[Bug-gnulib] Re: Autoconf and differences between GNU getopt and OpenBSD getopt

[Simon Josefsson]

Derek Robert Price <address@hidden> writes:

> Simon Josefsson wrote:
>
>>Derek Robert Price <address@hidden> writes:
>>
>>>>Do BSD really have getopt_long_only?  If so, then the getopt.m4 need
>>>>to be improved...
>>>
>>>
>>>According to Alex, yes, it does.
>>
>>
>>Ok, then getopt need to be fixed.
>>
>>But it seems bad style to replace a system getopt unless there is
>>something wrong with it?  In the POSIX sense, I mean.
>>
>>(Below I'm assuming the code Alex posted only exploited GNU specific
>>features.  If the code only relied on POSIX semantics, it seems that a
>>smaller fix to getopt.m4 is sufficient.)
>>
>>Perhaps we can create two modules: getopt and getopt-gnu.  This is
>>similar to getpass and getpass-gnu.  The getopt module would only be
>>enabled if POSIX compliant getopt and getopt_long_only are missing.
>>The getopt-gnu would be enabled when the system provided functions
>>don't implement the GNU interface of getopt*.
>>
>>Thoughts?
>
>
> Just thought I would point out that the current getopt module was
> attempting to be the "new" getopt-gnu you describe since prior to the
> BSD change, getopt_long_only meant a GNU implementation of getopt, in
> case anyone had missed it.
>
> Regardless, I wouldn't object to two modules, as long as one will
> provide a GNU-style getopt on recent Open & Free BSDs.

Ok, what do you (as in everybody) think about renaming 'getopt' to
'getopt-gnu' and fixing the test in getopt.m4?

Then if someone ever need the less intrusive 'getopt' module, it could
be added later on.  I guess nobody need it currently?

Do you have a tweaked getopt.m4 that would work better?  I guess it
has come to the point where it actually will have to compile and run
some small program.

Thanks.