php-src/Zend/ZEND_CHANGES
2002-01-14 12:14:18 +00:00

553 lines
19 KiB
Plaintext

Changes in the Zend Engine 2.0
* New Object Model.
The Zend Engine's handling of objects has been completely
changed in order to allow for new features, but also to increase
its performance.
Objects were handled in previous versions like primitive types
(for instance integers and strings). The drawback of this method
is, that semantically the whole object was copied when a
variable was assigned or parameters were passed to a method. The
new approach refers to objects by handle and not by value (one
can think of a handle as an object's ID).
Many PHP programmers aren't even aware of the copying quirks of
the old object model and, therefore, there is a relatively good
chance that the amount of PHP applications that will work out of
the box or after a very small amount of modifications would be
high.
[Not sure if the following will be implemented after all]
To simplify migration, the Zend Engine 2.0 supports an optional
'auto-clone' feature, which performs a cloning of the object
whenever it would have been copied in the Zend Engine 1.0.
Optionally, it emits an E_NOTICE message whenever such an
automatic clone occurs, in order to allow developers to
gradually migrate to the behavior of the Zend Engine 2 (without
automatic clones).
* Object Cloning
The Zend Engine 1.0 offered no way a user could decide what copy
constructor to run when an object is duplicated. During
duplication, the Zend Engine 1.0 did a bitwise copy making an
identical replica of all the object's properties.
Creating a copy of an object with fully replicated properties is
not always the wanted behavior. A good example of the need for
copy constructors, is if you have an object which represents a
GTK window and the object holds the resource of this GTK window,
when you create a duplicate you might want to create a new
window with the same properties and have the new object hold the
resource of the new window. Another example is if your object
holds a reference to another object which it uses and when you
replicate the parent object you want to create a new instance of
this other object so that the replica has its own separate copy.
An object copy is created by calling the object's __clone()
method.
Example:
<?php
$copy_of_object = $object->__clone();
?>
When the developer asks to create a new copy of an object, the
Zend Engine will check if a __clone() method has been defined or
not. If not, it will call a default __clone() which will copy
all of the object's properties. If a __clone() method is
defined, then it will be responsible to set the necessary
properties in the created object. For convenience, the engine
will supply a function that imports all of the properties from
the source object, so that they can start with a by-value
replica of the source object, and only override properties that
need to be changed. [The function hasn't been implemented yet]
Example:
<?php
class MyCloneable {
static $id = 0;
function MyCloneable() {
$this->id = self::$id++;
}
function __clone() {
$this->name = $clone->name;
$this->address = "New York";
$this->id = self::$id++;
}
}
$obj = new MyCloneable();
$obj->name = "Hello";
$obj->address = "Tel-Aviv";
print $obj->id . "\n";
$obj = $obj->__clone();
print $obj->id . "\n";
print $obj->name . "\n";
print $obj->address . "\n";
?>
* Forced deletion of objects.
The Zend Engine 1.0 had no means to force deletion of an object
if there are still references to it. The newly introduced delete
statement calls the object's destructor and frees it even if the
object is referenced by some other places in the engine. Other
references to the deleted object become stale and trying to
access them results in a fatal error.
Note that if you have a user-defined function delete() in an old
script, this script will yield a parser error with the Zend
Engine 2.0, since 'delete' is now a reserved word.
* Namespaces.
The Zend Engine 1.0 provided only three scopes: the global
scope, the class scope and the function scope. All scopes but
classes could contain variables, only the class and global
scopes could contain functions, while only the global scope
could contain constants and classes. This means that all of the
Zend Engine 1.0's scoping methods were inherently limited for
solving symbol name collision problems.
The Zend Engine 2.0 introduces the concept of namespaces to
manage the symbol collision problem by making it possible to
define multiple symbol tables able to contain all types of
symbols. The Zend Engine is aware of a current namespace,
defaulting to the current global one. The current namespace may
be changed on a file-by-file basis. Symbols in other namespaces
than the current one may be referenced using a new namespace
operator. It is possible to "import" symbols from one namespace
into another.
Namespaces and classes are the same with the Zend Engine 2.0,
except that you can't instantiate a namespace with "new". This
essentially also makes a class a namespace, so the scoping rules
for namespaces apply for classes. Some of the consequences of
this are: [Not finalized. Right now we basically have nested
classes so you can instantiate any nested class]
* Classes may contain classes.
