The compiler keeps going on Ctrl-C

[RalfJung]

Sometimes when things go really wrong the compiler emits a ton of error messages. This can actually take a while to render them all. And since somewhat recently, hitting Ctrl-C no longer interrupts that, I get a thousand error messages displayed after Ctrl-C.

I suspect this may have to do with #111769. The Ctrl-C handler does nothing at all on the first Ctrl-C unless the compiler is in the middle of const-eval.

Cc @saethlin

[saethlin]

Do you have a reproducer? I have not run into this myself.

The implementation should also ensure that the compiler exits promptly if you hit ctrl+c again. Does it?

[matthiaskrgr]

if stuff keeps going on ctrl+c, sometimes suspending the process (ctrl+z) and then killing the terminal does the trick somehow 😅

[RalfJung]

Do you have a reproducer? I have not run into this myself.

Unfortunately no, I encountered this during some rustc hacking where I ended up getting well over 1k error messages in the standard library build and that turned out to be hard to cancel.

The implementation should also ensure that the compiler exits promptly if you hit ctrl+c again. Does it?

I thought I had tried that, but I am not entirely sure.

[ChrisDenton]

Is it possible the shell is part of the problem? If, by the time you hit ctrl-c, it's working through a large amount of output it still has buffered then it might be awhile before you see a response.

[RalfJung]

It's possible, sure -- but usually it reacts to Ctrl-C immediately even when there's tons of output.

[fmease]

Have you tried reverting #111769 to check if that fixes it?

[RalfJung]

No. At the time I was busy getting a PR done and now I don't have a reproducing example any more.

[bjorn3]

#111769 only checks for ctrl-c during const eval, so if no const eval runs for an extended period of time, the ctrl-c will be silently ignored. This is not the only problem with that PR. It also caused #124265, and shows the "compilation was interrupted" error after the prompt is already drawn by bash.

[workingjubilee]

is there a good way to make rustc loop endlessly without ever touching const eval? maybe via the trait solver? 🤔

it would help for debugging this.

[fmease]

Yeah, just take any I-hang Issue: The compiler never terminates, due to infinite loops, deadlock, livelock, etc. issue (none are A-const-eval Area: Constant evaluation, covers all const contexts (static, const fn, ...) anyway).
I can reproduce this issue with #122443 for example:

• nightly-2024-03-26 (before #111769): Exits after a single ^C
• nightly-2024-03-27 (after #111769): Requires ^C^C to exit

bjorn3's analysis checks out.

[workingjubilee]

cool! I wasn't sure how much of the compiler's code flowed through const-eval, is why I asked.

[bjorn3]

Another case where this is observed: #125428

IMO we should revert #111769 or at least make the signal handler run exit after a short timeout.

[saethlin]

I'll post a PR that adds a timeout later today, but by all means if you feel strongly just put up a PR. The initial handler here was just added through the normal PR process, there's no reason to be more cautious on reverting or modifying it.