Hello Steph,

you forgot to mention the change in inheritance rules that affected
the way interface ArrayAccess can be used. Due to a missing check in our
inheritance rules 5.0.* series allowed to introduce return by reference
in derived classes. Thus it was possible to implement offsetGet($offset)
as function &offsetGet($offset).

right now []++ with overloaded objects is not possible due to a wrong
limitation in the engine. This should be a bug and assigned to andi or
dmitry.

marcus

Saturday, November 19, 2005, 3:20:42 PM, you wrote:

UPGRADE NOTES - PHP 5.1

1. Changes in reference handling
   a. Overview
   b. Code that worked under PHP 4.3, but now fails
   c. Code that was valid under PHP 4.3, but now throws an error
   d. Code that failed under PHP 4.3, but now works
   e. Code that 'should have worked' under PHP 5.0
   f. Warnings that came and went
2. String offset access
3. Reading []
4. instanceof, is_a(), is_subclass_of(), catch
5. Integer values in function parameters
6. Abstract private methods
7. Access modifiers in interfaces
8. Extensions
   a. Extensions that are gone from the PHP core
   b. Class constants in new PHP 5.1 extensions
9. Date/time support
10. Changes in database support
    a. PDO overview
    b. Changes in MySQL support
    c. Changes in SQLite support
11. Further migration information
12. Checking for E_STRICT errors

===============================================================================

1. Changes in reference handling
   ================================

1a. Overview

From the PHP script writer's point of view, the change most likely to impact
legacy code is in the way that references are handled in all PHP versions
post-dating the PHP 4.4.0 release.

Until and including PHP 4.3, it was possible to send, assign or return variables
by reference that should really be returned by value, such as a constant, a
temporary value (e.g. the result of an expression), or the result of a function
that had itself been returned by value, as here:

<?php

$foo = "123";

function return_value() {
global $foo;
return $foo;
}

$bar = &return_value();

?>>

Although this code would usually work as expected under PHP 4.3, in the general
case the result is undefined. The Zend Engine could not act correctly on these
values as references. This bug could and did lead to various hard-to-reproduce
memory corruption problems, particularly where the code base was large.

In PHP 4.4.0, PHP 5.0.4 and all subsequent PHP releases, the Engine was fixed
to 'know' when the reference operation is being used on a value that should
not be referenced. The actual value is now used in such cases, and a warning
is emitted. The warning takes the form of an E_NOTICE in PHP 4.4.0 and up,
and E_STRICT in PHP 5.0.4 and up.

Code that could potentially produce memory corruption can no longer do so.
However, some legacy code might work differently as a result.

1b. Code that worked under PHP 4.3, but now fails

<?php

function func(&$arraykey) {
return $arraykey; // function returns by value!
}

$array = array('a', 'b', 'c');
foreach (array_keys($array) as $key) {
$y = &func($array[$key]);
$z[] =& $y;
}

var_dump($z);

?>>

Running the above script under any version of PHP that pre-dates the reference
fix would produce this output:

array(3) {
[0]=>
&string(1) "a"
[1]=>
&string(1) "b"
[2]=>
&string(1) "c"
}

Following the reference fix, the same code would result in:

array(3) {
[0]=>
&string(1) "c"
[1]=>
&string(1) "c"
[2]=>
&string(1) "c"
}

This is because, following the changes, func() assigns by value. The value
of $y is re-assigned, and reference-binding is preserved from $z. Prior
to the fix, the value was assigned by reference, leading $y to be
re-bound on each assignment. The attempt to bind to a temporary value
by reference was the cause of the memory corruption.

Such code can be made to work identically in both the pre-fix and the
post-fix PHP versions. The signature of func() can be altered to return
by reference, or the reference assignment can be removed from the result
of func().

<?php

function func() {
return 'function return';
}

$x = 'original value';
$y =& $x;
$y = &func();
echo $x;

?>>

In PHP 4.3 $x would be 'original value', whereas after the changes it would
be 'function return' - remember that where the function does not return by
reference, the reference assignment is converted to a regular assignment.
Again, this can be brought to a common base, either by forcing func() to
return by reference or by eliminating the by-reference assignment.

