This PR was merged into the master branch.
Commits
-------
51223c0 added upgrade instructions
50e6259 adjusted tests
98f3ca8 [Routing] removed tree structure from RouteCollection
Discussion
----------
[Routing] removed tree structure from RouteCollection
BC break: yes (see below)
Deprecations: RouteCollection::getParent(); RouteCollection::getRoot()
tests pass: yes
The reason for this is so quite simple. The RouteCollection has been designed as a tree structure, but it cannot at all be used as one. There is no getter for a sub-collection at all. So you cannot access a sub-collection after you added it to the tree with `addCollection(new RouteCollection())`. In contrast to the form component, e.g. `$form->get('child')->get('grandchild')`.
So you can see the RouteCollection cannot be used as a tree and it should not, as the same can be achieved with a flat array!
Using a flat array removes all the need for recursive traversal and makes the code much faster, much lighter, less memory (big problem in CMS with many routes) and less error-prone.
BC break: there is only a BC break if somebody used the PHP API for defining RouteCollection and also added a Route to a collection after it has been added to another collection.
So
```
$rootCollection = new RouteCollection();
$subCollection = new RouteCollection();
$rootCollection->addCollection($subCollection);
$subCollection->add('foo', new Route('/foo'));
```
must be updated to the following (otherwise the 'foo' Route is not imported to the rootCollection)
```
$rootCollection = new RouteCollection();
$subCollection = new RouteCollection();
$subCollection->add('foo', new Route('/foo'));
$rootCollection->addCollection($subCollection);
```
Also one must call addCollection from the bottom to the top. So the correct sequence is the following (and not the reverse)
```
$childCollection->->addCollection($grandchildCollection);
$rootCollection->addCollection($childCollection);
```
Remeber, this is only needed when using PHP for defining routes and calling methods in a special order. There is no change required when using XML or YAML for definitions. Also, I'm pretty sure that neither the CMF, nor Drupal routing, nor Silex is relying on the tree stuff. So they should also still work.
cc @fabpot @crell @dbu
One more thing: RouteCollection wasn't an appropriate name for a tree anyway as a collection of routes (that it now is) is definitely not a tree.
Yet another point: The XML declaration of routes uses the `<import>` element, which is excatly what the new implementation of addCollection without the need of a tree does. So this is now also more analogous.
---------------------------------------------------------------------------
by Koc at 2012-11-26T17:34:15Z
What benefit of this?
---------------------------------------------------------------------------
by Tobion at 2012-11-26T17:56:53Z
@Koc Why did you not simply wait for the description? ^^
---------------------------------------------------------------------------
by dbu at 2012-11-26T18:33:09Z
i love PR that remove more code than they add whithout removing functionality.
---------------------------------------------------------------------------
by Crell at 2012-11-26T18:49:52Z
There's an issue somewhere in Drupal where we're trying to use addCollection() as a shorthand for iterating over one collection and calling add() on the other for each item. We can't do that, however, because the subcollections are not flattened properly when reading back and our current dumper can't cope with that. So this change would not harm Drupal at all, and would mean I don't have fix a bug in our dumper. :-) I cannot speak for any other projects, of course.
---------------------------------------------------------------------------
by Tobion at 2012-11-27T19:06:34Z
Ok, this is ready.
we don't need the logic to merge numeric keys, as we don't have them. I could also improve the genrated code by PhpMatcherDumper a little by saving a function call.
Only + and % are now encoded in generated routes, since they are the only characters that, if not encoded, could cause problems/conflicts when decoded. Namely + turns into a space, and % followed by numbers could do funky things.
The matcher decodes everything which works since nothing will have %NN without being escaped, and + are escaped as well.
* fixes a problem with security (/foo/bar and /foo///bar are not the same URL as far as security is concerned)
* this can still be done in your web server configuration or by adding a core.request listener
