[petsc-dev] GCC8 Fortran length changes from int to size_t

[Jed Brown]

Matthew Knepley <knepley at gmail.com> writes:

> On Wed, May 2, 2018 at 1:14 PM, Jed Brown <jed at jedbrown.org> wrote:
>
>> Matthew Knepley <knepley at gmail.com> writes:
>>
>> > On Wed, May 2, 2018 at 12:15 PM, Jed Brown <jed at jedbrown.org> wrote:
>> >
>> >> Lisandro Dalcin <dalcinl at gmail.com> writes:
>> >>
>> >> > On Wed, 2 May 2018 at 17:29, Satish Balay <balay at mcs.anl.gov> wrote:
>> >> >
>> >> >> So we need a 64bit arm with gcc8 - for this testcase failure?
>> >> >
>> >> >
>> >> > Or a big-endian machine/OS ?
>> >>
>> >> Shouldn't be necessary, but why are we so concerned about making a test
>> >> case fail instead of implementing correct behavior?
>> >>
>> >
>> > How would you know your fix was correct?
>>
>> Documentation and/or reading assembly.
>
>
> Its good that tests are superfluous for code you write because it conforms
> to standards and
> you check the assembly. We should take out all the SF tests. That would
> decrease runtime.

Dude, it's nothing alike.  This is about a calling convention (something
not defined by the standard, and where "getting it wrong" gives
undefined behavior rather than a defined error).  The current configure
does not attempt to get a failing test for HAVE_FORTRAN_MIXED_STR_ARG,
which is the closest analogy.  Your insistence on us demonstrating
failing code is like being informed of an obvious security hole (e.g.,
unvalidated sprintf) and insisting on a working exploit before patching.