PHP 8: Process & Fixing Insanity

POINT TECHNIQUE
PHP 8: Process
& Fixing Insanity

George Peter Banyard
● Mathematics student at Imperial
College London
● Currently hired by The Coding Machine
to work on php-src, i.e. the PHP Engine
● Find me on Twitter and elsewhere
@Girgias
● https://gpb.moe
2
Me at 3AM realising I need a photo for the presentation

How I got involved with the PHP project
• French translation PHP documentation
• PHP internal mailing list
• Proposed the “Deprecate PHP Short open tags” PHP RFC
• Started tinkering with PHP’s source code
• PoC: declare statement which promote warnings to Exceptions
• Made multiple patches
• php-src VCS karma granted
3

PHP a truly open source language: PHP RFCs
A PHP RFC is a proposal which aﬀects the PHP engine:
Behaviour change, feature addition, deprecations, extensions
Anyone can propose a PHP RFC, from QoL to Advanced feature
• Minimum 2 week discussion period
• 2 week voting period
• Everyone with a PHP VCS account can vote (1000+ people!)
PHP 8 is 30+ accepted RFCs from over 15 people
Many more which have been declined
See them all at: https://wiki.php.net/rfc
4

QoL
Trailing commas in parameter list and closure use list
$longArgs_longVars = function (
$longArgument,
$longerArgument ,
$muchLongerArgument ,
) use (
$longVar1,
$longerVar2,
$muchLongerVar3 ,
) {
// body
};
5
Allowed!

QoL
::class notation on variables
$class = $object::class;
// Identical to
$class = get_class($object);
DateTimeInterface::createFromInterface
public static function createFromInterface (DateTimeInterface $object): DateTime
public static function createFromInterface (DateTimeInterface $object): DateTimeImmutable
6

QoL
Non-capturing catches
try {
doStuffWhichMightThrow ();
} catch (Exception) { // The intention is clear: exception details are irrelevant
echo "Some error message" ;
}
Null safe operator ?->
$country = null;
if ($session !== null) {
$user = $session->user;
if ($user !== null) {
$address = $user->getAddress();
if ($address !== null) {
$country = $address->country;
}
}
}
// do something with $country
7
$country =
$session?->user?->getAddress()?-> country;
// do something with $country
Becomes

QoL
JSON Extension always enabled
Stable sorting
Functions:
function str_contains(string $haystack, string $needle): bool {}
function str_starts_with (string $haystack, string $needle): bool {}
function str_ends_with(string $haystack, string $needle): bool {}
function fdiv(float $dividend, float $divisor): float {} // IEEE 754 semantics
8

Just In Time Compiler (JIT)
• Compiles PHP code into machine language for x86 (Intel/AMD)
• Performance gain for CPU bounded (i.e. math)
• ⚠ May lead to performance loss on more “traditional” code
Part of OpCache:
• opcache.jit=on
• opache.jit_buﬀer_size=128M
10

Match
Similar to the switch statement
Major diﬀerences:
• Exhaustive, if value not matched throws UnhandledMatchError
• Uses strict comparison (===)
• No fallthrough
• Returns a value
• Match condition can be any expression
• Match statement must be a single expression (like arrow functions)
11

Match: a simple example
PHP 7.x
switch (1) {
case 0:
$result = 'Foo';
break;
case 1:
$result = 'Bar';
break;
case 2:
case 3:
$result = 'Baz';
break;
default:
$result = 'FooBar';
break;
}
echo $result;
//> Bar
PHP 8.0
echo match (1) {
0 => 'Foo',
1 => 'Bar',
2, 3 => 'Baz',
default => 'FooBar',
};
//> Bar
12

Match: complex example
echo match ($x) {
3, 5 => throw new Exception('Cannot be 3 or 5'),
foo() => "foo",
$this->bar() => "bar() check", // bar() isn't called if foo() matched with $x
};
If nothing is matched a UnhandledMatchError is thrown
13
Possible because throw is an
expression in PHP 8

Constructor promotion
PHP 7.x
class Point {
public float $x;
public float $y;
public float $z;
public function __construct(
float $x = 0.0,
float $y = 0.0,
float $z = 0.0,
) {
$this->x = $x;
$this->y = $y;
$this->z = $z;
}
}
PHP 8.0
class Point {
public function __construct(
public float $x = 0.0,
public float $y = 0.0,
public float $z = 0.0,
) { }
}
14

