path: root/UPGRADING

$Id$

PHP X.Y UPGRADE NOTES

1. Backward Incompatible Changes
2. New Features
3. Changes in SAPI modules
4. Deprecated Functionality
5. Changed Functions
6. New Functions
7. New Classes and Interfaces
8. Removed Extensions and SAPIs
9. Other Changes to Extensions
10. New Global Constants
11. Changes to INI File Handling
12. Windows Support
13. Other Changes


========================================
1. Backward Incompatible Changes
========================================

Language changes
================

Changes to variable handling
----------------------------

* Indirect variable, property and method references are now interpreted with
  left-to-right semantics. Some examples:

      $$foo['bar']['baz'] // interpreted as ($$foo)['bar']['baz']
      $foo->$bar['baz']   // interpreted as ($foo->$bar)['baz']
      $foo->$bar['baz']() // interpreted as ($foo->$bar)['baz']()
      Foo::$bar['baz']()  // interpreted as (Foo::$bar)['baz']()

  To restore the previous behavior add explicit curly braces:

      ${$foo['bar']['baz']}
      $foo->{$bar['baz']}
      $foo->{$bar['baz']}()
      Foo::{$bar['baz']}()
      
* The global keyword now only accepts simple variables. Instead of

      global $$foo->bar;

  it is now required to write the following:

      global ${$foo->bar};

* Parentheses around variables or function calls no longer have any influence
  on behavior. For example the following code, where the result of a function
  call is passed to a by-reference function

      function getArray() { return [1, 2, 3]; }

      $last = array_pop(getArray());
      // Strict Standards: Only variables should be passed by reference
      $last = array_pop((getArray()));
      // Strict Standards: Only variables should be passed by reference

  will now throw a strict standards error irregardless of whether parentheses
  are used. Previously no notice was generated in the second case.

* Array elements or object properties that are automatically created during
  by-reference assignments will now result in a different order. For example

      $array = [];
      $array["a"] =& $array["b"];
      $array["b"] = 1;
      var_dump($array);

  now results in the array ["a" => 1, "b" => 1], while previously the result
  was ["b" => 1, "a" => 1];

Relevant RFCs:
* https://wiki.php.net/rfc/uniform_variable_syntax
* https://wiki.php.net/rfc/abstract_syntax_tree

Changes to list()
-----------------

* list() will no longer assign variables in reverse order. For example

      list($array[], $array[], $array[]) = [1, 2, 3];
      var_dump($array);

  will now result in $array == [1, 2, 3] rather than [3, 2, 1]. Note that only
  the **order** of the assignments changed, but the assigned values stay the
  same. E.g. a normal usage like

      list($a, $b, $c) = [1, 2, 3];
      // $a = 1; $b = 2; $c = 3;
  
  will retain its current behavior.

* Empty list() assignments are no longer allowed. As such all of the following
  are invalid:

      list() = $a;
      list(,,) = $a;
      list($x, list(), $y) = $a;

* list() no longer supports unpacking strings (while previously this was only
  supported in some cases). The code

      $string = "xy";
      list($x, $y) = $string;

  will now result in $x == null and $y == null (without notices) instead of
  $x == "x" and $y == "y". Furthermore list() is now always guaranteed to
  work with objects implementing ArrayAccess, e.g.

      list($a, $b) = (object) new ArrayObject([0, 1]);

  will now result in $a == 0 and $b == 1. Previously both $a and $b were null.

Relevant RFCs:
* https://wiki.php.net/rfc/abstract_syntax_tree#changes_to_list
* https://wiki.php.net/rfc/fix_list_behavior_inconsistency

Changes to foreach
------------------

* Iteration with foreach() no longer has any effect on the internal array
  pointer, which can be accessed through the current()/next()/etc family of
  functions. For example

      $array = [0, 1, 2];
      foreach ($array as &$val) {
          var_dump(current($array));
      }

  will now print the value int(0) three times. Previously the output was int(1),
  int(2) and bool(false).

* When iterating arrays by-value, foreach will now always operate on a copy of
  the array, as such changes to the array during iteration will not influence
  iteration behavior. For example

      $array = [0, 1, 2];
      $ref =& $array; // Necessary to trigger the old behavior
      foreach ($array as $val) {
          var_dump($val);
          unset($array[1]);
      }

  will now print all three elements (0 1 2), while previously the second element
  1 was skipped (0 2).

