bcdiv using very small float with scientific notation cause “Division by zero” error

Question

Using bcdiv, i can't divide with small float using scientific notation :

Working code :

bcscale(30);
$a = '1' ;
$b = '0.00000001';
$result = bcdiv($a, $b);
var_dump($result);

Results in :

string(20) "100000000.0000000000"

Non-working code :

bcscale(30);
$a =  '1' ;
$b =  '1e-8';
$result = bcdiv($a, $b);
var_dump($result);

Results in :

Warning: bcdiv() [function.bcdiv]: Division by zero in C:\wamp\www\utilitaires\test_bcdiv.php on line XX NULL

How can i do this division properly, with the less precision loss ?

Answer 1

That is because, actually, bcmath doesn't support scientific notation. It isn't mentioned in manuals, but as you can see, in it's implementation argument conversion is used, it's named php_str2num:

static void php_str2num(bc_num *num, char *str TSRMLS_DC)
{
    char *p;

    if (!(p = strchr(str, '.'))) {
        bc_str2num(num, str, 0 TSRMLS_CC);
        return;
    }

    bc_str2num(num, str, strlen(p+1) TSRMLS_CC);
}

and so bc_str2num:

bc_str2num (bc_num *num, char *str, int scale TSRMLS_DC)
{
  int digits, strscale;
  char *ptr, *nptr;
  char zero_int;

  /* Prepare num. */
  bc_free_num (num);

  /* Check for valid number and count digits. */
  ptr = str;
  digits = 0;
  strscale = 0;
  zero_int = FALSE;
  if ( (*ptr == '+') || (*ptr == '-'))  ptr++;  /* Sign */
  while (*ptr == '0') ptr++;            /* Skip leading zeros. */
  while (isdigit((int)*ptr)) ptr++, digits++;   /* digits */
  if (*ptr == '.') ptr++;           /* decimal point */
  while (isdigit((int)*ptr)) ptr++, strscale++; /* digits */
  if ((*ptr != '\0') || (digits+strscale == 0))
    {
      *num = bc_copy_num (BCG(_zero_));
      return;
    }

  /* Adjust numbers and allocate storage and initialize fields. */
  strscale = MIN(strscale, scale);
  if (digits == 0)
    {
      zero_int = TRUE;
      digits = 1;
    }
  *num = bc_new_num (digits, strscale);

  /* Build the whole number. */
  ptr = str;
  if (*ptr == '-')
    {
      (*num)->n_sign = MINUS;
      ptr++;
    }
  else
    {
      (*num)->n_sign = PLUS;
      if (*ptr == '+') ptr++;
    }
  while (*ptr == '0') ptr++;            /* Skip leading zeros. */
  nptr = (*num)->n_value;
  if (zero_int)
    {
      *nptr++ = 0;
      digits = 0;
    }
  for (;digits > 0; digits--)
    *nptr++ = CH_VAL(*ptr++);

  /* Build the fractional part. */
  if (strscale > 0)
    {
      ptr++;  /* skip the decimal point! */
      for (;strscale > 0; strscale--)
    *nptr++ = CH_VAL(*ptr++);
    }
}

-not hard to see that it will fail on scientific notation (well-commented). Perhaps documentation needs to be updated (to mention that implicitly).

Possible solution will be to convert your string to plain view before applying bcmath functions