1c. Code that was valid under PHP 4.3, but now throws an error

<?php

class Foo {

function getThis() {
    return $this;
}

function destroyThis() {
    $baz =& $this->getThis();
}

}

$bar = new Foo();
$bar->destroyThis();
var_dump($bar);

?>>

In PHP 5.0.3, $bar evaluated to NULL instead of returning an object.
That happened because getThis() returns by value, but the value here
is assigned by reference. Although it now works in the expected way,
this is actually invalid code which will throw an E_NOTICE under
PHP 4.4 or an E_STRICT under PHP 5.0.4 and up.

1d. Code that failed under PHP 4.3, but now works

<?php

function &f() {
$x = "foo";
var_dump($x);
print "$x\n";
return($a);
}

for ($i = 0; $i < 3; $i++) {
$h = &f();
}

?>>

In PHP 4.3 the third call to var_dump produces NULL, due to the memory
corruption caused by returning an uninitialized value by reference.
This is valid code in PHP 5.0.4 and up, but threw errors in earlier
releases of PHP.

<?php

$arr = array('a1' => array('alfa' => 'ok'));
$arr =& $arr['a1'];
echo '-'.$arr['alfa']."-\n";

?>>

Until PHP 5.0.5, it wasn't possible to assign an array element by
reference in this way. It now is.

1e. Code that 'should have worked' under PHP 5.0

There are a couple of instances of bugs reported under PHP 5.0 prior
to the reference fixes which now 'work'. However, in both cases errors
are thrown by PHP 5.1, because the code was invalid in the first place.
Returning values by reference using self:: now works in the general
case but throws an E_STRICT warning, and although your mileage may
vary when assigning by reference to an overloaded object, you will
still see an E_ERROR when you try it, even where the assignment
itself appears to work.

1f. Warnings that came and went

<?php

function & foo() {
$var = 'ok';
return $var;
}

function & bar() {
return foo();
}

$a =& bar();
echo "$a\n";

?>>

Nested calls to functions returning by reference are valid code under both
PHP 4.3 and PHP 5.1, but threw an unwarranted E_NOTICE or E_STRICT under
the intervening PHP releases.

2. String offset access
   =======================

In PHP, both [] and {} can be used for accessing string offsets, e.g.

php -r "$str = "string"; echo $str{5}";
and
php -r "$str = "string"; echo $str[5]";

would both return the same result. This has led to many complaints over
inconsistent code in the past, and the [] syntax was deprecated some years
ago in an attempt to resolve the issue. However, it appears that [] is the
more popular means of accessing string offsets, so the decision has now
been made to deprecate the {} string offset syntax instead, with the
intention of removing it fully at a later date.

php -r "$str = "string"; echo $str{5}";
will now return an E_STRICT message to that effect in PHP 5.1.0 and up,
and you are strongly discouraged from using this syntax in new code.

3. Reading []
   =============

<?php

class XmlTest {

function test_ref(&$test) {
    $test = "ok";
}

function test($test) { }

function run() {
    $ar = array();
    $this->test_ref($ar[]);
    var_dump($ar);
    $this->test($ar[]);
}

}

$o = new XmlTest();
$o->run();

?>>

This should always have thrown a fatal E_ERROR, because [] cannot be used
for reading in PHP. It is invalid code in PHP 4.4.2 and PHP 5.0.5 upward.

4. instanceof, is_a(), is_subclass_of(), catch
   ==============================================

In PHP 5.0, is_a() was deprecated and replaced by the "instanceof" operator.
There were some issues with the initial implementation of "instanceof", which
relied on __autoload() to search for missing classes. If the class was not
present, "instanceof" would throw a fatal E_ERROR due to the failure of
__autoload() to discover that class. The same behaviour occurred in the
"catch" operator and the is_subclass_of() function, for the same reason.

None of these functions or operators call __autoload() in PHP 5.1, and
the class_exists() workarounds used in code written for PHP 5.0, while
not problematic in any way, are no longer necessary.

5. Integer values in function parameters
   ========================================

With the advent of PHP 5.0, a new parameter parsing API was introduced
which is used by a large number of PHP functions. In all versions of
PHP between 5.0 and 5.1, the handling of integer values was very strict
and would reject non-well formed numeric values when a PHP function
expected an integer. These checks have now been relaxed to support
non-well formed numeric strings such as " 123" and "123 ", and will
no longer fail as they did under PHP 5.0. However, to promote code
safety and input validation, PHP functions will now emit an E_NOTICE
when such strings are passed as integers.