* When iterating arrays by-reference, modifications to the array will continue
  to influence the iteration. However PHP will now do a better job of
  maintaining a correct position in a number of cases. E.g. appending to an
  array during by-reference iteration

      $array = [0];
      foreach ($array as &$val) {
          var_dump($val);
          $array[1] = 1;
      }

  will now iterate over the appended element as well. As such the output of this
  example will now be "int(0) int(1)", while previously it was only "int(0)".

* Iteration of plain (non-Traversable) objects by-value or by-reference will
  behave like by-reference iteration of arrays. This matches the previous
  behavior apart from the more accurate position management mentioned in the
  previous point.

* Iteration of Traversable objects remains unchanged.

Relevant RFC: https://wiki.php.net/rfc/php7_foreach

Changes to parameter handling
-----------------------------

* It is no longer possible to define two function parameters with the same name.
  For example, the following method will trigger a compile-time error:

      public function foo($a, $b, $unused, $unused) {
          // ...
      }

  Code like this should be changed to use distinct parameter names, for example:

      public function foo($a, $b, $unused1, $unused2) {
          // ...
      }

* The func_get_arg() and func_get_args() functions will no longer return the
  original value that was passed to a parameter and will instead provide the
  current value (which might have been modified). For example

      function foo($x) {
          $x++;
          var_dump(func_get_arg(0));
      }
      foo(1);

  will now print "2" instead of "1". This code should be changed to either
  perform modifications only after calling func_get_arg(s)

      function foo($x) {
          var_dump(func_get_arg(0));
          $x++;
      }

  or avoid modifying the parameters altogether:

      function foo($x) {
          $newX = $x + 1;
          var_dump(func_get_arg(0));
      }

* Similarly exception backtraces will no longer display the original value that
  was passed to a function and show the modified value instead. For example

      function foo($x) {
          $x = 42;
          throw new Exception;
      }
      foo("string");

  will now result in the stack trace

      Stack trace:
      #0 file.php(4): foo(42)
      #1 {main}

  while previously it was:

      Stack trace:
      #0 file.php(4): foo('string')
      #1 {main}

  While this should not impact runtime behavior of your code, it is worthwhile
  to be aware of this difference for debugging purposes.

  The same limitation also applies to debug_backtrace() and other functions
  inspecting function arguments.

Relevant RFC: https://wiki.php.net/phpng

Changes to integer handling
---------------------------

* Invalid octal literals (containing digits larger than 7) now produce compile
  errors. For example, the following is no longer valid:
  
      $i = 0781; // 8 is not a valid octal digit!

  Previously the invalid digits (and any following valid digits) were simply
  ignored. As such $i previously held the value 7, because the last two digits
  were silently discarded.

* Bitwise shifts by negative numbers will now throw a warning and return false:

      var_dump(1 >> -1); // bool(false)
      // Warning: Bit shift by negative number

* Left bitwise shifts by a number of bits beyond the bit width of an integer
  will always result in 0:

      var_dump(1 << 64); // int(0)

  Previously the behavior of this code was dependent on the used CPU
  architecture. For example on x86 (including x86-64) the result was int(1),
  because the shift operand was wrapped.

* Similarly right bitwise shifts by a number of bits beyond the bit width of an
  integer will always result in 0 or -1 (depending on sign):

      var_dump(1 >> 64);  // int(0)
      var_dump(-1 >> 64); // int(-1)

Relevant RFC: https://wiki.php.net/rfc/integer_semantics

Changes to string handling
--------------------------

* Strings that contain hexadecimal numbers are no longer considered to be
  numeric and don't receive special treatment anymore. Some examples of the
  new behavior:

      var_dump("0x123" == "291");     // bool(false)     (previously true)
      var_dump(is_numeric("0x123"));  // bool(false)     (previously true)
      var_dump("0xe" + "0x1");        // int(0)          (previously 16)

      var_dump(substr("foo", "0x1")); // string(3) "foo" (previously "oo")
      // Notice: A non well formed numeric value encountered

  filter_var() can be used to check if a string contains a hexadecimal number
  or convert such a string into an integer:

