[RFC] Floating Point support in GMP extension

Ok, add precision param to gmp_init.

gmp_init ( mixed $number [, int $base = 0 ], int $precision = 0]] )
$a = gmp_init (3); // type float, by precision = 0
$b = gmp_init (3.333, 10, 3); // float
$c = $a + $b; // float

2014-04-22 0:07 GMT+03:00 Johannes Schlüter johannes@schlueters.de:

For backward compatibility in php.ini set default values
​​gmp.default_precision = 0

No. INI setting is not backwards compatibility. An ini setting means
that everybody has to care about it and mind it and either check or
set it.

An alternative is a runtime setting (gmp_set_compatiility(true) ), in
case the change is really important. I haven't looked further but
behavior should not depend on ini. We made the mistake and should learn
not repeat.

johannes

Yes, now I understand better the sense in GMPInt, you are right that there
should be different classes.

We use BCMath, but bulky code BCMath:
bcmul(bcadd($a, $b, $scale), 2, $scale)
I want to write simple:
$a + $b * 2

I'm looking forward to implementing GMPFloat

2014-04-24 11:17 GMT+03:00 Nikita Popov nikita.ppv@gmail.com:

On Mon, Apr 21, 2014 at 10:40 PM, Park Framework <park.framework@gmail.com

wrote:

Concept is simple.
Need, remove GMPInt, leave only GMPFloat.
For backward compatibility in php.ini set default values
​​gmp.default_precision = 0
Anyone who wants to work with float, must specify precision > 0
What do you think about this?

This is not possible, because GMP integers and GMP floats are
fundamentally and inherently different types.

A GMP integer isn't just a GMP float with zero precision. Actually, what I
think you're missing here is that the GMP floating point type is just that:
A floating point type. This means that it is not precise. What you
specify as the GMP float precision actually translates to the size of the
mantissa/significand, not the number of places after the comma.

GMP integers on the other hand are precises. Furthermore a primary
application for GMP integers are computations over cyclic groups, as well
as other number theoretic computations. The operations for these are not a
available for the floating point type (and frankly, wouldn't really make
sense).

Maybe you're looking for the BCMath extension?

Nikita