Re: [pdf-devel] PDF_FORCE_BIGNUM

[Jeffrey Walton]

Hi Jose,

>   Its not enough to '#define PDF_FORCE_BIGNUMS' in types.h (apparently,
>   some modules do not treat pdf_i64_t opaquely, or do not honor
>   PDF_USE_BUILTIN_64BIT_SUPPORT).
>
> Which modules are those?  We definitely want to fix that.
I think this one is now tracked down. pdf-time.c would crash on
occasion (depending on how I nudged things). I was able to get it to
crash using both built in int64_t and BIGNUM. The first issue appeared
in pdf_time_clear when the pdf_i64_assign_quick was invoked. Once
things got unstable, lots of seg faults followed. pdf_time_clear was
trying to zero out a pdf_i64_t.

I found by adding a true function, pdf_i64_zero() cleared the issue.
I've got too many changes to types.h/types.c (and so little knowledge
of bzr) that I can't offer a patch. Fortunately, the changes were
small so the code is below.

Jeff

// pdf-time.c
/* Clear contents of the pdf_time_t object */
pdf_status_t
pdf_time_clear (pdf_time_t time_var)
{
  PDF_ASSERT_BASE (NULL != time_var);
  if (NULL == time_var)
    {
      return PDF_ERROR;
    }

  pdf_status_t status = PDF_OK;

  /* Set time as January 1st, 1970 */
  status = pdf_i64_zero (&(time_var->seconds));
  PDF_ASSERT_BASE (status == PDF_OK);

  /* UTC */
  time_var->gmt_offset = 0;

  return status;
}

// pdf-type.h
INLINE pdf_status_t pdf_i64_zero (pdf_i64_t* bignum);

// pdf-types.h
INLINE pdf_status_t
pdf_i64_zero (pdf_i64_t * bignum)
{
  PDF_ASSERT_BASE (NULL != bignum);
  if (NULL == bignum)
    {
      return PDF_ERROR;
    }

#if defined(PDF_USE_BUILTIN_64BIT_SUPPORT)
  *bignum = 0;
#else
  bignum->high = 0;
  bignum->low = 0;
#endif

  return PDF_OK;
}

On Thu, Feb 4, 2010 at 1:26 PM,  <address@hidden> wrote:
>
> Hi Jeff.
>
>   In config.h near line 790, there's the following comment:
>
>   /* Force the usage of the 64bits bignum implementation even if the system
>      support a suitable native type */
>   /* #undef PDF_FORCE_BIGNUMS */
>
>  [SNIP]