Cancel that offer!

[Tom Tkacik]

In article <448 at manta.pha.pa.us> brant at manta.pha.pa.us (Brant Cheikes) writes:
>In a previous article, I offered GCC 1.31 for anonymous uucp from
>manta.  That offer is canceled; sorry, I jumped the gun this time.
>Further testing indicates that there are still some problems.  Among
>other things, I tried compiling Gnu Emacs 18.52 with GCC---the
>resulting executable dumps core with a Memory Fault.  Personally, I
>feel that if you can't compile with gcc -O and have it work, it's not
>worth using.

I agree with Alex Crain about the problems people are seeing with gcc
when using the optimizer.  I have seen only a few problems with my
current version (gcc-1.28), and have been able to write 4 line programs
which show the problem, (looking at the wrong assembly code generated).
Now, I cannot say with certainty that Brant is wrong about gcc-1.31 having
problems, but if the only proof is that emacs crashes when run, there are
many things that can cause that.

Gcc has many flags which cause it to act differently.  The one that I find
I need most frequently is the -fwritable-strings option.
Without it, gcc puts all strings in read-only memory.  The result is a core
dump whenever you try to write to an initialized string.  I would guess
that this may be the problem.  Another flag to try is -traditional.
This makes the pre-processor act slightly differently, and is another
frequent problem I have found.

When using gcc, if the executable crashes, but doesn't when compiled with pcc,
these are now the first things I try, and usually, (but not always) the result
is a working program.

I think that when talking about bugs remaining in gcc, it would be helpful
to give a specific problem, and not just a claim of gcc acting wrong.
It would certainly make me believe that the author is correct, and not just
using gcc improperly.

---
Tom Tkacik
GM Research Labs,     Warren, MI 48087
{umix, uunet!edsews}!rphroy!megatron!tkacik