Subject: Re: egcs / koffice
To: None <rearnsha@arm.com, richard.earnshaw@arm.com>
From: Todd Vierling <tv@pobox.com>
List: port-arm32
Date: 10/25/1998 11:58:22
On Sun, 25 Oct 1998, Richard Earnshaw wrote:

: > So it means that moving to EGCS on ARM entails a severe transition, because
: > installing a new shared libc runs the risk of breaking every call to ldiv and
: > so on.

: Yes it looks like it, unless we set these up to make them behave the same 
: way as we are thinking of making the softfloat code work; however I don't 
: think putting these attributes on things that are visible to  "userland" 
: would be a good move.  We can get away with it with the softfloat code, 
: since that just provides compiler support functions.

We can't get away with it at all.  You're making the implied assumption that
no one has compiled any third party code, particularly shared libraries, on
the old gcc.  Should I introduce you to pkgsrc?

: BTW, the code that was in gcc before was just plain broken -- for example, 
: structures of size 4, 8 or 16 bytes would get returned in registers; but 
: 12 bytes structures would get returned in memory.

Whatever the "problem" was, it's now our arm32 ABI.  If you want an
APCS-compatible mechanism, it has to be able to return in *both* registers
and stack by default, and assume that the default is registers for
compatibility.  Use of -f*-return ONLY would change that behavior and
expectation.

Until this is resolved, I cannot enable egcs for arm32.

-- 
-- Todd Vierling (Personal tv@pobox.com; Bus. todd_vierling@xn.xerox.com)