Re: static type-references

Dear Rasmus, On Thu, Mar 14, 2013 at 5:26 PM, Rasmus Schultz <rasmus@mindplay.dk> wrote: > Hey List, > > What do you think about adding a typeof() operator to PHP? > > It could go something like this: > > class User > { > public $name; > } > > /** > * @var ReflectionClass $user_type > * @var ReflectionProperty $user_name_property > */ > > $user_type = typeof(User); > > $user_name_property = typeof(User::$name); > > Mind you, this is an operator and not a function - User and User::$name are > interpreted as symbols. > > This would allow for static analysis of class and property-references. > > Why? Because you can't use static analysis on code like this: > > $user_type = new ReflectionClass('User'); > > $user_name_property = new ReflectionProperty('User', 'name'); > > 'User' and 'name' are just strings, and strings are dead matter when it > comes to static analysis. > It is certainly possible for static analyses to reason about literal strings. Handling reflective code is not that hard when its arguments are literals. > > Most form builders and ORMs (and other reflective meta-programming > components) require references to properties or classes, e.g.: > > $form->textInputFor($user, 'name'); > > There is no knowledge that 'name' actually refers to the symbol User::$name > and hence to way to perform static analysis or use refactoring tools on > this code. > > Compare to: > > $form->textInputFor($user, typeof(User::$name)); > To me this code makes no sense. I expect typeof to return a type, since we have no first-class types in PHP, I expect it to return a string representing the type. If I understand your example, here it returns a Reflection* ? What about typeof($a->b) or even typeof($a->$foo) ? I assume both of them would be forbidden ? > This code can be analyzed and refactored, because the reference to the > User::$name property is now literal. > > Difficulty to statically analyze code comes from its dynamic nature, to me typeOf(User::$name) is no less dynamic than new ReflectionProperty("User", "name"). > Actually having to type out typeof() may not be the most elegant approach, > and using this syntax it does resemble a function - having a more distinct > syntax might be better. > > But those things aside, what do you think about having a way to statically > reference types and members? > Thanks, > Rasmus Schultz > -- Etienne Kneuss