Tool crash when building RPi code

The archive from July 21 that I’ve been building repeatedly, while working on the IIC code, builds fine.

The archive from yesterday shows one of the tools crashing during the build:

gnu.diff -df >null: CLIB:h.swis derived.swis

Internal error: undefined instruction at &0001CBC6

Postmortem requested
1cbc2 in anonymous function
d2b0 in anonymous function
d380 in anonymous function
e270 in anonymous function
c80c in anonymous function
  Arg5: 00000000 0
  Arg4: 00000000 0
  Arg3: 0x00023915 145685
  Arg2: 0x00023909 145673
  Arg1: 00000000 0
b600 in anonymous function
  Arg2: 0x000238dc 145628 -> [0x000238fc 0x00023905 0x00023909 0x00023915]
  Arg1: 0x00000004 4 -> [0xe59ff2b8 0xe59ff2b8 0xe59ff2b8 0xe59ff2b8]
9660 in anonymous function
  Arg2: 0x00008704 34564 -> [0xe1a0c00d 0xe92ddbf3 0xe24cb004 0xe15d000a]
  Arg1: 0x000233e0 144352 -> [0x2e756e67 0x66666964 0x66642d20 0x756e3e20]
fc13f018 in shared library function
8300 in function ___init
AMU: *** ignored non-zero exit (1) ***

Alignment exceptions are off. The conditions are, TTBOMK, identical to those under which the July 21 archive built successfully.

Is this a known problem with a known solution? Or, since the build process seems to carry on anyway, do I just ignore it?

Looks like the copy of diff in CVS isn’t ARMv6/v7 safe, although I don’t remember anyone complaining about it in the past. At some point I’ll have to try doing a build on a BB myself and work out how to rebuild all the old tools.

It looks like it’s safe to ignore this particular crash – the clib makefile only uses diff to check if the exported swis.h needs updating, and since the tool crashing has resulted in a non-zero return code, it will result in the desired behaviour of updating the file.

OK, thanks, Jeffrey.