On Wed, 5 Aug 1998, David Edelsohn wrote:

: Todd> Doesn't matter how fast it runs - there are some types of constructs at
: Todd> higher optimization levels that will make the resulting code fall over.  In
: Todd> short, you can use it to generate reliable code on powerpc if you never use
: Todd> -O2 or -O3.
: 
: 	I don't think that you can or should make this type of
: generalization.  egcs-1.0.3 has problems as does gcc-2.7.2.2 and
: gcc-2.8.1.  It all depends on how you are using the compiler and the
: source code in question.

I'm referring to the various reports I have read on egcs-bugs@cygnus.com,
which I've read since the egcs beta days.  Out of all the
platforms that 1.0.3 supports, only i386 and sparc have fairly solid C
codegen; other platforms are missing some bits in optimization (and some,
like arm, will sometimes generate broken code without optimization).  The
generalization about 1.0.3 is pretty valid - just ask random people at
Cygnus how much they would trust 1.0.3 for big production projects.  :)

I've been wary of running egcs 1.0.3 on anything but i386 for a while
because of this.  There have been too many "this is broken" reports with
valid code that have been replied to with, "Fixed in 1.1 mainline."  C++ is
worse - it's rather broken even on i386.

You'll note that I keep saying i386 works well in C:  that's Linux's
mainstay architecture, and also the one that had the worst liberties taken
by programmers that you later mentioned.

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