Re: gcc-2.95.3, but not gcc-2.95.2, requires gas on hppa2.0-hp-hpux10.20

[Martin Buchholz <martin at xemacs dot org>]

Executive summary:  
Confirmed that Bernd's patch solves the HP as build problem.  The old
`make compare' problem remains.

>>>>> "J" == Jeffrey A Law <law@redhat.com> writes:

J>   In message <14957.7716.184344.679629@mule.m17n.org>you write:
>> I reran the build.  Here is what I get this time:
>> 
>> make[3]: Entering directory `/usr/local/xemacs/gcc-2.95.3/gcc'
>> ./xgcc -B/usr/local/hppa2.0-hp-hpux10.20/bin/ -B./ -I/usr/local/hppa2.0-hp-
>> hpux10.20/include -O2   -DIN_GCC -DHAIFA    -O2 -O2 -I./include  -fPIC -g1 
>> -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED   -O2 -I. -I/project/xemacs/import/g
>> cc-2.95.3.test2/gcc -I/project/xemacs/import/gcc-2.95.3.test2/gcc/config -I
>> /project/xemacs/import/gcc-2.95.3.test2/gcc/../include \
>> -c /project/xemacs/import/gcc-2.95.3.test2/gcc/cp/tinfo.cc
>> cc1plus: warning: -g is only supported when using GAS on this processor,
>> cc1plus: warning: -g option disabled.
>> as: error 7403: undefined label - L$0175 (7403)
>> 
>> It does seem like a gasism has infected the generated assembler code.
>> 
>> To test this hypothesis, I built a binutils-2.10 for my
>> hppa2.0-hp-hpux10.20 system, and then retried bootstrapping
>> gcc-2.95.3.test2 again, adding --with-as=/path/to/as to the
>> configure command.
>> 
>> This time, tinfo.cc built just fine.
J> It's entirely possible the EH code isn't compatible with HP's assembler.

J> My recommendation, don't use HP's assembler.  It has many significant
J> problems, including some which make it impossible for GCC to emit correct
J> code for certain programs.

J> GAS (of course) doesn't have these problems, in addition is supports debugging.

Wearing my hat of XEmacs maintainer, I will take Jeff's advice and use
gcc+gas, so that I can debug XEmacs.

But wearing my gcc beta tester hat, gcc (and the rest of the "Cygnus
toolchain") should be able to bootstrap on a vanilla system using the
crufty HP compiler and assembler.

Now building gcc 2.95.3 with HP `as' fails in the same way as 2.95.2 -
the compare step fails with errors like this:

cp/method.o differs
cp/parse.o differs
cp/pt.o differs
cp/ptree.o differs
cp/repo.o differs
cp/rtti.o differs
cp/search.o differs
cp/semantics.o differs
make[1]: *** [compare] Error 1
make[1]: Leaving directory `/project/xemacs/import/test-gcc/2.95/hppa2.0-hp-hpux10.20/build/gcc'
make: *** [bootstrap] Error 2

This is only a minor problem, since `make install' seems to work just
fine anyways.  Although this is not a regression from gcc-2.95.2, it's
still clearly a bug.  It would be nice if there were a simple way to
suppress or fix the compare step when building with HP as.

I build my compilers using 

configure && make bootstrap && make install

It would be bad to have to replace the && with ;