From owner-help-octave at bevo dot che dot wisc dot edu  Fri Nov 10 00:48:40 1995
Subject: Re: gcc __complex__ data types 
From: John Utz <spaz at u dot washington dot edu>
To: niles at axp745 dot gsfc dot nasa dot gov
Cc: Wonkoo Kim <wkim+ at pitt dot edu>, Octave Users <help-octave@bevo.che.wisc.edu>,         Jim Van Zandt <jrv@mbunix.mitre.org>, niles@axp745.gsfc.nasa.gov,         niles@axp745.gsfc.nasa.gov
Date: Thu, 9 Nov 1995 22:48:23 -0800 (PST)

Hi ;

I wanted to add my 2 cents from what i remeber reading on the gcc mailing 
list ...


 On Fri, 10 Nov 1995 niles at axp745 dot gsfc dot nasa dot gov wrote:

> 
> >I don't know what was broken in gcc's complex data support.  I played a 
> >bit. It looks like normal operators are okay but not for math functions.
> >Is this the reason why gcc's __complex__ data support is considered a 
> >way of broken?  Do you know a bit more detail about what are "broken"?

	As i recall ( i saved the message because it was *really* 
interesting, now i can't find it! grr.. ) gcc will handle a singular 
complex value ok, but if the value is in array it chops off the complex 
part. While sounding kind of trivial to fix on the surface.. the 
specifics seem to be that ( i get vague here ) the portion of code that 
handles complex vars has a specific amount of registers assigned to it, 
and that was not enuf of then to do the job ( thus the complex part got 
hacked off ) and to free up the required register was gonna take a 
complete rewrite, and that nobody had the time for it anytime soon.


> No but the word is offically from Jim Wilson and Richard Kenner, they
> are GCC maintainers, if your really interested contact them.  I'm sure
> there e-mail addresses are somewhere in the GCC dist.  However, think
> twice before you do.  These guys are really busy and might not
> appreciate taking the time out if your only "interested".  However, if
> your really hardcore and think you might be able to help fix the
> problems I'm sure everyone would appreciate the contrabution.

hoh boy, and i was gonna take rick to task for telling the list to mail 
wilson and kenner, but he thoughtfully neglected to add there addresses 
to his mail. :-)

	seriously folks, if u even asked the question "whats a register" 
*stay away* from the gcc guys. Matter of fact, even if u do know how to 
make a compiler that works on umpteengazillion different platforms, and u 
are not willing to basically drop everything and try and fix this, dont 
bother them.

	Rest assured, i have seen them at work thru my little secret 
peephole and they know about the problem and it is on somebody's list.

	But oh man, unless u *like* an inbox full of flame, stay away!

( yes, as ungodly embarrasing as it is to admit to the whole list, i 
stuck my toungue in and got it ripped out by the roots.. :-( ) )
                                                            ^
                                                            no toungue!


> I don't know what is exactly broken, but I'm fairly sure using
> __complex__ can generate bad code and/or crash the compiler.  These
> we've experienced in the g77 work.
> 
> Hope this helps,
> 	Rick Niles.
> 
> 

*******************************************************************************
 John Utz	spaz at u dot washington dot edu
	idiocy is the impulse function in the convolution of life