Attributes: syntax TBD
Using doc comments
/** @Entity */
class User {
/**
* @Id
* @Column(type="integer")
* @GenerateValue
*/
private $id;
}
Attributes with Rust-like syntax
use DoctrineORMAttributes as ORM;
#[ORMEntity]
class User {
#[ORMId]
#[ORMColumn("integer")]
#[ORMGenerateValue]
private $id;
}
15
Class name
Constructor argument

Attributes: syntax TBD
Hack-style
<<ORMEntity>>
class User {
<<ORMId>>
<<ORMColumn("integer")>>
<<ORMGenerateValue>>
private $id;
}
@@-style
@@ORMEntity
class User {
@@ORMId
@@ORMColumn("integer")
@@ORMGenerateValue
private $id;
}
16

Attributes
$rc = new ReflectionClass (User::class);
foreach ($rc->getAttributes() as $attr) {
var_dump($attr->getName());
// => "DoctrineORMAttributesEntity"
var_dump($attr->getArguments());
// => []
var_dump($attr->newInstance());
// object(DoctrineORMAttributesEntity) ...
}
17
Attribute validation happens here

Named arguments
// Using positional arguments:
array_fill(0, 100, 50);
// Using named arguments:
array_fill(start_index: 0, num: 100, value: 50);
array_fill(value: 50, num: 100, start_index: 0);
// Skip default values
htmlspecialchars ($string, double_encode: false);
// Same as
htmlspecialchars ($string, ENT_COMPAT|ENT_HTML401, 'UTF-8', false);
Combined with Constructor promotion: Object Initialization
Named arguments work with attributes
Renaming internal function args and matching the documentation
18

Type System Improvements
Union types, mixed, static return, internal functions, opaque objects
19

Existing union types
• ?Type = Type|null
• iterable = Traversable|array
20

Union types
PHP 7.4
class Number {
/** @var int|float */
private $number;
/** @return float|int */
public function getNumber()
{
return $this->number;
}
/** @param float|int $number */
public function setNumber($number): void
{
$this->number = $number;
}
}
PHP 8.0
class Number {
private float|int $number;
public function getNumber(): float|int
{
return $this->number;
}
public function setNumber(float|int $number): void
{
$this->number = $number;
}
}
21

Union types: exotic types
function shell_exec(string $command): string|false|null {}
22
Common in
standard library
?Type is a shorthand to
Type|null

Union types: handling in weak mode
In weak mode the handling is a bit tricky:
• Type is part of the union
• Scalars are coerced to int, float, string, bool in order of preference
23

Union types: weak and strict mode
declare(strict_types=0);
function test(int|float|bool $arg) {
var_dump($arg);
}
test(45); // int(45)
test(45.8); // float(45.8)
test("45"); // int(45)
test("45.8"); // float(45.8)
test(""); // bool(false)
test("X"); // bool(true)
test([]); // TypeError
declare(strict_types=1);
function test(int|float|bool $arg) {
var_dump($arg);
}
test(45); // int(45)
test(45.8); // float(45.8)
test("45"); // TypeError
test("45.8"); // TypeError
test(""); // TypeError
test("X"); // TypeError
test([]); // TypeError
24

mixed
Distinguish between:
• Type is missing because it wasn’t yet added
• This function accepts/returns anything
25
function var_dump(mixed $value, mixed ...$value): void {}
function unserialize(string $value, array $options = []): mixed {}

mixed
// For argument types:
// No type is equivalent to mixed
class Parent {
public function foo(mixed $arg) {}
}
class Child extends Parent {
public function foo($arg) {}
}
// For return types:
// No type means mixed|void
class Parent {
public function foo(): mixed {}
}
class Child extends Parent {
public function foo() {}
}
26
Forbidden: widening type

Static return type
// Named constructor
class ParentClass {
public static function createFromWhatever($whatever): static {
return new static($whatever);
}
}
class ChildClass extends ParentClass {}
var_dump(ChildClass::createFromWhatever(null));
//object(ChildClass)#1 (0) {
//}
27

Static return type
// Named constructor
class ParentClass {
public static function createFromWhatever($whatever): static {
return new self($whatever);
}
}
class ChildClass extends ParentClass {}
var_dump(ChildClass::createFromWhatever(null));
// TypeError: ParentClass::createFromWhatever(): Return value must be of type
ChildClass, ParentClass returned
28
self vs static

