precision 0 -INF

[hormus]

Description

<?php

ini_set('precision', 0);
ini_set('serialize_precision', 0);
$nif = -INF;
$pif = INF;
$float = -INF;
$string = '';
if(is_float($float)) {
if(INF === INF) {
if($pif === $float) {
$string = 'INF';
} elseif($nif === $float) {
$string = '-INF';
} else {
$string = $float . '';
}
}
}
$error = '';
$string3 = $pif . '';
$arr = preg_split('//', $string3);
$count = count($arr) - 2;
$false = true;
foreach($arr as $key => $value) {
if($key > 0 && $key <= $count) {
$ord = ord($arr[$key]);
if(1 == $key) {
$y = 'I';
$z = 'i';
} elseif(2 == $key) {
$y = 'N';
$z = 'n';
} elseif(3 == $key) {
$y = 'F';
$z = 'f';
}
if(!($ord == ord($y) || $ord == ord($z))) {
$false = false;
break;
}
}
}
if(!($false && $pif === INF)) {
$error = $error . 'INF';
}
$string4 = $nif . '';
$arr2 = preg_split('//', $string4);
$count = count($arr2) - 2;
$false2 = true;
foreach($arr2 as $key2 => $value2) {
if($key2 > 0 && $key2 <= $count) {
$ord = ord($arr[$key2]);
if(1 == $key2) {
$y = '-';
$z = '-';
} elseif(2 == $key2) {
$y = 'I';
$z = 'i';
} elseif(3 == $key2) {
$y = 'N';
$z = 'n';
} elseif(4 == $key2) {
$y = 'F';
$z = 'f';
}
if(!($ord == ord($y) || $ord == ord($z))) {
$false2 = false;
break;
}
}
}

if(!($false2 && $nif === -INF)) {
$error = $error . '-INF';
}
if(!(INF === INF)) {
$error = $error . 'Not === INF';
}

$string2 = $float . '';
var_dump($string, $string2, $error);

?>

Correct:
string(4) "-INF" string(4) "-INF" string(0) ""
Correct without loose precision for special case INF, -INF or NAN.

PHP Version

php 8.1.7

Operating System

Irrelevant

[damianwadley]

Next time, instead of posting a lot of complicated and rather bizarre code, please keep it simple. For example,

Starting with PHP 8.1.0, when precision=0, -INF serializes to "-". Negative numbers are unchanged.

https://3v4l.org/S7mjm

Looks like INF also serializes to "I", and unsurprisingly NAN/-NAN are affected too.

[hormus]

Hi thanks @damianwadley for your contribution sorry but I have passed the 10 lines of code since reading a float number sequence is simple, more than anything else in a string variable we use EG(precision) while var_dump uses PG(precision) although the explicit cast shows a unintentional behavior it is okay to use EG(precision).

PHP uses “raw” IEEE 754 value internally regardless of precision settings.

[cmb69]

The behavioral change has been introduced with 6048481. The relevant code:

php-src/Zend/zend_strtod.c

Lines 4521 to 4529 in 21507ef

	if (decpt == 9999) {
	/*
	* Infinity or NaN, convert to inf or nan with sign.
	* We assume the buffer is at least ndigit long.
	*/
	snprintf(buf, ndigit + 1, "%s%s", (sign && *digits == 'I') ? "-" : "", *digits == 'I' ? "INF" : "NAN");
	zend_freedtoa(digits);
	return (buf);
	}

This could easily be fixed (pass (sign && *digits == 'I') + 4 instead of ndigit + 1), but there is no guarantee that the buffer has a sufficient size (in this case it is called from zend_double_to_str() where the size is sufficient, but the function may be called from somewhere else). If we don't directly fix zend_gcvt(), though, we would need to fix other functions than zend_double_to_str() as well.