Build failure for 5.1.0-RC2 on Mac OS 10.8

Torbjorn Granlund tg at
Tue Dec 11 23:15:35 CET 2012

Jack Howarth <howarth at> writes:

  +  .set LC1, L1-Ltab
  +  .long	LC1
  and his reasoning as follows...
  The deal is, you want the assembler to evaluate the deltas and not
  pass relocations on to the linker. You can see the compiler doing this
  is you just compile a simple switch like:
Thank for the information.

This behaviour is highly unusual, and I know not all Apple assembler
releases require that, and I would expect some Apple assemblers to
actually reject that syntax.

It should be noted that no other concurrent assemblers fail to compute
sym1-sym2 when both symbols are in the same segment.  They don't
require any contortions in order to trigger symbol arithmetic that is
fully foldable.

We know that the code of RC1 works with some Apple assemblers/linkers,
but not with ones older and newer.  This make me take an extremely
conservative stance wrt Apple porting.

We will never get stable Apple support, however hard we try.  Only if
Apple decided for what to support, and wrote a good test suite
enforcing the decision, and educated their staff, than we could reach
the same GMP quality with Apple as with other platforms.  But the
damage is already done with the mess of inconsistent tools out there.


More information about the gmp-bugs mailing list