[Bug-gnulib] Re: new module: getsubopt

[Simon Josefsson]

Bruno Haible <address@hidden> writes:

> While doing this I noted an issue that may be a glibc bug:
>
> Does the POSIX:2001 description paragraph "If the string at *optionp ...
> value for the suboption is an error." hold only for the case that
> the "token" is found among the keylist, i.e. when the return value is >= 0?
>
> Or does it hold also when the return value is -1?
>
> In the first case, an application must be aware that *OPTIONP and *VALUEP
> are undefined after getsubopt() returns -1. (Or maybe they should both be
> unchanged??) I.e. the search loop cannot be continued by the calling program,
> as in the other case.
>
> In the second case, glibc is in error if it sets *VALUEP specially.

I know it is merely informative, but looking at the example code on
the OpenGroup web page suggest *VALUEP should be set, even if -1 is
returned, as the code appear to dereference *VALUEP in that case.

To me it seems more useful to set *VALUEP even for -1.

> I've added your patch.

Thanks,
Simon