impl FromStr for sys::Signal

[quodlibetor]

This implements both ALLCAPS and lowercase full name (including SIG) and short
name.

This matches what kill accepts in many shells, and it also allows the Debug
representation of Signal, which means it can be round-tripped, if desired.

[quodlibetor]

The failure looks spurious:

error: failed to build archive: No space left on device

[asomers]

I fixed the ENOSPC and restarted the builds.

[asomers]

I don't agree with this change. If nix were an interactive shell it would make sense. But nix is a library. Why would we want to intentionally broaden our API by publishing multiple spellings of common symbols? I've worked on projects that did things like this. I just creates inconsistency and confusion, not to mention more maintenance.

[quodlibetor]

Fair enough. Would you be more receptive to a change that only accepts a subset of these spellings? For example, just the exact matches (only SIGHUP, not SIGHUP and HUP and sighup and hup)? I'm happy to adjust. This is also just a way to reduce boilerplate and a newtype in one of my projects so I wouldn't be terribly upset if you just want to reject this.

[Susurrus]

I don't like that there isn't a ToStr implementation to go with this. We should either have both or neither. I'm also of the opinion that there should only be one canonical representation of each signal, the full capitalized version.

@quodlibetor What's your use case here? You want to parse command-line arguments specifying signals or are these strings in a data stream?

[quodlibetor]

@Susurrus I'm happy to implement Display and AsRef<str> for these. My use case is command-line flags (here).

This PR verifies that the Debug impl does roundtrip, which isn't as neat as adding to_string but does work. I agree that round-tripping is important.

I would slightly prefer to keep both all-caps versions (e.g. SIGHUP and HUP), but that's just because I think of them as synonyms and I would be happy to remove everything if coming in with a limited feature spec would be more likely to get accepted.

[Susurrus]

What I would suggest is that we only support the SIG[A-Z]* format for signal identifiers. Additionally we implement Display for Signal and only test that that can be round-tripped. Additionally our round-tripped tests should do it for every variant and some ones that are clearly erroneous, rather than testing just a singular one.

[quodlibetor]

I've done most of that. I implemented AsRef<str> as well, because it's possible to implement Display in terms of AsRef so the total amount of code is almost the same, but with all the benefits of not needing to allocate for the common-case in rust of just needing an &str.

[quodlibetor]

I would still prefer to include the short-all-caps names, but if you're happy with this then so am I.

[Mange]

I'd love to see this shipped. Is there anything I can help with?

[Susurrus]

@Mange Take it for a spin in your own code, feel free to leave a review or feedback.

[Susurrus]

This looks good to me at this point. @asomers any comments?

Before merging, it does need the following:

• Probably a rebase
• Squashed into a single commit with a good commit message
• A CHANGELOG "Added" entry

[asomers]

I agree with susurrus. It looks good except for the CHANGELOG entry and the squash.

[Susurrus]

bors r+

[bors]

Build succeeded

• buildbot/nix-rust/nix amd64_fbsd11
• buildbot/nix-rust/nix i386_fbsd11
• continuous-integration/travis-ci/push