Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:96439
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: error (pb1.pair.com: domain belski.net from 85.214.73.107 cause and error)
To: "'PHP Internals'" <internals@lists.php.net>
Cc: "'Davey Shafik'" <davey@php.net>,
	"'Joe Watkins'" <krakjoe@php.net>
Date: Tue, 18 Oct 2016 20:46:28 +0200
Message-ID: <073601d2296f$f27135e0$d753a1a0$@belski.net>
MIME-Version: 1.0
Content-Type: text/plain;
	charset="us-ascii"
Content-Transfer-Encoding: 7bit
Thread-Index: AdIpZYnqPN74wDpmQ5SYtCAmxYoAuQ==
Content-Language: en-us
Subject: Constants for better double edge case handling 
From: anatol.php@belski.net ("Anatol Belski")

Hi,

I would like to suggest adding the following constants 

    REGISTER_MAIN_LONG_CONSTANT("PHP_DBL_DIG", DBL_DIG, CONST_PERSISTENT |
CONST_CS);
    REGISTER_MAIN_DOUBLE_CONSTANT("PHP_DBL_MAX", DBL_MAX, CONST_PERSISTENT |
CONST_CS);
    REGISTER_MAIN_DOUBLE_CONSTANT("PHP_DBL_MIN", DBL_MIN, CONST_PERSISTENT |
CONST_CS);
    REGISTER_MAIN_DOUBLE_CONSTANT("PHP_DBL_EPSILON", DBL_EPSILON,
CONST_PERSISTENT | CONST_CS);

The goal of this is to improve the handling of double in the user land. Here
are a couple of usages to illustrate the idea.

The comparison of double values.

$d0 = sin(M_PI/6.0);
$d1 = .5;
var_dump(
	$d0, $d1,
	abs($d0 - $d1),
	$d0 == $d1,
	abs($d0 - $d1) < PHP_DBL_EPSILON
);

float(0.5)
float(0.5)
float(5.5511151231258E-17)
bool(false)
bool(true)


The rounding behavior, max possible value representable by the string
conversion.

$d = .2345234523453245324323465;
echo $d, " ", round($d, 20), " ", round($d, PHP_DBL_DIG);

0.23452345234532 0.23452345234532 0.23452345234533


Producing INF. There's currently no explicit way to produce INF and NAN,
whereby NAN is gettable with sqrt(-1).

echo PHP_DBL_MAX*PHP_DBL_MAX, " ", -PHP_DBL_MAX*PHP_DBL_MAX;
INF -INF


In general, it is more about the possibility to handle the edge cases
properly. While such cases would cause unnecessary overhead and likely a BC
breach with a direct core implementation, they'd be fine to handle in the
scripts where it comes to it. I think, at least DBL_DIG and DBL_EPSILON
should be mapped to the constants, to provide a base for more flexibility.
The change itself is pretty outspoken, so I'm not sure it requires an RFC. I
would target at least master with this. Or 7.1, if RMs are ok.

Regards

Anatol