Search the Community
Showing results for tags 'mediator'.
Hello fellow wirecionados, looking at the roadmap to 3.0 always gives me the good feeling of being on the sunny, futureproof side when using Processwire. Everything we already have together with the features planned is all I ever needed from a CMS. At least this is what I thought until that next idea came to my mind. Actually, there's one thing about Processwire that I have been struggling with from the start and until now: the hook system. I know, the hook system is deeply enrooted in Processwire (as can be seen from the fact that it's implemented in the Wire class that virtually every other class extends) and therefore is one of the key aspects for its success in the first place. It allows for the relatively loose coupling of components, with a lot of the core functionalities being implemented as modules. This makes for the great extensibility that Processwire is famous for. There aren't many parts of Processwire that can't be tweaked and modified to anyones needs by developing a module that alters some core functionality. I'm really glad about that and probably wouldn't have started to use Processwire in the first place, if the hook system didn't exist. That all being said, I also see some more or less problematic downside to the current implementation. It's sometimes pretty hard to find out, where functions are actually implemented. How long did it take you to find out that the actual logic for Page::render() is implenented in PageRender::renderPage()? While this kind of flexibility is just great, the implementation is not very verbose. This makes things harder than they have to be – especially for beginners. Just to make the argument above worse, there are several different ways to define a hook: addHook(), addHookAfter() & addHookBefore(), and all can be called either with a Class name or on an object. This makes it even harder to find modules, that hook into a particular function. I know, we have Soma's fabulous Captain Hook, and I use it all the time, but it still feels more like a workaround than like a solution. Speaking of Captain Hook, there's still a „Crocodile Hooked“ missing, a central repository that keeps track of all methods that are hooked to another method, including (private) site modules – and if possibly available at run time. Hooks for classes are saved in a static variable of Wire, but local hooks can only be fetched from that particular instance of an object. The convention with the three underscores in front of a function name makes functions harder to find. It is a nightmare to debug. If you ever tried to follow a request through the system with Xdebug, maybe involving some hooks placed by modules, you probably know what I mean. You have to go through Wire::runHooks() and __call() again and again to get where you want. It also prevents some of the convenient features of your favourite IDE™ (e.g. code completion, parameter hints when using a function, go to the definition of a function by clicking on it, etc.) from working in some cases. While one could definitely argue, that this is the fault of the IDE used, it's still slowing down work and not likely to be changed anywhere soon. Please excuse me if this was too ranty and even more correct me if anything I mentioned is an error on my side. I have to emphasize that none of these downsides prevents anyone from getting something done, it just makes it a tad less obvious and comfortable. Thinking through the things that I don't like about the current implementation of the hooks system, it came to my mind, that what we currently have here is kind of a non-verbose version of a „traditional“ event system and maybe what's missing could be more verbosity, so maybe such an event system (that implements the mediator pattern) could be of help? What I have in my mind here is the Symfony EventDispatcher or the Drupal8 EventDispatcher component, which is similiar to the CakePHP Events System and the handling of Events in Laravel. Let's look at how a solution to the problems mentioned might work by way of example using the Symfony EventDispatcher. In this particular example, the difference would be that Page::render() would actually be implemented in the Page class but just dispatches an Event.render event. Functionally no real difference, but much more verbose and therefore it is much more clear, that you will find the implementation anywhere a listener for Event.render is added. A new implementation could use just one method to add listeners (e.g. EventDispatcher::addListener()). Execution before or after an event will have to be handled by multiple events where this makes sense. Because the event dispatcher is the application-wide sole instance where event listeners are stored, it's easy to get all registered event listeners at runtime (using EventDispatcher::getListeners()). No more underscore's, just calls to dispatch events and calls to listeners. Debugging get's as smooth as it get's with loose component coupling. No more running through __call() all the time, always be able to check the defined event listeners. Having separated before and after events prevents you from some running through runHooks(). The added verbosity also helps to grasp the existing architecture quickly. Most of the IDE comfort functionality that doesn't work now should work with the Event System in place, because functions are always really where they are expected and with the exact same name. As an added benefit this system is very well known to a lot of developers already. But there are also some downsides for sure: Added verbosity is, well, verbose. More keys will have to be hit to get something done. Method and property would have to be implemented differently. This can be done by sending events on missing methods/properties, but it certainly looses some of the appeal it has now. The key argument would for sure be, that it breaks compatibility with a lot of core and user contributed modules. This is definitely true, so there shouldn't be an easy decision on this one. I thought I'll use the time before 3.0 to bring this one up, because as I understand it, the next major will break some stuff either way. Also I think that it should be possible to do the switch by in the beginning only deprecating the old hook system while keeping it functional, by translating calls to the old functions to adding and calling event listeners. This will have a performance impact, but would make a gentle transition possible. Now I'm really interested on your opinions on this topic. Do you already like the current implementation and wouldn't want to loose it? Do you have other problems with hooks? What kind of solution could you imagine? Is such a deep change realistic (and worth it) at all? Thanks a lot for your feedback Marc