A couple of bugs

[Dean Clamons]

1. gmp won't quite build correctly on my HP system. Here are the details about 
my system:

GMP version: 4.1
Configuration options: CC=cc
Compiler: cc
Compiler Version: cpp: HP92453-01 A.10.32.03 HP C Preprocessor
                           ccom: HP92453-01 G.10.32.05 HP C Compiler
uname -a: HP-UX null B.10.20 A 9000/898 1657613331 two-user license
.config.guess: hppa2.0-hp-hpux10.20
./configfsf.guess: hppa2.0-hp-hpux10.20
The problem is manifested in the message:
Make: Don't know how to make SCANF_OBJECTS.  Stop.
The reason for the problem is that the HP make doesn't deal with the line 
immediately preceding the definition of SCANF_OBJECTS in the Makefile. That 
line end with \ (a backslash) with nothing on the following line. Apparently 
this version of make does not like a blank continuation line. It was easy for 
me to fix the problem by just removing the trailing backslash. The more general 
solution involves the Makefile.in file, and I did not pursue the problem that 
far.
2. This compiler handles 'long long' types, but the configure file will not let 
it be used. I'm not sure why. I see that it checks the version of the compiler 
rather than doing an actual check of 'long long's. Is there some kind of known 
problem with this version of the compiler?
3. There is a bug in the test routine t-sizeinbase in tests/mpz.
The problem is that in routine mpz_fake_bits there is the line:

PTR(z) = (&n) - (SIZ(n)-1);

That's not legal since it points to unallocated space. I suggest the following 
instead:

--------------------------------------------------------
/* Create a fake mpz consisting of just a single 1 bit, with totbits being
   the total number of bits, inclusive of that 1 bit.  */
void
mpz_fake_bits (mpz_ptr z, unsigned long totbits)
{
  static mp_limb_t  n;
  unsigned long     zero_bits, zero_limbs;

  zero_bits = totbits - 1;
  zero_limbs = zero_bits / GMP_NUMB_BITS;
  zero_bits %= GMP_NUMB_BITS;

  SIZ(z) = zero_limbs + 1;
  z->_mp_d[SIZ(z) - 1] = CNST_LIMB(1) << zero_bits;

  ASSERT_ALWAYS (mpz_sizeinbase (z, 2) == totbits);
}


/* This was seen to fail on a GNU/Linux powerpc32 with gcc 2.95.2,
   apparently due to a doubtful value of mp_bases[10].chars_per_bit_exactly
   (0X1.34413509F79FDP-2 whereas 0X1.34413509F79FFP-2 is believed correct).
   Presumably this is a glibc problem when gcc converts the decimal string
   in mp_bases.c, or maybe it's only a function of the rounding mode during
   compilation.  */
void
check_sample (void)
{
  unsigned long  totbits = 198096465;
  int        base = 10;
  size_t     want = 59632979;
  size_t     got;
  mpz_t      z;

  mpz_init2( z, totbits );
  mpz_fake_bits (z, totbits);
  got = mpz_sizeinbase (z, base);
  if (got != want)
    {
      printf ("mpz_sizeinbase\n");
      printf ("  base    %d\n",  base);
      printf ("  totbits %lu\n", totbits);
      printf ("  got     %u\n",  got);
      printf ("  want    %u\n",  want);
      abort ();
    }
  mpz_clear( z );
}
-------------------------------------------------

Thanks very much,

Dean Clamons
Code 7420
Naval Research Lab
Washington, DC 20375
202-767-2732