[5.4] Fluent partial resource routes

[jaripekkala]

Discussed in laravel/ideas#412

In 5.4 you can register routes more fluently e.g.

Route::name('users.index')->get('users', 'UserController@index');

It would be nice to get same behaviour for registering partial resource routes e.g.

Route::resource('users', 'UserController')->only('index', 'show');

Route::resource('comments', 'CommentController')->except('destroy');

Route::resource('posts', 'PostController')->only(
    'index', 'show', 'store', 'update', 'destroy'
);

[JosephSilber]

This relies heavily on GC.

Is it guaranteed to run before the actual route is matched? Or before the routes are cached?

[GrahamCampbell]

@JosephSilber is correct. This implement is not guaranteed to be correct.

[GrahamCampbell]

👎

[jaripekkala]

Implemented now in the other way proposed laravel/ideas#412 (comment)

Now only and except removes registered routes from the RouteCollection.

[taylorotwell]

If the PHP manual is correct, then yes the implementation is guranteed to be correct:

"The destructor method will be called as soon as there are no other references to a particular object"

[JosephSilber]

I doubt that is to be taken so literal.

There's no way for the runtime to constantly know that there are no longer any references to an object. The only way to know that is at the end of a GC cycle.

[GrahamCampbell]

I don't think the PHP manual is correct. It is up to the garbage collector to look for objects with a zero refcount.

[taylorotwell]

So you're going to submit a PR to the PHP documentation to fix it? It states it very specifically in several places :)

[taylorotwell]

And if this is the case, should we take ANYTHING in the PHP manual literally? If so, why?

[browner12]

I'll add my two cents from personal experience, I've had less than stellar results when dealing with destructors. It seemed to be very difficult to rely on when it was run.

[taylorotwell]

I would actually be very interested if someone can show a situation where it is not run as soon as there are no other references to the object like the documentation states.

[GrahamCampbell]

I would actually be very interested if someone can show a situation where it is not run as soon as there are no other references to the object like the documentation states.

By its very nature, it's non-deterministic, so, it's hard to provide an example and make any guarantees about when garbage collection will take place.

[GrahamCampbell]

That said, I was unable to get PHP to call these destructors in an "incorrect" order: https://3v4l.org/qUD51. Maybe there's an ordering by the garbage collector, just due to the implementation specifics?

[GrahamCampbell]

There are definitely user error cases I can think of though. Will write an example.

[GrahamCampbell]

Here we go: https://3v4l.org/ihpuf.

The application runs to completion, before the route gets registered.

[GrahamCampbell]

In terms of the PR, I actually already have a big problem with it, the register method is not supposed to return anything, but now it does...

(and if the programmer actually leaves a reference to the returned value somewhere, they are screwed)

[GrahamCampbell]

For that reason, I think this should be a 5.5 change, because it changes the contract.

[GrahamCampbell]

An implementation to avoid using destructors, would be for the registra to keep a queue of pending routes, and call some resolve method on them all, before we try to match anything.

[jaripekkala]

Thank you @GrahamCampbell for your time!

I implemented this once again based on your suggestion.

[taylorotwell]

This really was cleaner using destructors tbh, even though some are scared they aren't called when the PHP docs say they are called. Graham's code example is not really reflective of an actual Laravel routes file.

[GrahamCampbell]

Graham's code example is not really reflective of an actual Laravel routes file.

There's no reason to assume that people do routing in that way. I don't always.

[GrahamCampbell]

This solution is robust. And I don't think it's any worse than the destructor way tbh.

[GrahamCampbell]

this PR should target 5.5, because this is a breaking change to the interface, requiring implementations to return this object

[jaripekkala]

A newbie question: does this require any actions by me? My first PR 🥇

[taylorotwell]

Is this not changing the entire order of when your routes are registered though? Now the resource routes aren't actually registered until a later time and that makes a difference when registering things.

[taylorotwell]

Also I still haven't seen an actual case that won't work with the destructor approach. 😬

[GrahamCampbell]

Also I still haven't seen an actual case that won't work with the destructor approach.

Yeh, u have. That example where I showed the variable living beyond the scope we wanted it to. I mean, I could write a laravel app that demonstrates it, but I can't really be arsed.

The point is, that if a route remains referenced, after routing happens, then it won't have ever been registered. It's likely people would do this by mistake, but it's very hard for them to debug, if they don't understand the internals.

[JosephSilber]

The point here is more than that. Even if you don't have a reference to it anymore, there's no guarantee when GC will run. PHP's docs be damned.

[GrahamCampbell]

there's no guarantee when GC will run.

While that is true, I haven't been to actually get PHP to not garbage collect in time in my testing to try to break it.

[JosephSilber]

@GrahamCampbell that's not enough to actually build a feature on top of.

[GrahamCampbell]

that's not enough to actually build a feature on top of.

I agree, that's why I asked for this to be implemented without using that, and it now has been.

[jaripekkala]

The resources are currently registered only once so calling only or except methods after registration has no effect.

Should they? See 1048ce7

Pending resources could just be marked as registered and exception thrown if tried to limit/exclude methods afterwards.

Registering resources in the RouteCollection each time needed is a little overkill imo. There might e.g. be several route('users.index') calls in the templates.

[taylorotwell]

If you want to re-submit this using the original, simple approach I suggested go for it. I can't really subscribe to the view that the PHP documentation is not to be trusted because then we can't program anything in this language at all to be honest. I will function as if the documentation is true. If it is not, it should be fixed.

[GrahamCampbell]

For what it's worth 👎 On the original approach. This new approach is much better IMO. (ultimately, do what taylor says tho, he da boss)