Re: Newbie having a egcs1.1.1 compile problem

[Paul Derbyshire <pderbysh at usa dot net>]

At 01:16 AM 2/3/99 +0100, you wrote:
>I am somewhat frustrated, because I have been trying to build cvs-egcs
>on linuxppc unsuccessfully since around December 12. 

If it were December 31 I'd be tempted to suspect a Y2K problem... showing
up a year early perhaps because of data that goes ahead one year. Not that
I have any idea why a compiler would bother with such data. Although I can
see a Y2K-buggy make or a make on a Y2K-buggy filesystem having trouble
around the start of 2000 thinking newly built files are very very old and
need recompiling, over and over and over... :-)

>Before that date, I had no problems building cvs-egcs. After that, for a
>while even compiling cvs-egcs was impossible. Now I can compile it using
>a specially patched egcs-1.1.1.

1. On and around December 12 did you change part of your system,
   especially system headers, environment settings, or where it
   expects to find system headers.
2. What exactly did you do to patch egcs-1.1.1 and why? I'm guessing
   maybe the patch is buggy.

>The resulting gcc doesn't, however,
>produce correct code: It compiles, for example, a linux kernel without
>error, but this kernel does not run. And I cannot do a make bootstrap.

Maybe the machine specs have gotten scrogged, whichever specs determine
assembly code and scheduling.

What is fatal signal 11 on your system? (That's the signal your bogus gcc
compiled cpp caught.) If it's an alignment fault, you probably somehow
ended up with a scrogged assembler or linker/the wrong assembler or
linker/a scrogged specification file that tells the assembler or linker
how/whether to align. So the occasional misaligned data access is being
generated.
All this is moot if the architecture you're running on doesn't care much
about alignment.

>So am all the more thrilled to hear that even an egcs-1.1.1 (do you know
>if this rawhide egcs rpm is made from egcs-1.1.1 release or from some
>later snapshot?) on i386-linux shows the same symptoms. So either both
>of us are doing something very stupid (which I doubt, of course), or
>this is a serious problem in egcs that should be solved.

Why don't you get a known good gcc BINARY distribution from redhat.com and
install it, then try to bootstrap egcs-1.1.1 from that? Do they have PPC
binaries?

BTW, the likely reason you have better performance with bootstrapped
compilers than binary distributions is that you probably have set things up
so your compiles are optimized tuned for the exact PPC chip you're using. A
generic PPC-optimized binary will run more slowly than one you built
specifically for your own machine. Just like a generic Pentium optimized
binary will be outperformed by one aimed at an MMX PII on an MMX PII
machine.
-- 
   .*.  "Clouds are not spheres, mountains are not cones, coastlines are not
-()  <  circles, and bark is not smooth, nor does lightning travel in a
   `*'  straight line."    -------------------------------------------------
        -- B. Mandelbrot  |http://surf.to/pgd.net
_____________________ ____|________     Paul Derbyshire     pderbysh@usa.net
Programmer & Humanist|ICQ: 10423848|