6. Abstract private methods
   ===========================

Abstract private methods were supported between PHP 5.0.0 and PHP 5.0.4,
but were then disallowed on the grounds that the behaviours of 'private'
and 'abstract' are mutually exclusive.

7. Access modifiers in interfaces
   =================================

Under PHP 5.0, function declarations in interfaces were treated in exactly
the same way as function declarations in classes. This has not been the case
since [Oct 13th 2004], at which point only the 'public' access modifier was
allowed in interface function declarations. Since [April 26th 2005], the
'static' modifier has also been allowed. However, the 'protected' and 'private'
modifiers will now throw an E_ERROR, as will 'abstract'. Note that this change
should not affect your existing code, as none of these modifiers makes sense
in the context of interfaces anyway.

8. Extensions
   =============

8a. Extensions that are gone from the PHP core

One of the first things you're likely to notice when you download PHP 5.1 is that
several of the older extensions have disappeared. Those extensions that are still
actively maintained are available in the PHP Extension Community Library (PECL),
at http://pecl.php.net. Windows binaries are built regularly, and you can obtain
the binaries for PECL extensions built against PHP 5.1 from
http://pecl4win.php.net/list.php/5_1.

Extension Alternative/status
========= ========================
ext/cpdf pecl/pdflib
ext/dbx pecl/dbx
ext/dio pecl/dio
ext/fam not actively maintained
ext/ingres_ii pecl/ingres
ext/ircg not actively maintained
ext/mcve pecl/mcve
ext/mnogosearch not actively maintained
ext/oracle ext/oci8 or ext/pdo_oci
ext/ovrimos not actively maintained
ext/pfpro not actively maintained
- alternatives at
http://pecl.php.net/packages.php?catpid=18&catname=Payment
ext/w32api pecl/ffi
ext/yp not actively maintained
sapi/activescript pecl/phpscript

Modules in PECL that are not actively maintained (i.e. have not been supported
for some time, have no active maintainer working on them currently, and do not
have any PECL package releases), are still available in CVS at
http://cvs.php.net/pecl/. However, unreleased PHP modules are by their nature
unsupported, and your mileage may vary when attempting to install or use them.

8b. Class constants in new PHP 5.1 extensions

The Zend Engine 2.1 API allows extension developers to declare class constants
in object oriented extensions. New extensions written for PHP 5.1, including SPL,
PDO, ext/XMLReader and ext/date, have their constants in the format

PDO::CLASS_CONSTANT

rather than in the C format

PDO_CLASS_CONSTANT

in order to minimise pollution of the global namespace in PHP.

Note that the new Date class exists at this point purely to allow the core date
extension to adhere to the above convention, although extended functionality
is planned for the the class in the future.

9. Date/time support
   ====================

Date/time support has been fully rewritten in PHP 5.1, and no longer
uses the system settings to 'know' the timezone in operation. It will
instead utilize, in the following order:

• The timezone set using the date_default_timezone_set() function (if any)
• The TZ environment variable (if non empty)
• The date.timezone ini option (if set)
• "magical" guess (if the operating system supports it)
• If none of the above options succeeds, UTC

To ensure accuracy (and avoid an E_STRICT warning), you will need to define
your timezone in your php.ini using the following format:

date.timezone = Europe/London

The supported timezones are listed, in this format, in the PHP manual at
http://www.php.net/manual/en/timezones.php.

10. Changes in database support
    ==============================

10a. PDO overview

PHP Data Objects (PDO) were introduced as a PECL extension under PHP 5.0,
and became part of the core PHP distribution in PHP 5.1. The PDO extension
provides a consistent interface for database access, and is used alongside
database-specific PDO drivers. Each driver may also have database-specific
functions of its own, but basic data access functionality such as issuing
queries and fetching data is covered by PDO functions, using the driver
named in PDO::__construct().

You are encouraged to use PDO when creating new projects in PHP 5.1. Legacy
code will generally rely on the pre-existing database extensions, which are
still maintained.

Note that the PDO extension, and its drivers, are intended to be built as
shared extensions. This will enable straightforward driver upgrades from
PECL, without the need to rebuild PHP 5.1.

There is more in-depth information about the PDO extension in the manual
at http://www.php.net/manual/ref.pdo.php.

