. Text overriding display of editor

Expected behavior

No output should be displayed on the editor with the LSP running in a background task.
I would propose that the build tooling is modified to direct to stderr or other buffer that should avoid this 🤞

Your Environment

• risc0-zkvm version: 0.2 and above
• Rust version: 1.75 and above
• Platform/OS: Linux

Additional context

I have not pushed to get the Helix side of things patched to mitigate this, but I would think other editors like nvim might have the same issue too...?

[linear]

ZKVM-409 [BUG] Guest build.rs logs output overwrites Helix editor via Rust analyzer

[nuke-web3]

Ah I think I found the culprit:

risc0/risc0/build/src/lib.rs

Lines 454 to 458 in df99cdd

	// HACK: Attempt to bypass the parent cargo output capture and
	// send directly to the tty, if available. This way we get
	// progress messages from the inner cargo so the user doesn't
	// think it's just hanging.
	let tty_file = env::var("RISC0_GUEST_LOGFILE").unwrap_or_else(|_| "/dev/tty".to_string());

risc0/risc0/build/src/lib.rs

Lines 476 to 481 in df99cdd

	for line in BufReader::new(stderr).lines() {
	match &mut tty {
	Some(tty) => writeln!(tty, "{}: {}", pkg.name, line.unwrap()).unwrap(),
	None => eprintln!("{}", line.unwrap()),
	}
	}

So I would expect all tty based editors to have this issue, would you agree?

Thus we need some way to detect the process is run by the LSP and not write to the tty I think, not 100% sure how best to do that while preserving the behavior you want... Open to implement based on a suggestion from the team here 😀