PHP RFC: Constructor behaviour of internal classes

• Version: 1.0
• Date: 2015-03-01
• Author: Dan Ackroyd, Danack@php.net
• Status: Under Discussion
• First Published at: https://wiki.php.net/rfc/internal_constructor_behaviour

Several internal classes in extensions that ship with PHP exhibit some highly surprising behaviour. This RFC has two aims:

i) to make the behaviour of these classes behave more consistently with the behaviour that most people would expect them to have.

ii) setting the coding standards that should be followed by internal classes.

If other internal classes that are not listed in this RFC are found to have behaviour that is not consistent with the behaviour listed in this RFC they should be treated as bugs and fixed at the next appropriate release.

For the internal classes:

* finfo * PDO * Collator * IntlDateFormatter * MessageFormatter * NumberFormatter * ResourceBundle * IntlRuleBasedBreakIterator

when their constructor is called with parameters that cannot be used to create a valid instance of the class, the constructor returns null. This is inconsistent with classes which are declared in userland which are not capable of returning null value, but instead have to indicate an error by throwing an exception.

Having a different behaviour for internal classes is inconsistent and highly suprising when users encounter it:

``` $mf = new MessageFormatter('en_US', '{this was made intentionally incorrect}'); if ($mf === null) {

  echo "Surprise!";

} ```

This RFC proposes setting a standard that internal classes should either return a valid instance of the class or throw an exception. Also for the classes listed for them to be changed in PHP 7 so that they follow this behaviour.

The classes IntlDateFormatter, MessageFormatter, NumberFormatter, ResourceBundle and IntlRuleBasedBreakIterator all have a static factory method that can create an instance of the class, as well as the constructor. The factory methods do not currently throw an exception when invalid arguments are passed to it. It is the position of this RFC that it is correct for those factory methods return either null or a valid usuable instance of the class.

PHP is a multi-paradigm programming language which support writing code in either procedural or OO style. Having the factory methods behave like this is perfectly consistent with userland code.

``` class NumberFormatter {

  static function create($locale , $style , $pattern = null) {
      try {
          return new NumberFormatter($locale, $style, $pattern);
      }
      catch(NumberFormatterException $nfe) {
          trigger_error($nfe->getMessage(), E_USER_ERROR);
          return null;
      }
  }

} ```

There is also a function `finfo_open` which is also a procedural way of instantiating an finfo object. This function would not be modified.

The constructors of several classes check the parameters that they are given and just give a warning when they are not acceptable. This leaves the object instantiated but in an unusable state.

``` <?php Reflection function is not meant to accept an array $reflection = new ReflectionFunction([]); Outputs Warning: ReflectionFunction::construct() expects parameter 1 to be string, array given in... var_dump($reflection->getParameters()); Fatal error: ReflectionFunctionAbstract::getParameters(): Internal error: Failed to retrieve the reflection object in.. ``` This RFC proposes that this behaviour should be changed so that the constructor should throw an exception if the class cannot be created in usuable state i.e. convert the current warning to an exception of the appropriate. The list of classes that show this behaviour is: * UConverter * SplFixedArray * ReflectionFunction * ReflectionParameter * ReflectionMethod * ReflectionProperty * ReflectionExtension * ReflectionZendExtension * Phar * PharData * PharFileInfo ==== Constructor generates fatal error ==== The class PDORow gives a fatal error when a user attempts to instantiate it. ``` $foo = new PDORow(); PHP Fatal error: PDORow::construct(): You should not create a PDOStatement manually in test.php on line 5 ```

Fatal errors should only be used for fatal errors. This RFC proposes that the constructor for PDORow should be changed to throw an appropriate exception rather than giving a fatal error.

I don't believe that most users are even aware that the classes listed can show this behaviour. For these people none of the changes would be a BC break.

For people who are aware that the constructor can fail there would be a small BC break.

For the classes that have a factory creation method the code that currently tests against the constructor returning null:

``` $mf = new MessageFormatter('en_US', $thisMightBeInvalid); if ($mf === null) {

  // error handling code

} ```

could be changed to using the factory method:

``` $mf = MessageFormatter::create('en_US', $thisMightBeInvalid); if ($mf === null) {

  // error handling code

} ```

For the other classes which do not have an equivalent procedural method which will still return null, the user would need to wrap the call to the constructor in a try/catch block.

For the classes where the instance is created but in an unusable state there would theoretically be a small BC break in that the location where the code breaks for invalid parameters would be changed.

These changes would be for PHP 7

None.

The standard of either returning a usuable instance or throwing an exception in an objects constructor would apply to all extensions that ship with the official PHP release. Other extensions (e.g. those that are hosted on PECL) may wish to follow this standard, but it is not required.

If anyone feels strongly about splitting the vote into separate parts, please say so.

The vote requires 2/3 to pass.

Should the standard paradigm for constructors for internal objects to be to either return a usable instance of a class on success, and throw an exception if they encounter an error, and should the code for the classes listed below be modified to follow this standard?

These clases will be corrected by making the constructor throw an exception rather than return null if the construction of the object fails.

* finfo * PDO * Collator * IntlDateFormatter * MessageFormatter * NumberFormatter * ResourceBundle * IntlRuleBasedBreakIterator

These classes would be corrected by detecting the invalid data in the constructor and throwing an exception at object construction time, rather giving an error when the created instance is used. * UConverter * SplFixedArray * ReflectionFunction * ReflectionParameter * ReflectionMethod * ReflectionProperty * ReflectionExtension * ReflectionZendExtension * Phar * PharData * PharFileInfo

The class PDORow will be changed to give an exception if an attempt is made to instantiate it from userland.

I hope to fix all of these in a branch before opening the vote, so that people can see the details of the changes.

At the very least all of the exception to be used for each class will be determined, and will be an appropriate exception i.e. not \Exception.

After the project is implemented, this section should contain

1. the version(s) it was merged to
2. a link to the git commit(s)
3. a link to the PHP manual entry for the feature