10b. Changes in MySQL support

In PHP 4, MySQL 3 support was built-in. With the release of PHP 5.0 there
were two MySQL extensions, named 'mysql' and 'mysqli', which were designed
to support MySQL < 4.1 and MySQL 4.1 and up, respectively. With the
introduction of PDO, which provides a very fast interface to all the
database APIs supported by PHP, the PDO_MYSQL driver can support any
of the current versions (MySQL 3, 4 or 5) in PHP code written for PDO,
depending on the MySQL library version used during compilation. The
older MySQL extensions remain in place for reasons of back compatibility,
but are not enabled by default.

10c. Changes in SQLite support

In PHP 5.0, SQLite 2 support was provided by the built-in sqlite
extension, which was also available as a PECL extension in PHP 4.3
and PHP 4.4. With the introduction of PDO, the sqlite extension doubles
up to act as a 'sqlite2' driver for PDO; it is due to this that the
sqlite extension in PHP 5.1 has a dependency upon the PDO extension.

PHP 5.1 ships with a number of alternative interfaces to sqlite:

The sqlite extension provides the "classic" sqlite procedural/OO API
that you may have used in prior versions of PHP. It also provides the
PDO 'sqlite2' driver, which allows you to access legacy SQLite 2
databases using the PDO API.

PDO_SQLITE provides the 'sqlite' version 3 driver. SQLite version 3
is vastly superior to SQLite version 2, but the file formats of the
two versions are not compatible.

If your SQLite-based project is already written and working against
earlier PHP versions, then you can continue to use ext/sqlite without
problems, but will need to explicitly enable both PDO and sqlite. New
projects should use PDO and the 'sqlite' (version 3) driver, as this is
faster than SQLite 2, has improved locking concurrency, and supports
both prepared statements and binary columns natively.

11. Further migration information
    ================================

For general information about migrating from PHP 4 to PHP 5, please refer to
the relevant section in the PHP manual at
http://www.php.net/manual/migration5.php.

12. Checking for E_STRICT errors
    ================================

If you only have a single script to check, you can pick up E_STRICT
errors using PHP's commandline lint facility:

php -d error_reporting=4095 -l script_to_check.php

For larger projects, the shell script below will achieve the same task:

#!/bin/sh

directory=$1

shift

These extensions are checked

extensions="php inc"

check_file ()
{
echo -ne "Doing PHP syntax check on $1 ..."

Options:

ERRORS=/www/php/bin/php -d display_errors=1 -d html_errors=0 -d error_prepend_string=" " -d error_append_string=" " -d error_reporting=4095 -l $1 | grep -v "No syntax errors detected"

if test -z "$ERRORS"; then
echo -ne "OK."
else
echo -e "Errors found!\n$ERRORS"
fi

echo
}

loop over remaining file args

for FILE in "$@" ; do
for ext in $extensions; do
if echo $FILE | grep ".$ext$" > /dev/null; then
if test -f $FILE; then
check_file "$FILE"
fi
fi
done
done

Best regards,
Marcus

There is no pecl/phpscript, it's pecl/activescript.
I'm still wary of sending too strong a message about adopting PDO in
this release; there are almost certainly some features missing from
PDO that are present in the traditional drivers. This feature gap
will get smaller after the 5.1 release is made and we can start up
development again.

If you can find a way of putting that in your notes, I'd feel happier.

--Wez.

Guys and guyess,

Hopefully this is the final version of the upgrade notes. Please could you
scroll through it (particularly if you've been involved in developing any of
the affected areas) and get back to me ASAP if you find any misconceptions,
missing information about changes that will affect legacy code, or downright
errors.

These notes need to be 100% complete and agreed by 18:00 hours UTC Sunday -
no inet access for me after this until post-release.

Thanks all, especially Jani for the helpful multi-file script.

• Steph

I suggest a little more damage control is needed...

Steph Fox wrote:

Guys and guyess,

...

4. instanceof, is_a(), is_subclass_of(), catch
   ==============================================

In PHP 5.0, is_a() was deprecated and replaced by the "instanceof" operator.
There were some issues with the initial implementation of "instanceof", which

they way I remember it was that there weren't 'some issues' but that pretty much
everyone but Pierre refused to admit for a very long time that the
instanceof implementation (above all) was useless to the point that a depreciated
is_a() was often peoples only recourse.

interesting to see the turn around, pity that its seems to have been carried out
so low under the radar - (I followed this subject closely and I can't recall any decision
being made in the open regarding reversal of position on this subject)

• an occasional "we we're wrong" might not be a bad idea from a marketing perspective.

relied on __autoload() to search for missing classes. If the class was not
present, "instanceof" would throw a fatal E_ERROR due to the failure of
__autoload() to discover that class. The same behaviour occurred in the
"catch" operator and the is_subclass_of() function, for the same reason.

