Routing: Legacy Router Rule for com_users

[Hackwar]

This implements the current routers behavior of com_users as a legacyrouter. This is meant for those that want to keep the old behavior at all costs. Later, parameters will be introduced to switch between the legacy behavior and new features.

This router is copied verbatim from the current router to make sure that the same behavior is kept. No changes should be done to it until Joomla 4.0, when this is supposed to be deleted. The component routers right now are so fragile, that a slow transition will be next to impossible. To prevent any breaks in backwards compatibility, this legacy router is introduced, so that people having problems with the new router can switch back.

How to test

1. Check your site before applying the patch and see the URLs of com_users.
2. Apply the patch.
3. See the URLs of com_users to stay the same.

Dependencies

This PR requires #5446 to be accepted by the project before it can be applied.

This was made possible through the generous donation of the people mentioned in the following link via an Indiegogo campaign: http://joomlager.de/crowdfunding/5-contributors

[johanjanssens]

@Hackwar By having this method your are coupling the categories implementation to the routing API. Do we need to have a this method here ? Wouldn't this be something either for a more specialised router that understands categories, or for a specific implementation in a component router ?

[Hackwar]

I'm indeed pondering if this method should be there or not. As I wrote in the description, I'm interested in having get methods to retrieve the necessary data and this would be one of those methods. Right now, most of our core frontend components need this method and a lot of third party extensions work with categories, too. So I could either have this here or implement another class between this one and the actual router to simply implement that class. Trying to strike a balance here between abstraction and over-engineering, I put it in this class. Also, I could quite frankly not come up with yet another class name for such an intermediate class. 😉

Now, I would leave it at that, you would move it into its own class, so I would call for a third party to cast a vote what to do. 😄

[johanjanssens]

Could these methods not be exposed by the rules JComponentRouterRulesCategory object. If they need to be accessible through the router you could use magic calls to get them making the rules object work like a mixin. That gives you flexibility without cluttering the initial router.

[Hackwar]

No, the get"ViewName" methods need to be accessible from inside other rules and thus have to be implemented in the router. If you, as a dev that needs to modify an existing routers, need other methods (for example when you add your own view to a component) you can create your own router which extends from the component router and set it with JRouterSite::setComponentRouter() as a replacement for the original one.

I thought about using magic calls, but I have the feeling that this complicates the implementation again and makes it less intuitive. Joomla devs are not used to that...

[johanjanssens]

@Hackwar Would consider calling this method registerView() instead of register to better declare it's purpose. All other view related methods in the proposed API also have 'View' in their name which makes it easier to see the correlation between the methods.

[Hackwar]

Will do.

[johanjanssens]

Great!

[johanjanssens]

@Hackwar see my other comment about the getCategory() method. If you don't add it you also don't need to add the getName() method. Additionally you could ask yourself if this method has use outside of the implementation as proposed in getCategory(). If not it might be better to remove this and implement the logic in getCategory().

[Hackwar]

This method has a use outside of JCategories. It is also used to find the right menu items for the lookup and for other filter tasks that we have. Yes, it is named like our magic getters, but consider it one of the reserved words that can not be used as a name for a view.

[johanjanssens]

In that case, maybe consider giving the method a protected scope and renaming it to getRouterName() to allow the router to have a getName() method if it needs to ? Otherwise this becomes a need to know that you cannot expose a 'name' property in your routes ?

[johanjanssens]

@Hackwar By making the property public you are allowing it to be changed from the outside. Is that intentional ? Couldn't this result in unwanted side effects ?

[Hackwar]

That is indeed unintentional. How could that slip through? Thanks, will change.

[johanjanssens]

Thanks!

[johanjanssens]

@Hackwar Bit of a code smell here. You should try to check against and interface and not an implementation. This gives developers more flexibility, Eg would be better if it would be

public function __construct(JComponentRouterAdvancedInterface $router)

[Hackwar]

I choose the concrete implementation, because the rules are pretty tightly coupled to this class. I also again tried to find a balance between abstraction and over-engineering. My heart is not attached to this, but I'd rather not add another folder to this one simply to stuff the interface in there for the autoloader...

As with the last cases, I'm happy to be overruled by similar opinions.

[johanjanssens]

Typo, should have been public function __construct(JComponentRouterInterface $router)Additionally would consider adding this to the JComponentRouterRulesInterface to enforce the constructor.

[Hackwar]

JComponentRouterInterface does not have the interface that is required by the rules. You definitely need an interface like JComponentRouterAdvanced. Remember that this advanced router is optional. If a component doesn't want it, we are not forcing it upon the devs. So JComponentRouterInterface does not have the features that are required for the advanced component router.

[johanjanssens]

@Hackwar I understand, it would still be good to have an interface for this so we don't need to work against an implementation. Can we move the API as defined in the advanced router into an interface ?

[johanjanssens]

@Hackwar Do you want to allow this ? By letting any Itemid being passed in a developer can hijack a route. Connecting a component to any random Itemid. Shouldn't that be prevented ?

[Hackwar]

This is necessary for the menu system to work. The menu module hands in "index.php?Itemid=XX" and this is sent through preprocess. If we don't allow Itemids to be set from the outside, we will get an issue here. So I've allowed that.

[johanjanssens]

Oki.