    $str = "0xffff";
    $int = filter_var($str, FILTER_VALIDATE_INT, FILTER_FLAG_ALLOW_HEX);
    if (false === $int) {
        throw new Exception("Invalid integer!");
    }
    var_dump($num); // int(65535)

* Due to the addition of the Unicode Codepoint Escape Syntax for double-quoted
  strings and heredocs, "\u{" followed by an invalid sequence will now result in
  an error:

      $str = "\u{xyz}"; // Fatal error: Invalid UTF-8 codepoint escape sequence

  To avoid this the leading backslash should be escaped:

      $str = "\\u{xyz}"; // Works fine

  However, "\u" without a following { is unaffected. As such the following code
  won't error and will work the same as before:

      $str = "\u202e"; // Works fine

Relevant RFCs:
* https://wiki.php.net/rfc/remove_hex_support_in_numeric_strings
* https://wiki.php.net/rfc/unicode_escape

Other language changes
----------------------

* The yield language construct no longer requires parentheses when used in an
  expression context. It is now a right-associative operator with precedence
  between the "print" and "=>" operators. This can result in different behavior
  in some cases, for example:

      echo yield -1;
      // Was previously interpreted as
      echo (yield) - 1;
      // And is now interpreted as
      echo yield (-1);

      yield $foo or die;
      // Was previously interpreted as
      yield ($foo or die);
      // And is now interpreted as
      (yield $foo) or die;

  Such cases can always be resolved by adding additional parentheses.

  . Removed ASP (<%) and script (<script language=php>) tags.
    (RFC: https://wiki.php.net/rfc/remove_alternative_php_tags)
  . Removed support for assigning the result of new by reference.
  . Removed support for scoped calls to non-static methods from an incompatible
    $this context. See details in https://wiki.php.net/rfc/incompat_ctx.
  . Removed support for #-style comments in ini files. Use ;-style comments
    instead.
  . $HTTP_RAW_POST_DATA is no longer available. Use the php://input stream instead.

Standard library changes
========================

  . call_user_method() and call_user_method_array() no longer exists.
  . ob_start() no longer issues an E_ERROR, but instead an E_RECOVERABLE_ERROR in case an 
    output buffer is created in an output buffer handler.
  . Added hybrid sorting algo zend_sort for better performance.
  . Added stable sorting algo zend_insert_sort.
  . Removed dl() function on fpm-fcgi.

Other
=====

- Curl:
  . Removed support for disabling the CURLOPT_SAFE_UPLOAD option. All curl file
    uploads must use the curl_file / CURLFile APIs.

- Date:
  . Removed $is_dst parameter from mktime() and gmmktime().

- DBA
  . dba_delete() now returns false if the key was not found for the inifile
    handler, too.

- GMP
  . Requires libgmp version 4.2 or newer now.
  . gmp_setbit() and gmp_clrbit() now return FALSE for negative indices, making
    them consistent with other GMP functions.

- Mcrypt
  . Removed deprecated mcrypt_generic_end() alias in favor of
    mcrypt_generic_deinit().
  . Removed deprecated mcrypt_ecb(), mcrypt_cbc(), mcrypt_cfb() and mcrypt_ofb()
    functions in favor of mcrypt_encrypt() and mcrypt_decrypt() with an
    MCRYPT_MODE_* flag.

- Session
  . session_start() accepts all INI settings as array. e.g. ['cache_limiter'=>'private']
    sets session.cache_limiter=private. It also supports 'read_and_close' which closes
    session data immediately after read data.
  . Save handler accepts validate_sid(), update_timestamp() which validates session
    ID existence, updates timestamp of session data. Compatibility of old user defined
    save handler is retained.
  . SessionUpdateTimestampHandlerInterface is added. validateSid(), updateTimestamp()
    is defined in the interface.
  . session.lazy_write(default=On) INI setting enables only write session data when
    session data is updated.

- OpenSSL:
  . Removed the "rsa_key_size" SSL context option in favor of automatically
    setting the appropriate size given the negotiated crypto algorithm.

- PCRE:
  . Removed support for /e (PREG_REPLACE_EVAL) modifier. Use
    preg_reaplace_callback() instead.

