Re: sysexits.h codes? (was: Re: [RFC][PATCH] `newuser` prompt theme)

["Daniel Shahaf" <d.s@xxxxxxxxxxxxxxxxxx>]

Marlon Richert wrote on Fri, 30 Apr 2021 19:40 +00:00:
> On Wed, Apr 14, 2021 at 5:09 PM Daniel Shahaf <d.s@xxxxxxxxxxxxxxxxxx> wrote:
> > > >> @@ -0,0 +1,197 @@
> > > >> +readonly -ga sysexits=(
> > > >> +    USAGE
> > > > ⋮
> > > >> +    CONFIG
> > > >
> > > > Two interrelated scripts on zsh.org hardcode these constants too.
> > > > I wonder if we should provide these constants in a standard autoloaded
> > > > function or preset variable.
> > >
> > > A preset variable would be great. In C code, you can get the names and values from `sysexits.h`, which is available in both GNU and BSD.
> >
> > An associative array mapping symbolic names to integers?
> 
> I think we could settle for just an array of names, to be consistent
> with $signals.

With $signals there's the annoyance that there's an off-by-one between
the signals' value and its array index.

With a $sysexits that's just an array of names there'd be an off-by-64.
That'd be a magic number.

Not a deal breaker, but not ideal either.

> 
> > Just those available on the target system, or?
> 
> sysexits.h tries to standardize exit codes. Having different $sysexits
> on different target systems would defeat its purpose.
> 

That's two separate issues: one, whether sysexits.h should define the
same numeric values everywhere; two, whether $sysexits should reflect
the local sysexits.h or a "well-known" one, in case they're different.

> > > It would be great if, besides printing the name, it could also add a
> > > short description of what the exit code actually means. Something like
> > > “SIGPIPE (140): write on a pipe with no reader”.
> >
> > There's precedent to work with:
> >
> >     % =pwd | true
> >     zsh: broken pipe  =pwd
> >
> > I suppose the default should be about this level of verbosity.
> >
> > Other degrees of verbosity could be considered.
> 
> Having an option to make error messages more verbose would be great
> for new users. "broken pipe", for example, sounds a lot worse than it
> really is. :)
> 

"broken pipe" is just what libc gives, so, shouldn't such an option be
implemented in libc itself?  Besides, libc's message may be localized,
and there's no reason to require users to choose between a verbose
message and a localized one.

> Verbose descriptions could be copied from BSD's signal.h or man 2
> sigaction. I think those are pretty decent.
> 
> Speaking of which, this reminds me of what Paul wrote in workers/48236:
> 
> On Wed, Mar 24, 2021 at 4:29 PM Paul <GammaFunction@xxxxxxxxxxx> wrote:
> > - I half-agree with not unsetting clobber. noclobber is better default
> >   behavior (it prevents much more damaging surprises), but just seeing
> >   "file exists" or "no such file or directory" without any '>|' or '>>|'
> >   suggestion is frustrating.
> 
> That could use a more helpful message, too.

Yeah.  The code should say what it was trying to do when it called the
syscall which returned that error; i.e., «zerrnam(foo, "while doing bar:
%e", errno)» rather than «zerrnam(foo, "%e", errno)».

Daniel
(replying out of order)