PHP Stubs
PHP stubs specify signature for internal functions/methods
● Becomes the source of truth
● Generates C code function registration with arg info
● Data is now available through Reflection
○ ReflectionFunction::getReturnType();
○ ReflectionParameter::getType();
○ ReflectionParameter::getDefaultValue();
29

PHP Stubs: what it means for you
// Stub
class DateTime implements DateTimeInterface
{
/** @return DateTime */
public function add(DateInterval $interval) { }
}
// Your code
class MyDateTime extends DateTime
{
public function add(DateInterval $interval): DateTime { }
}
30
Now allowed!
A real return type would force all
extending classes to specify it due
to covariance, BC Break!

Resource to object conversion
Why?
• Better type safety
• Better internal support
• Long term: convert all resources to objects
• That’s why there is no ‘resource’ type
• Opaque objects with no API
• Oriented Object API might be added later
31

Notable conversions in PHP 8
• GdImage
• Curl
• ZIP
• Inflate/Deflate from Zlib
• OpenSSL (W.I.P.)
32

PHP 7.x only
$curlHandler = curl_init($url);
if (!is_resource($curlHandler))
{
throw new Exception();
}
PHP 7.x and PHP 8.0
$curlHandler = curl_init($url);
if ($curlHandler !== false) {
throw new Exception();
}
33
This is an Object in case of
success or false on failure

Notable Backward
Compatibility (BC) breaks

Notable BC breaks: error handling
• E_ALL by default
• Might need to adjust error handler and INI settings
• PDO now throws exception by default
• Might break application not using PDO’s throw mode
• Can de disabled by passing: PDO::ERRMODE_SILENT during construction
35

Notable BC breaks: locale
• Default locale is now always the C locale
• Float to string conversions are now locale independent
36

Notable BC breaks: locale
PHP 7.4
setlocale(LC_ALL, "de_DE");
$f = 3.14;
(string) $f; // 3,14
strval($f); // 3,14
print_r($f); // 3,14
var_dump($f); // float(3,14)
debug_zval_dump($f); // float(3,14)
settype($f, "string"); // 3,14
implode([$f]); // 3,14
xmlrpc_encode($f); // 3,14
PHP 8.0
setlocale(LC_ALL, "de_DE");
$f = 3.14;
(string) $f; // 3.14
strval($f); // 3.14
print_r($f); // 3.14
var_dump($f); // float(3.14)
debug_zval_dump($f); // float(3.14)
settype($f, "string"); // 3.14
implode([$f]); // 3.14
xmlrpc_encode($f); // 3.14
37

Notable BC breaks: warning promotion
• Consistent TypeError for internal functions
• This allows the PHP Stubs to exist
• LSP errors always throw (type safety)
• Engine notices and warnings reclassified
• Complete list at: https://wiki.php.net/rfc/engine_warnings
• ValueErrors
• A good attempt to list all aﬀected functions:
https://nahan.fr/nouvelle-erreur-valueerror-dans-php-8/
These only aﬀect programming errors
Handling the error is nonsensical, code needs to be fixed
38

Warning promotion: what benefits?
var_dump(strlen([]));
// Warning: strlen() expects parameter 1 to be string, array given
// NULL
function strlen(string $str): int|null {}
var_dump(array_fill(0, -100, 'foobar'));
// Warning: array_fill(): Number of elements can't be negative
// bool(false)
function array_fill(int $start_key, int $num, mixed $val): array|false {}
39

Warning promotion: what benefits?
var_dump(strlen([]));
// Fatal error: Uncaught TypeError: strlen():
// Argument #1 ($str) must be of type string, array given
function strlen(string $str): int {}
var_dump(array_fill(0, -100, 'foobar'));
// Fatal error: Uncaught ValueError: array_fill():
// Argument #2 ($num) must be greater than or equal to 0
function array_fill(int $start_key, int $num, mixed $val): array {}
40

Warning promotion: what stays the same
var_dump(fopen("does_not_exist.txt", "r'"));
// Warning: fopen(does_not_exist.txt): Failed to open stream: Operation not
permitted
// bool(false)
41
Actually I lied, this is capitalized in PHP 8.0

