Re: powerpc-linux: C++ looks completly broken

[Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>]

At 15:44 21.02.2001, Andreas Jaeger wrote:

>Trying to run today's gcc-3.0 branch from CVS on powerpc with glibc
>2.2, binutils 2.10.91.0.2 (from HJ - similiar to binutils from CVS) I
>get a huge number of failures for C++.  I'm sending the test log
>separatly to gcc-testresults@gcc.gnu.org.  Looking at the archives I
>noticed that Franz has only a few failures - but everybody else
>reporting powerpc-linux bootstrap has these over 900 failures.
>
>Franz, why are your results better than those from the other folks?
>Are you using extra patches?  Can you make them available and get them
>integrated into GCC, please?
>
>Looking through g++.log I notice that most failures are like the
>following:
>
>FAIL: g++.benjamin/tem06.C (test for excess errors)
>/usr/src/packages/BUILD/gcc/obj/gcc/testsuite/../libgcc_s.so: undefined 
>reference to `__SBSS2_END__'
>/usr/src/packages/BUILD/gcc/obj/gcc/testsuite/../libgcc_s.so: undefined 
>reference to `__init'
>/usr/src/packages/BUILD/gcc/obj/gcc/testsuite/../libgcc_s.so: undefined 
>reference to `__SBSS_END__'
>/usr/src/packages/BUILD/gcc/obj/gcc/testsuite/../libgcc_s.so: undefined 
>reference to `__SDATA2_START__'
>/usr/src/packages/BUILD/gcc/obj/gcc/testsuite/../libgcc_s.so: undefined 
>reference to `__SDATA_START__'
>collect2: ld returned 1 exit status
>
>What's going on with __SBSS2_END__ ?  It seems that at the moment C++
>support for PowerPC is broken.

Heh, I forgot to add this to my prepend file for test_summary, but 
currently you need something like:

[fsirl@entropy:~/cvsx/gcc3]$ cvs diff gcc/config/rs6000/t-ppccomm
Index: gcc/config/rs6000/t-ppccomm
===================================================================
RCS file: /cvs/gcc/egcs/gcc/config/rs6000/t-ppccomm,v
retrieving revision 1.8
diff -u -p -r1.8 t-ppccomm
--- t-ppccomm   2000/04/14 23:16:25     1.8
+++ t-ppccomm   2001/02/21 14:54:17
@@ -6,7 +6,7 @@ CROSS_LIBGCC1 =

  # These are really part of libgcc1, but this will cause them to be
  # built correctly, so... [taken from t-sparclite]
-LIB2FUNCS_EXTRA = eabi.S tramp.S
+LIB2FUNCS_EXTRA = tramp.S

  # We want fine grained libraries, so use the new code to build the
  # floating point emulation libraries.

to let C++ work correctly on powerpc-linux-gnu. This is a fallout of the 
shared libgcc thing and I discussed it a little bit with Mark on the list. 
I haven't put any pressure on it yet as long as the shared libgcc 
discussion is still as heated as it is now :-).

Apart from that, I only have a pending code correctness issue with 
glibc-2.2 being miscompiled (time_output() in ld-time.c to be exact, maybe 
the code is not alias safe though) by gcc-3.0pre. That's caused by Bernd's 
patch to init_alias_analysis() early January.

Franz.