None of these functions or operators call __autoload() in PHP 5.1, and
the class_exists() workarounds used in code written for PHP 5.0, while

does this mean class_exists() loses its second parameter? or has the default
value changed fom true to false?

either way updates to the docs to reflect the current state would go a long way
to avoiding alot of headaches and confusion (read: 'more irritation').

not problematic in any way, are no longer necessary.

5. Integer values in function parameters
   ========================================

With the advent of PHP 5.0, a new parameter parsing API was introduced
which is used by a large number of PHP functions. In all versions of
PHP between 5.0 and 5.1, the handling of integer values was very strict
and would reject non-well formed numeric values when a PHP function
expected an integer. These checks have now been relaxed to support
non-well formed numeric strings such as " 123" and "123 ", and will
no longer fail as they did under PHP 5.0. However, to promote code
safety and input validation, PHP functions will now emit an E_NOTICE
when such strings are passed as integers.

another annoyance [for the average phper] in the name of a completely
non-quantified drive to 'promote safety and input validation'.

exactly who benefits from this 'promotion'? and what proof of the benefit
is there? given that there cannot be any proof I would suggest such changes
in warning/notice behaviour belongs within E_STRICT where it's definitely
not going to affect anyone who isn't asking explicitly for it.

6. Abstract private methods
   ===========================

Abstract private methods were supported between PHP 5.0.0 and PHP 5.0.4,
but were then disallowed on the grounds that the behaviours of 'private'
and 'abstract' are mutually exclusive.

if one thinks they are mutually exclusive then one can not use them...
but I don't see the reason to change the actually behaviour. a moving target is
worse than an imperfect one, I would have preferred it if this had been left alone,
the only harm it was doing was by offending someone's idea of aesthetics and/or
correctness - given that even when 99.9% of a given group agree on something that something
does not equate to THE TRUTH it seems to me counter productive to impose
the views on those that don't share them (and may possibly have code that
relied on the current behaviour).

7. Access modifiers in interfaces
   =================================

Under PHP 5.0, function declarations in interfaces were treated in exactly
the same way as function declarations in classes. This has not been the case

I really like to be given a solid technical reason why that had to be changed,
even a 'we required X, but we couldn't implement X because of Y so we removed
the capability to Z which made X possible' rather than 'your not allowed to
do it anymore because we think you are wrong'.

since [Oct 13th 2004], at which point only the 'public' access modifier was
allowed in interface function declarations. Since [April 26th 2005], the
'static' modifier has also been allowed.

WTF - static is allowed on interface methods??? to quote some of the comments
made to me by Wez on the subject (he, Andi and others have said the same thing
many times):

"interfaces technically only apply to objects, so no static calls make

sense in an interface context."

and:

"An interface is a public contract that describes the behaviour of an object.

If you're not using interfaces in that way, you're doing something
fundamentally wrong."

so basically what we seem to have is that 'static' is allowed on interface methods
but its 'fundamentally wrong' AND 'protected'/'private' are not allowed on interfaces
because they are 'fundamentally wrong' - that is not consistent (and personally I'd like
back the ability to be 'fundamentally wrong' in all cases not just some), and it becomes
very irritating when the behaviour of interfaces keeps changes with every new version
(major or minor).

regardless on what the current implementation does/allows it might be worth
getting everyone on one line regarding what it should do before making more
[academic] changes.

However, the 'protected' and 'private'
modifiers will now throw an E_ERROR, as will 'abstract'. Note that this change
should not affect your existing code, as none of these modifiers makes sense
in the context of interfaces anyway.

that sounds arrogant. fundamentally speaking the 'does not make sense' argument
is a matter of consensus regardless of any argumentation used to back it up. in ten years
time a totally new/hot/popular development paradigm could contradict this completely -
who is to say what is right and wrong?

PS - Wez' notion of 'fundamentally wrong' is ridiculous. the concept of 'wrong'
is dualistic, dualism exists by way of division - for which there must be something to
divide - ergo dualism is not fundamental, as a result 'fundamentally wrong' is a
contradiction in terms. given that he intends on taking over the world I can
understand the premature attempt to force his views upon others ;-)