[RFC][DISCUSSION] Object type hint

On Sun, Oct 23, 2016 at 1:39 AM, Michał Brzuchalski <michal@brzuchalski.com> wrote:

I generally don't mind keywords/new classes being added to minor versions. However I am nervous about this one as I am sure there are a lot of `class Object` floating around. Maybe this would be best suited for 8.0 with E_DEPRECATED being emitted to warn people of the up-coming change? Also, can the RFC be edited to give an example of how people will be affected? Other than this issue I support this RFC.

Hi!

I don't see why it would be important that the variable is an object but not important which kind of an object. There's very small number of things you can do with an object without knowing the exact type. For generic functions, like serialization, there's no reason why they shouldn't be able to serialize integers or nulls or booleans. On the contrary, many objects would not be serializable. For less generic functions, like DI containers, object type is next to useless since you can't just pass any object to implement a service - it should actually implement the required service. So adding "object" does not add real specificity to it.

They probably think more "strictness" and more checks means better code. It is usually not true per se. E.g. in the RFC examples the type does not seem to be serving a good purpose - it's either wrong (like in the case of json_decode - it won't always return an object), or nonspecific - like in factory, very rarely one has a factory returning just arbitrary objects, no matter what they are.

On Sun, Oct 23, 2016 at 3:39 AM, Michał Brzuchalski <michal@brzuchalski.com> wrote:

Type hinting, along with the assert statement, provide a means to establish expectations for a methods. It's part of a coding paradigm known as 'design by contract' and it's critical for large teams and projects that are used among multiple teams. Checking to see if a variable is an object in no way provides a real test of what can be done with it. Can it be iterated on? Probably, but not always. Does a method exist on the object? No way to know without further testing. Defining what an object can do is the reason interfaces exist. This is also why objects are allowed to implement multiple interfaces, so that they can be labeled according to all the roles they can fulfill. My principle worry with this RFC is it encourages an anti-pattern - not using interfaces and thereby not clearly labeling what objects can and cannot do. The one situation this might be useful requires additional testing anyway - that would be a method that will receive objects of one or more classes that don't have a common interface and rewriting the class to add the desired interface isn't viable (for example, the objects come from different packages). In that event though type hinting for an object is only going to be the first step in determining if the code state is sane, you'll also need to check if the object belongs to one of the several classes the code can work with. This gets even hairier when dealing with a collection of objects. I wrote this generic assertion for Drupal 8. https://api.drupal.org/api/drupal/core%21lib%21Drupal%21Component%21Assertion%21Inspector.php/function/Inspector%3A%3AassertAllObjects/8.2.x In your own code you'll likely need to do something similar with your argument. Incidentally, these sorts of tests, which can only fail if the code is wrong, are what assert statements are to be used for, not exceptions. This way the check for correct objects can be turned off in production to avoid the cpu overhead of the test (assuming your function doesn't need to switch to different branches depending on the exact object used). In summary, -1 to this as it encourages bad code writing.

Hi, On 23.10.16 09:39, Michał Brzuchalski wrote:

I just wished a few weeks ago I was able to typehint that. Actually I even tried just to lern it doesn't work. It's a unit test helper method to make a method on an object public, signature: protected function makeMethodPublic(object $obj, string $methodName): ReflectionMethod So, I can come up with exactly one (to me) useful case, but it's arguable not "that important" and I certainly wouldn't and didn't add an is_object check for this, so there goes my priority for it. thanks, - Marus

Hi Michal! First of all, many thanks to you and Dan for pushing this forward. It is indeed a very needed and long due addition to the engine. I just wanted to clear up the use-case scenarios for everyone in here that keeps stating that this will lead to bad code. It is very unlikely to type-hint "object" inside your business domain. Where this functionality is useful is any of the following (with examples): - generic hydrators/property access (these exist, and they are widely used inside form components) class SillyLibraryHidrator { public function hydrate(object $object, array $someData) { // pseudo: foreach ($objectReflectionProperties as $p) { $p->setValue($object, $someData[$p->getName()]); } } } - serializers - also widely used, can generally tell them to produce an object of a certain type. Consumers would then wrap it, and use a more specific type-hint interface JsonSerializer { public function serialize(object $object) : string; public function deSerialize(string $data, string $className) : object; } - identity maps - typically part of ORMs interface IdentityMap { public function add(string $id, object $object) : void; public function getId(object $object) : ?string; public function getObject(string $id) : ?object; } - code generators interface WrapThingIntoAProxy { public function wrap(object $object) : object; } - instantiation logic for generic factories interface AGenericFactory { public function create(array $data, string $className) : object; } - table data gateways/repositories interface TableDataGateway { public function fetchById(array $identifier) : ?object; } These are just some of the examples that get direct benefit from this sort of type-hint, or at least the scenarios that I work with on a daily basis. Please don't tell me that I have to get back to `is_object()`, because: * it is implicitly part of the implementation, and not part of the reflectable signature *AS IT SHOULD BE* * it is more code for me to test * I cannot throw a TypeError myself, if a non-object is given (TypeError can't be instantiated in userland, AFAIK), and every time I have to design custom exceptions (also to be tested/documented) * this way is actually faster than a manual `is_object()` check * intent is conveyed to the consumer directly, and not via documentation docblock * allows for explicit nullability, yet splitting object/non-object APIs * many library-level codebases rely on generic code that doesn't know about one specific type, but just about "objects" The alternative is going with full generics, but that's not viable for now. This additional type hint is pretty much covering the scenarios above in a very approachable way. Greets, Marco Pivetta http://twitter.com/Ocramius http://ocramius.github.com/ On Sun, Oct 23, 2016 at 12:39 AM, Michał Brzuchalski <michal@brzuchalski.com

Hi, Michał Brzuchalski wrote:

I like this proposal. We already have an object type internally for the same reasons (functions acting on a generic object), so it makes sense to extend this to userland which may want to do the same kinds of things. One potential benefit is that functions can be more generic. I imagine there's some functions out there typed against \stdClass which could really accept any object if given the option. Best of luck.