Re: GMP 4.1 on Solaris, HP-UX

[Torbjorn Granlund]

Albert Chin <address@hidden> writes:

  >   =====================
  >   14 of 52 tests failed
  >   =====================
  
I am unable to reproduce these errors on a PA8500 system
running HPUX 11.11, except the t-sizeinbase one.

The t-sizeinbase.c test case is really bogus, and cannot be
expected to work on a segmented system like HP-UX.  ANSI C
allows a pointer to point within an object and just after
it.  Here we create a pointer point far under the object.

  Ok, the patch below fixes 13 of the failures. The patch
  forces ".level 1.1" in the ASM files.

It is really puzzling how this can have the effect you've
observed.  I could understand how a .level declaration could
make files not assemble, but how they can pass the assembler
and then not run, is really puzzling.

I will not install your patch since part of it is
undesirable (disabling pa7100 optimized code) and part of it
seems dubious (.level stuff).  Older assembler did not
understand the .level directive.  Only if someone can
explain to me how .level can avoid a mis-assembly, and tell
me which assembly version has the problem, I am likely to be
convinced.  Still, your patch isn't right for older
assemblers; something more sophisticated is needed.

-- 
Torbjörn