Re: Complex Division Patch (reprise)

[craig at jcb-sc dot com]

>The existing gcc code for complex arithmetic is not "broken."  It
>implements the so-called usual mathematical definition.  As I have
>already pointed out, that is specifically mandated by ISO C9X.  I have
>also already pointed out exactly how the proposed change violates
>IEEE and where it violates ISO C9X.

Okay, then we'll disable the change for languages other than g77,
as I've already proposed.  Any objections?

>As language implementors you are not supposed to be making up the
>rules.  You are supposed to be implementing the rules that have
>already been written in the standard.  The burden of proof is on you,
>to show that your proposal obeys the standard.

I don't have a copy of ISO C9X around.  I doubt Toon does.

In fact, if you want egcs/gcc development to come to a standstill,
convince everyone working on it that they must read *every* pertinent
language and numerical standard before changing a line of code.

Please.  Take a deep breath, stop *criticizing*, and try *helping*.

>If fortran does not already have a definition of complex arithmetic,
>then I think it's a safe bet that the current standards efforts coming
>from Ada, LIA, NCEG, C9X, etc. etc. will soon find their way into
>fortran.  I think you would be better off adopting a standard method
>than arguing for a non-standard solution.

Toon's solution is not non-standard.  It complies perfectly with existing
standards -- including K&R C, ANSI C, ISO C, ANSI FORTRAN 77, ANSI/ISO
Fortran 90, ANSI/ISO Fortran 95, and, I believe, IEEE 754 and IEEE 854
as well.

So please stop claiming Toon's algorithm is non-standard.

If you're saying gcc is *already* completely C9X-standard, and that
this one change would break that, please say *that* more specifically.
It'd certainly be news to me.

In the meantime, until g77 decides to adhere to some standard, of
which I'm personally unaware, that mandates something different
than Toon's (and libf2c's) algorithm for complex divide, that's
what we'll use.

        tq vm, (burley)