Malcolm Wallace (malcolm-nhc@cs.york.ac.uk)
Wed, 21 Feb 2001 18:49:04 +0000
[compile nhc98 on AIX 4.3.3 using gcc 2.95.2.1] Zoltan, > First, I added -D__PPC__ to the OPT flag into top level Makefile.inc, > because it seemed to me that some header file requires this define. This suggests to me that the endian check at configure-time has failed somehow. Can you check the line from the output of ./configure that looks something like This machine's endian-ness is: -DHIGH_BYTE_FIRST If it actually reports -DLOW_BYTE_FIRST, then we have made a wrong assumption about PPC systems. I know the Apple Mac PPC is big-endian, but perhaps IBM PPC machines are little-endian? > After I replaced harch with config.guess from some GNU program to be > able to correctly identify the powerpc-ibm-aix4.3.3.0 target instead > of the 005406DA4C00-AIX-3 target string So that I can improve 'harch', can you report the results of each of these commands for me please?: uname -p uname -m uname -s uname -r arch machine > Tried to evaluate beyond end of function. > Instruction pointer at 10009 > What, run() returned! These error messages are a classic symptom that we have guessed the endian-ness of the machine wrongly. To be absolutely certain, could you please compile and run the following little C program and report its output? #include <stdio.h> main() { union { unsigned i; char s[4]; } convert; convert.i = 0x34333231; fwrite(&convert.s,sizeof(char),4,stdout); fputc('\n',stdout); } Another test whose output would be very useful in allowing us to add a symbol like __PPC__ automatically at configuration time is: $ cpp -dM /dev/null Regards, Malcolm