Re: Unexpected zero return code from `throw std::runtime_error`

[Adam Dinwoodie <adam@dinwoodie.org>]

On Sat, Jul 02, 2022 at 10:19:44PM +0200, Csaba Raduly wrote:
> On Sat, 2 Jul 2022 at 15:22, Adam Dinwoodie  wrote:
> >
> > I'm currently experimenting with compiling rdfind for Cygwin, and one of
> > the testcases is failing because std::runtime_error is expected to
> > result in the compiled program exiting with a non-zero return code,
> > but on Cygwin, it just seems to cause the program to terminate with a
> > zero return code.
> >
> > I've attached a simple test case.  Compare the output on Cygwin...
> >
> >     $ ./test.sh
> >     + cat
> >     + g++ test.cc
> >     + [[ -x a.exe ]]
> >     + ./a.exe
> >     + rc=0
> >     + ((  rc == 0  ))
> >     + echo 'Unexpected zero return code from execution'
> >     Unexpected zero return code from execution
> >     + exit 1
> >
> > ...with the output from one of my Debian boxes...
> >
> >     $ ./test.sh
> >     + cat
> >     + g++ test.cc
> >     + [[ -x a.exe ]]
> >     + [[ -x a.out ]]
> >     + ./a.out
> >     terminate called after throwing an instance of 'std::runtime_error'
> >       what():  Test error
> >     ./test.sh: line 21: 566327 Aborted                 ./a.out
> >     + rc=134
> >     + ((  rc == 0  ))
> >     + echo 'Expected non-zero return code received: 134'
> >     Expected non-zero return code received: 134
> >     + exit 0
> >
> > I'm not massively familiar with C++, so I could well be missing
> > something obvious, but this seems like an unexpected difference between
> > Cygwin and other *nix platforms.  Is this a Cygwin bug, or am I doing
> > something wrong?
> 
> 
> Works just fine for me, unless I misunderstood something.
> 
> $ /cygdrive/c/Users/Csaba/Downloads/test.sh
> + cat
> + g++ test.cc
> + [[ -x a.exe ]]
> + ./a.exe
> terminate called after throwing an instance of 'std::runtime_error'
>   what():  Test error
> /cygdrive/c/Users/Csaba/Downloads/test.sh: line 21:   641 Aborted
>            (core dumped) ./a.exe
> + rc=134
> + ((  rc == 0  ))
> + echo 'Expected non-zero return code received: 134'
> Expected non-zero return code received: 134
> + exit 0

That's definitely closer to the behaviour I'd expect!  I don't think it
should be dumping a core, though, at least based on the behaviour I see
on Debian.

I'd like to have a look at what's different about your environment;
could you share the output from `cygcheck -srv >cygcheck.out` on your
system?

> The if goes in the "else" case because rc is not 0, as it should be.
> (shoudn't the condition be $rc == 0 ?)
> After
> 
> $ diff -u /cygdrive/c/Users/Csaba/Downloads/test.sh test.sh
> --- /cygdrive/c/Users/Csaba/Downloads/test.sh   2022-07-02
> 22:09:21.506377100 +0200
> +++ test.sh     2022-07-02 22:15:08.670809700 +0200
> @@ -20,7 +20,8 @@
>         exit 2
>  fi
> 
> -if (( rc == 0 )); then
> +if [[ $rc == 0 ]]
> +then
>         echo 'Unexpected zero return code from execution'
>         exit 1
>  else
> 
> I get
> 
> $ ./test.sh
> + cat
> + g++ test.cc
> + [[ -x a.exe ]]
> + ./a.exe
> terminate called after throwing an instance of 'std::runtime_error'
>   what():  Test error
> ./test.sh: line 21:   660 Aborted                 (core dumped) ./a.exe
> + rc=134
> + [[ 134 == 0 ]]
> + echo 'Expected non-zero return code received: 134'
> Expected non-zero return code received: 134
> + exit 0

`(( rc == 0 ))` is comparing the value of `$rc` against the number 0;
the `$` isn't necessary in `(( ... ))` tests.  `(( rc == 0 ))` is
essentially identical to `[[ $rc -eq 0 ]]`.

Your change means the comparison is done as a string comparison rather
than a numeric comparison.  That doesn't make any difference here: '0'
is '0' regardless of whether you're testing it as a number or as a
string.