Example:
<?php
class DB::MySQL {
var $host = "";
function db_connect($user) {
print "Connecting to MySQL database '$this->host' as $user\n";
}
}
class DB::Oracle {
var $host = "localhost";
function db_connect($user) {
print "Connecting to Oracle database '$this->host' as $user\n";
}
}
$MySQL_obj = new DB::MySQL();
$MySQL_obj->db_connect("Susan");
$Oracle_obj = new DB::Oracle();
$Oracle_obj->db_connect("Barbara");
?>
* Classes may contain constants.
Example:
<?php
class foo {
const hey = "hello";
}
print foo::hey;
?>
* Current namespace's symbol tables are searched first for
constants and functions.
Example:
The following code prints "foobar", not "foo", because
the class constant overrides the "global" constant of
the same name.
<?php
define("foo", "bar");
class FooClass {
const foo = "foobar";
function printFoo() {
print foo;
}
}
?>
* In the scope of a function, the current namespace is that
of the containing class/namespace.
Example:
<?php
class FooClass {
function foo() {
$this->bar();
bar();
}
function bar() {
print "foobar\n";
}
}
$obj = new FooClass;
$obj->foo();
$obj->foo();
?>
This prints "foobar" two times, since a bar() method exists
in the current namespace.
Old code that does not take advantage of namespaces will run
without modifications.
* Unified Constructors.
The Zend Engine allows developers to declare constructor methods
for classes. Classes which have a constructor method call this
method on each newly-created object, so it is suitable for any
initialization that the object may need before it can be used.
With the Zend Engine 1.0, constructor methods were class methods
that had the same name as the class itself. Since it is very
common to call parent constructors from derived classes, the way
the Zend Engine 1.0 worked made it a bit cumbersome to move
classes around in a large class hierarchy. If a class is moved
to reside under a different parent, the constructor name of that
parent changes as well, and the code in the derived class that
calls the parent constructor has to be modified.
The Zend Engine 2.0 introduces a standard way of declaring
constructor methods by calling them by the name __construct().
Example:
<?php
class BaseClass {
function __construct() {
print "In BaseClass constructor\n";
}
}
class SubClass extends BaseClass {
function __construct() {
parent::__construct();
print "In SubClass constructor\n";
}
}
$obj = new BaseClass();
$obj = new SubClass();
?>
For backwards compatibility, if the Zend Engine 2.0 cannot find
a __construct() function for a given class, it will search for
the old-style constructor function, by the name of the class.
Effectively, it means that the only case that would have
compatibility issues is if the class had a method named
__construct() which was used for different semantics.
* Destructors.
Having the ability to define destructors for objects can be very
useful. Destructors can log messages for debugging, close
database connections and do other clean-up work.
No mechanism for object destructors existed in the Zend Engine
1.0, although PHP had already support for registering functions
which should be run on request shutdown.
The Zend Engine 2.0 introduces a destructor concept similar to
that of other object-oriented languages, such as Java: When the
last reference to an object is destroyed the object's
destructor, which is a class method name __destruct() that
recieves no parameters, is called before the object is freed
from memory.
Example:
<?php
class MyDestructableClass {
function __construct() {
print "In constructor\n";
$this->name = "MyDestructableClass";
}
function __destruct() {
print "Destroying " . $this->name . "\n";
}
}
$obj = new MyDestructableClass();
?>
Like constructors, parent destructors will not be called
implicitly by the engine. In order to run a parent destructor,
one would have to explicitly call parent::__destruct() in the
destructor body.
* Exceptions.
The Zend Engine 1.0 had no exception handling. The Zend Engine 2.0
introduces a exception model similar to that of other programming
languages.
Example:
<?php
class MyException {
function __construct($exception) {
$this->exception = $exception;
}
function Display() {
print "MyException: $this->exception\n";
}
}
class MyExceptionFoo extends MyException {
function __construct($exception) {
$this->exception = $exception;
}
function Display() {
print "MyException: $this->exception\n";
}
}
try {
throw new MyExceptionFoo("Hello");
}
catch (MyException $exception) {
$exception->Display();
}
?>
Old code that has no user-defined functions 'catch', 'throw' and
'try' will run without modifications.
* Derefencing objects returned from functions.
Example:
<?php
class Circle {
function draw() {
print "Circle\n";
}
}
class Square {
function draw() {
print "Square\n";
}
}
function ShapeFactoryMethod($shape) {
switch ($shape) {
case "Circle": return new Circle();
case "Square": return new Square();
}
}
ShapeFactoryMethod("Circle")->draw();
ShapeFactoryMethod("Square")->draw();
?>
* Static member variables of static classes can now be
initialized.
Example:
<?php
class foo {
static $my_static = 5;
}
print foo::$my_static;
?>
* Supporting default values for by-reference function parameters.
