Add POSIX compliant ctrl-c handling

[sloane-shark]

first pass at addressing #872

based on/cribbed from ghci

the main contribution of this feature is more user-friendly signal handling, particularly making it so that ucm responds to SIGINT similarly to other shells, like ghci. if you're typing some input and hit ^c, you will now cancel that input and be given a new line to type on. you can also cancel execution/evaluation of running code in ucm the same way. ^d will emit an EOF that will kill the shell the way ^c used to.

this only handles posix signals at the moment, but could be feasibly be extended with the same (CPP-based) windows handling that ghci uses.

i haven't made many contributions to codebases this size, so let me know if some of this would be better added somewhere else or packaged up in a more modular way.

[aryairani]

Thanks @matthewess, LGTM. Didn't notice it was ready for review.

For all of us who haven't looked closely at the issue and maybe haven't tried it out, could you add a comment in the PR to describe:

• How specifically does it change the behavior of ucm & how can we tell it's doing what it's expected to: e.g., is there a reproducible case that it handles better?
• The PR description calls it a "first-pass" — what would you say is left to do? Just the Windows port, eventually, or something more?

Thanks again!

[mitchellwrosen]

LGTM, I think this is ready to merge once conflicts are resolved.

@aryairani, in a nutshell, it changes the default behavior of the GHC runtime (translate the first SIGINT into a UserInterrupt delivered to the main thread) into ghci-like behavior (translate every SIGINT into a UserInterrupt delivered to the main thread).

Elsewhere in the commandline main you're catching UserInterrupt and re-polling for user input (hence why you currently get the nice re-prompt behavior, but only once nvm, that was added in this patch, too).

[mitchellwrosen]

I think it would probably be more correct to re-throw non UserInterrupt exceptions rather than just quit.

[aryairani]

Should we wait on this?

[mitchellwrosen]

I think it could be fixed up post-merge, it's good enuf ;)

[sloane-shark]

added a bit more context but @mitchellwrosen has got it. @aryairani yes, the remaining work is Windows compatibility, I believe. but as mitchell called out there should probably be more delicate handling of non UserInterrupt exceptions, as well.

[aryairani]

Great! Thanks!

[aryairani]

If you wouldn’t mind creating tickets to track the remaining work, that would be awesome. :)