Numeric strings
Numeric strings: “12”, “15.6”, “5e6”, “ 10”
Leading-numeric strings: “10 ”, “14abc”
Non-numeric strings: “string”
42
Emit E_NOTICE "A non well formed
numeric value encountered"
Leading white-spaces are
considered more numeric
than trailing white-spaces

Numeric strings
Where are numeric strings used?
• Type declarations in weak mode
• is_numeric() function
• String to string comparisons
• Arithmetic and bitwise operations
• Increment/Decrement operation (++ and --)
• String offsets
• NOT array offsets, they have different semantics
43

Numeric strings: bringing back sanity
All details: Saner Numeric Strings RFC:
https://wiki.php.net/rfc/saner-numeric-strings
TL;DR:
• E_NOTICE "A non well formed numeric value encountered" to
E_WARNING "A non-numeric value encountered" with value
preserved
• Other E_WARNINGs promoted to TypeError
44

Numeric strings: bringing back sanity
PHP 7.4
function foo(int $i) { var_dump($i); }
foo(" 123"); // int(123)
foo("123 "); // int(123) with E_NOTICE
foo("123abc"); // int(123) with E_NOTICE
var_dump(123 + "123 ");// int(246) with E_NOTICE
var_dump(123 + "123abc");// int(246) with E_NOTICE
var_dump(123 + "str"); // int(123) with E_WARNING
var_dump("123" == " 123"); // bool(true)
var_dump("123" == "123 "); // bool(false)
var_dump("123" == "123abc"); // bool(false)
PHP 8.0
function foo(int $i) { var_dump($i); }
foo(" 123"); // int(123)
foo("123 "); // int(123)
foo("123abc"); // int(123) with E_WARNING
var_dump(123 + "123 ");
var_dump(123 + "123abc"); // int(246) with E_WARNING
var_dump(123 + "str"); // TypeError
var_dump("123" == " 123"); // bool(true)
var_dump("123" == "123 "); // bool(true)
var_dump("123" == "123abc"); // bool(false)
45

String to number comparisons
$validValues = ["foo", "bar", "baz"];
$value = 0;
var_dump(in_array($value, $validValues));
// bool(true) WTF???
var_dump(0 == "foo");
// bool(true) WTF???
Are always performed numerically
46

Other BC breaks:
• Precedence of the concatenation operator .
• Default values before mandatory ones are now deprecated
• final on private methods (except on __construct())
• Treat namespaces names as a single token
• Full list in the UPGRADING document
https://github.com/php/php-src/blob/master/UPGRADING
48

Performance gain
• strtolower() now uses a SIMD implementation when using the "C"
LC_CTYPE locale
• array_slice() on an array without gaps will no longer scan the
whole array to find the start oﬀset. This may significantly reduce
the runtime of the function with large oﬀsets and small lengths
• Removal of old style constructor in PHP 8.0 seems to yield 20%
performance increase in WordPress
https://idk.dev/improving-performance-of-php-for-arm64-and-im
pact-on-amazon-ec2-m6g-instances/
49

Micro-optimizations
DISCLAIMER:
I have not tested the performance of the
following changes, these are just tidbits I
remember which have been committed
to php-src
50

Micro-optimizations
• Optimize $x === null into is_null($x)
https://github.com/php/php-src/commit/36afe4e39ec724eab19c54c1492db030721d0ec1
• Yes is_null() generated a “better” TYPE_CHECK OpCode
• Optimize instanceof checks
https://github.com/php/php-src/commit/c858d17f06179aa25f6e8aa06965313fd519d8e9
https://github.com/php/php-src/commit/c63a0e005abe4b00ab097dc47ca53d20788a6361
• Optimize int === int/float === float
https://github.com/php/php-src/commit/e8525c2f68d938162af055fd843239778a13565d
• Weirdly enough by copying the implementation of ==
• Optimize is_scalar($x) into a TYPE_CHECK opcode
https://github.com/php/php-src/commit/937fa6d9e2f38b5e70a2a510b87ce0be4839a404
• Optimize return type checking
https://github.com/php/php-src/commit/0bac7854779595603e0431521c60744665641734
51

PHP 8: Process & Fixing Insanity

More Related Content

What's hot (20)

Similar to PHP 8: Process & Fixing Insanity (20)

Recently uploaded (20)

PHP 8: Process & Fixing Insanity