6.0.0a -Fails to build on my powerpc box (debian wheezy)

[Torbjorn Granlund]

Richard Biener <rguenther at suse.de> writes:

  Well, I don't control the build bots (and the failing machine could
  have been a power6 host as well, for example).  But I've seen that
  for example gmp thinks that it builds for a 64 bit ABI unless
  I specify --host=i586-linux-gnu when running inside a 32bit virtual
  machine on a 64bit host (because /proc/cpuinfo obviously shows a
  64bit capable cpu).  Maybe GMP is trying to be a bit too clever here?
  What's wrong with using ./config.guess results, at least for
  --enable-fat binaries?  Thus, does the above (power7) mean that
  I get power7 code generated by default?  Mind that the machine I am
  building on is _not_ necessarily the lowest common dominator that
  I want gmp to be able to run on.

GMP tries to compile for the best ABI which the CPU and compiler
supports.  If you have a CPU and compiler which supports ABI=foo which
GMP considers to be better than ABI=bar, then it will choose foo and not
bar, only of either fails will GM consider ABI=bar.  If foo binaries
don't actually run on the system, then that's not something GMP takes
into account.  For what it is worth, the user might compile for a
different system, and the compiler available to GMP is a cross-compiler.

THere are documented ways of overriding the ABI default and compiler
choice.


Torbjörn
Please encrypt, key id 0xC8601622