- Standard:
  . Removed string category support in setlocale(). Use the LC_* constants
    instead.
  . Removed set_magic_quotes_runtime() and its alias magic_quotes_runtime().

- JSON:
  . Rejected RFC 7159 incompatible number formats in json_decode string -
        top level (07, 0xff, .1, -.1) and all levels ([1.], [1.e1])
  . Calling json_decode with 1st argument equal to empty PHP string or value that 
    after casting to string is empty string (NULL, FALSE) results in JSON syntax error.

- Stream:
  . Removed set_socket_blocking() in favor of its alias stream_set_blocking().

- XSL:
  . Removed xsl.security_prefs ini option. Use XsltProcessor::setSecurityPrefs()
    instead.

========================================
2. New Features
========================================

- Core
  . Added group use declarations.
    (RFC: https://wiki.php.net/rfc/group_use_declarations)
  . Added null coalesce operator (??).
    (RFC: https://wiki.php.net/rfc/isset_ternary)
  . Support for strings with length >= 2^31 bytes in 64 bit builds.
  . Closure::call() method added.
  . Added \u{xxxxxx} Unicode Codepoint Escape Syntax for double-quoted strings
    and heredocs.
  . define() now supports arrays as constant values, fixing an oversight where define() did not support arrays yet const syntax did.
  . Added the comparison operator (<=>), aka the spaceship operator.
    (RFC: https://wiki.php.net/rfc/combined-comparison-operator)

- OpenSSL
  . Added "alpn_protocols" SSL context option allowing encrypted client/server
    streams to negotiate alternative protocols using the ALPN TLS extension when
    built against OpenSSL 1.0.2 or newer. Negotiated protocol information is
    accessible through stream_get_meta_data() output.

========================================
3. Changes in SAPI modules
========================================

- FPM
  . Fixed bug #65933 (Cannot specify config lines longer than 1024 bytes).
  . Listen = port now listen on all addresses (IPv6 and IPv4-mapped).

========================================
4. Deprecated Functionality
========================================

- OpenSSL
  . The "capture_session_meta" SSL context option is now deprecated. Meta
    data concerning active crypto on a stream resource is now accessible
    through the return result from stream_get_meta_data().

========================================
5. Changed Functions
========================================

- parse_ini_file():
- parse_ini_string():
  . Added scanner mode INI_SCANNER_TYPED to yield typed .ini values.
- unserialize():
  . Added second parameter for unserialize function 
    (RFC: https://wiki.php.net/rfc/secure_unserialize) allowing to specify
    acceptable classes: 
    unserialize($foo, ["allowed_classes" => ["MyClass", "MyClass2"]);


========================================
6. New Functions
========================================
- GMP
  . Added gmp_random_seed().

- Standard
  . Added intdiv() function for integer division.
  . Added error_clear_last() function to reset error state.


========================================
7. New Classes and Interfaces
========================================


========================================
8. Removed Extensions and SAPIs
========================================

- sapi/aolserver
- sapi/apache
- sapi/apache_hooks
- sapi/apache2filter
- sapi/caudium
- sapi/continuity
- sapi/isapi
- sapi/milter
- sapi/phttpd
- sapi/pi3web
- sapi/roxen
- sapi/thttpd
- sapi/tux
- sapi/webjames
- ext/mssql
- ext/sybase_ct
For more details see https://wiki.php.net/rfc/removal_of_dead_sapis_and_exts

========================================
9. Other Changes to Extensions
========================================


========================================
10. New Global Constants
========================================

- Core
  . PHP_INT_MIN added.

========================================
11. Changes to INI File Handling
========================================

- Core
  . Removed asp_tags ini directive. Trying to enable it will result in a fatal
    error.
  . Removed always_populate_raw_post_data ini directive.

========================================
12. Windows Support
========================================

- Core
  . Support for native 64 bit integers in 64 bit builds.
  . Support for large files in 64 bit builds.

========================================
13. Other Changes
========================================

- Core
  . Instead of being undefined and platform-dependent, NaN and Infinity will
    always be zero when casted to integer.
  . Calling a method on a non-object no longer raises a fatal error; see
    also: https://wiki.php.net/rfc/catchable-call-to-member-of-non-object.
  . Error messages for zend_parse_parameters, type hints and conversions now always say "integer" and "float" instead of "long" and "double".