Example:
<?php
function my_function(&$var = null) {
if ($var === null) {
die('$var needs to have a value');
}
}
?>
Changes in the Zend Engine 1.0
The Zend Engine was designed from the ground up for increased speed,
reduced memory consumption and more reliable execution. We dare say
it meets all of these goals and does so pretty well. Beyond that,
there are several improvements in the language engine features:
* References support.
$foo = &$a; would make $foo and $a be two names to the same
variable. This works with arrays as well, on either side; e.g.,
$foo = &$a[7]; would make $foo and $a[7] be two names to the
same variable. Changing one would change the other and vice
versa.
* Object overloading support.
This feature allows various OO libraries to use the OO notation
of PHP to access their functionality. Right now, no use is made
of that feature, but we'd have a COM module ready by the time
PHP 4.0 is released. A CORBA module would probably follow.
* include() and eval() are now functions, and not statements.
That means they return a value. The default return value from
include() and eval() is 1, so that you can do if (include())
without further coding. The return value may be changed by
returning a value from the global scope of the included file or
the evaluated string. For example, if 'return 7;' is executed in
the global scope of foo.inc, include("foo.inc") would evaluate
to 7.
* Automatic resource deallocation.
Several people have been bitten by the fact that PHP 3.0 had no
concept of reference counting. The Zend Engine adds full
reference counting for every value in the system, including
resources. As soon as a resource is no longer referenced from
any variable, it is automatically destroyed to save memory and
resources. The most obvious example for the advantage in this is
a loop that has an SQL query inside it, something like '$result
= sql_query(...);'. In PHP 3.0, every iteration resulted in
another SQL result-set allocated in the memory, and all of the
result sets weren't destroyed until the end of the script's
execution. With the Zend Engine, as soon as we overwrite an old
result set with a new one, the old result set which is no longer
referenced, is destroyed.
* Full support for nesting arrays and objects within each other,
in as many levels as you want.
* true and false are now constants of type boolean.
Comparing any other value to them would convert that value to a
boolean first, and conduct the comparison later. That means, for
example, that 5==true would evaluate to true (in PHP 3.0, true
was nothing but a constant for the integer value of 1, so
5==true was identical to 5==1, which was false).
* Runtime binding of function names.
This complex name has a simple explanation - you can now call
functions before they're declared!
* Added here-docs support.
* Added foreach.
Two syntaxes supported:
foreach(array_expr as $val) statement
foreach(array_expr as $key => $val) statement
* A true unset() implementation.
A variable or element that is unset(), is now sent to oblivion
in its entirely, no trace remains from it.
* Output buffering support.
Use ob_start() to begin output buffering, ob_end_flush() to end
buffering and send out the buffered contents, ob_end_clean() to
end buffering without sending the buffered contents, and
ob_get_contents() to retreive the current contents of the output
buffer. Header information (header(), content type, cookies) are
not buffered. By turning on output buffering, you can
effectively send header information all throughout your file,
regardless of whether you've emitted body output or not.
* Full variable reference within quoted strings:
${expr} - full indirect reference support for scalar
variables
{variable} - full variable support
For example:
$foo[5]["bar"] = "foobar";
print "{$foo[5]["bar"]}"; // would print "foobar"
* Ability to call member functions of other classes from within
member functions or from the global scope.
You can now, for example, override a parent function with a
child function, and call the parent function from it.
* Runtime information for classes (class name, parent, available
functions, etc.).
* Much more efficient syntax highlighter - runs much quicker,
performs more reliably, and generates much tighter HTML.
* A full-featured debugger has been integrated with the language
(supports breakpoints, expression evaluation, step-in/over,
function call backtrace, and more).
The Zend Engine claims 100% compatability with the engine of PHP
3.0, and is shamelessly lying about it. Here's why:
* Static variable initializers only accept scalar values
(in PHP 3.0 they accepted any valid expression). The impact
should be somewhere in between void and non existent, since
initializing a static variable with anything but a simple
static value makes no sense at all.
* The scope of break and continue is local to that of an
include()'d file or an eval()'d string. The impact should
be somewhat smaller of the one above.
* The return statement no longer works from a require()'d file. It
hardly worked in PHP 3.0, so the impact should be fairly small. If
you want this functionality - use include() instead.
* unset() is no longer a function, but a statement.
* The following letter combination is not supported within
encapsulated strings: "{$". If you have a string that includes
this letter combination, for example, print "{$somevar"; (which
printed the letter { and the contents of the variable $somevar in
PHP 3.0), it will result in a parse error with the Zend Engine.
In this case, you would have to change the code to print
"\{$somevar"; This incompatability is due to the full variable
reference within quoted strings feature added in the Zend
Engine.