[johanjanssens]

@Hackwar As commented before. If the you need dummy methods here you can ask yourself if you really need those methods in the interface at all ?

[Hackwar]

My first implementation did not have an interface but instead you could simply hand in callbacks for rules. You also had to manually assign them to parse and build rules. That was before I introduced preprocess and before I expected the rule to be an instantiated object with the router handed in through the constructor. That said, I decided against the ultra-flexible solution, since I think it is more error-prone for third party developers and choose this interface. It requires less methods to register rules and less code to store and execute them and at the same time gives devs a construct to work from.

[johanjanssens]

Not following you here. Does that mean you agree in getting it out of the interface or not ?

[Hackwar]

This means that I think the interface is good the way it is.

[johanjanssens]

@Hackwar As commented before. If the you need dummy methods here you can ask yourself if you really need those methods in the interface at all ?

[Hackwar]

See my previous comment. 😄

[johanjanssens]

@Hackwar Since you are using the result of buildLookup() in this code block I would make buildLookup() return the lookup array. It can still cache it but it makes sense it returns it for use in the code block here.

[Hackwar]

This might sound stupid, but I wanted to optimize away a function-call here. I also choose the direct access of the lookup property in the code further down in the hopes that it saves a bit of memory and time instead of copying a pointer each time into a new variable in the methods scope etc. Yes, it might be a bit over-optimized, but I would stick to it.

However I see a bug in there, that the $language is not correctly set if the lookup is already created and the query language parameter is set. Will have to split that into 2 ifs.

[johanjanssens]

I would choose code readability over the few microseconds you might be saving here. Especially since you are already caching.

[johanjanssens]

@Hackwar Not sure about the proposed function name here. getMenuItemMap() might be a better name to reflect what this method is doing. I would also not make it return void but return the result instead. The result could be filtered by the language passed in through the method parameter.

[Hackwar]

Its the name that we used so far in the *HelperRoute classes for this. I simply choose it for the sake of familiarity with those methods.

[johanjanssens]

Since you are adding it new here and we are refactoring and removing the HelperRoute classes anyway, I would consider renaming it to benefit a clearer API.

[johanjanssens]

@Hackwar Getting the active menu item is a common operation in a router. Could consider adding a getActiveMenuItem() method to return this.

[Hackwar]

Sorry, but I would call that over-engineered code. The method would simply contain this one line and that is it and I don't see any benefit in that.

[johanjanssens]

I don't see why that would be over engineering. This is the L in SOLID, or the Liskov substitution principle and Design by Contract Those are well established programming principles.

By not exposing the getter you are relying on implementation instead of on an interface. Having a getActiveMenuItem() or getActiveMenu() as a getter creates a declarative API and make the code more easily extendable. It also better communicates the intention of the API by having the method as part of it.

I would still reconsider this and try to stick to SOLID as much as possible when creating new API's.

[johanjanssens]

@Hackwar Getting the default menu item is a common operation in a router. Could consider adding a getDefaultMenuItem() method to return this.

[Hackwar]

See my above note.

[johanjanssens]

See my note. Please consider it.

[johanjanssens]

@Hackwar Does this need to be accessible from object scope ? If not might consider making it private to prevent developers for using it directly ?

[Hackwar]

In general, rules should not be extended upon. They are self-contained pieces of code. But if someone really wants to, he/she should be able to mess around here. But since its not accessible from the component router, I would say that it is okay the way it is.

[johanjanssens]

Oki, you could still consider making it private for now. If someone needs to access this is could be made protected easily. Having it private gives you less to worry about in case the code needs to be refactored later.

[wilsonge]

Probably need to change the text there then :)

[johanjanssens]

@Hackwar If the method to add a rule if called attachRule(), if would make sense that the method to remove it is called detachRule(), detach being the opposite of attach. Alternatively you could consider renaming attachRule to addRule().

[Hackwar]

What would you prefer? I can do either.

[johanjanssens]

Attach/Detach is mostly used in the context of an object. Since the rule is an object, attach/detach make most sense here. Does that help ?

[johanjanssens]

@hannes Good work. Feedback as requested :

• If you have special setters it's best to have getters too and make the class variables protected or even private. Your implementation now allows for the class variables to be set directly which can have unwanted side-effects.
• I would consider adding an isNestable() method to be able to check if the views is nestable or not.
• What is the difference between setName() and setViewName() ? These methods are a bit confusing to me, is there a better way we can name them ?

[johanjanssens]

@Hackwar I would consider changing this to getCategoryIdyou are getting the category identifiers, not it's key. They key would be the name of the identifier, aka 'id'.

[johanjanssens]

@Hackwar Oki, fair comments. Additional feedback :

• I still don't feel category handling should be in the this router. It would make more sense to me that this router is part of the categories component. It could be an abstract router one can extend from. By the chosen implementation you are tightly coupling categorie handling with routing.
• I would consider renaming JComponentRouterAdvanced to JComponentRouterView. Advanced isn't a very concrete name. This router deals specifically with component routing based on registering views. Would make sense the name reflect this, this also matches with the JComponentRouterViewconfigurationObject.

[Hackwar]

I've combined the changes from this and all other routing related PRs into a new PR: #7615 Please review and comment in the new PR. I'm closing this one, so that we can focuse on the new PR.