make bcpowmod stricter by not returning false, instead throw exception

[vv12131415]

There are 2 points on this

1. bcpowmod is the only function left on bcmath lib that may return false on failure. Let's fix that
2. next release is major, so we can break things a littlebit

The only one question that I have for you guys - what kind of exception should I throw when we have negative exponent?
i left todo in code, will remove after I get answers

[vv12131415]

I see that CI failed, but I don't understand what exactly failed

[nikic]

You need to update

php-src/ext/opcache/Optimizer/zend_func_info.c

Line 792 in 6bc375f

F1("bcpowmod", MAY_BE_FALSE | MAY_BE_STRING),

[vv12131415]

@nikic by the way.
Since bcpow can have exponent as negative number, I think throwing exception is kind of workaround.

I'm curious why I can't change this function

php-src/ext/bcmath/libbcmath/src/raisemod.c

Lines 62 to 63 in 6bc375f

	int
	bc_raisemod (bc_num base, bc_num expo, bc_num mod, bc_num *result, int scale)

to look like this

int
bc_raisemod (bc_num base, bc_num expo, bc_num mod, bc_num *result, int scale)
{
  bc_num power, exponent, modulus, temp;

  /* Check for correct numbers. */
  if (bc_is_zero(mod)) return -1;

  power = bc_copy_num (base);
  exponent = bc_copy_num (expo);
  modulus = bc_copy_num (mod);
  temp = bc_copy_num (BCG(_one_));
  bc_raise(power, exponent, &temp, scale);
  bc_free_num (&power);
  bc_free_num (&exponent);
  bc_modulo(temp, modulus, &temp, scale);
  /* Assign the value. */
  bc_free_num (&modulus);
  bc_free_num (result);
  *result = temp;
  return 0;	/* Everything is OK. */

Am I missing something?
I have failed test ext/bcmath/tests/bcpowmod.phpt
this one

echo bcpowmod("10", "2147483648", "2047");

gives error

Fatal error: Allowed memory size of 134217728 bytes exhausted (tried to allocate 1052672 bytes)

I know that it's memory leak somewhere. Going to try to debug it myself, but appreciate if you can help with this

[vv12131415]

Ok, I kind of figured out that it's bcpow
I've tried to do

echo bcpow("10", "2147483648");

and it also failed

---

upd: valgrind haven't found any memory leaks.

Those algorithms implmenented in bcpow and bcpowmod implemented differently

[nikic]

@vladyslavstartsev The point of modular exponentiation is exactly that it does not require computing the full exponentiation. This is why bcpowmod exists, instead of letting people use bcpow + bcmod.

I don't think it makes sense to support negative exponents in this function. It will likely only confuse people, because it would interpret the negative exponent in the sense of a division, not in the sense of a modular multiplicative inverse, which is what people usually mean when talking about modular exponentiation with negative exponents.

[nikic]

Looks fine, maybe @kocsismate wants to check this aligns with error conventions.

[kocsismate]

@nikic It looks all fine for me as well :)

[carusogabriel]

Ok, wait, what my name is doing here?

[kocsismate]

Oops, too late, sorry :S You can remove it from master though :)