Commits
-------
1474aa5 [Form] Fixed consideration of Twig's template inheritance and added another performance-improving check
b4ec7f5 Fixed my rubbish English
d11f8b5 [Form] Fixed passing of variables in the FormRenderer
629093e [Form] Extracted common parts of FormHelper and FormExtension into separate classes
216c539 [Form] Implemented a more intelligent caching strategy in FormHelper (PHP +100ms, Twig +100ms)
Discussion
----------
[Form] Merged FormHelper and FormExtension and implemented a better caching strategy
Bug fix: no
Feature addition: no
Backwards compatibility break: no
Symfony2 tests pass: yes
Fixes the following tickets: -
Todo: -
This PR extracts common parts of `FormHelper` and `FormExtension` into implementations of the new interfaces `FormRendererInterface` and `FormRendererEngineInterface`. The implemented `AbstractRendererEngine` features a more intelligent caching strategy than the one used before. When this strategy was implemented directly in `FormHelper`, the performance of [this specific, heavy form](http://advancedform.gpserver.dk/app_dev.php/taxclasses/1) could be improved from **2.5** to **2.25 seconds** on my machine for PHP templates.
Due to the abstraction and delegation, the performance gain is not that big anymore, but we still have a performance gain of about **0.1 seconds** for both PHP and Twig in the above example. The second, big improvement of this PR is maintainability - the differences between PHP and Twig templates are now contained in relatively small classes - and extendability (it is very easy now to support different template engines).
---------------------------------------------------------------------------
by stof at 2012-07-14T13:47:19Z
should a similar refactoring be done for the [Twig rendering](https://github.com/symfony/symfony/blob/master/src/Symfony/Bridge/Twig/Extension/FormExtension.php) ?
---------------------------------------------------------------------------
by bschussek at 2012-07-14T13:49:25Z
Yes. I would like to merge the common parts of Twig's FormExtension and PHP's FormHelper into an abstract class. Before that I need to have a [working, heavy Twig Form](https://twitter.com/webmozart/status/224135287377371138) in order to measure whether I don't actually decrease the performance with Twig. Can you help me there?
---------------------------------------------------------------------------
by vicb at 2012-07-16T21:48:24Z
Would it make sense to create a 'renderer' folder in the form component and move related classes there ?
---------------------------------------------------------------------------
by stof at 2012-07-16T22:06:58Z
@vicb It makes sense to keep the Twig renderer in the brisge. This is what the bridge is about. Moving the Twig class to the component would not be consistent. And the PHP renderer is already in the component (but it could make sense to move the helper from FrameworkBundle to the TemplatingExtension of the Form component though)
---------------------------------------------------------------------------
by vicb at 2012-07-16T22:16:50Z
@stof I was only referring to the classes located in the Component/Form folder.
---------------------------------------------------------------------------
by vicb at 2012-07-16T22:27:27Z
Overall I don't really know what to think of this PR. PHP and Twig use a different way to support blocks:
- PHP has one block per file,
- Twig could have many blocks per templates.
I am not sure if this PR is optimal for Twig and improves maintainability ?
---------------------------------------------------------------------------
by stof at 2012-07-16T22:46:11Z
@vicb it avoids duplicating the whole rendering logic for each engine (there is at least a third one in [SmartyBundle](https://github.com/noiselabs/SmartyBundle/blob/master/Extension/FormExtension.php) btw)
---------------------------------------------------------------------------
by bschussek at 2012-07-17T07:16:42Z
@vicb I don't think a renderer subfolder makes sense. The interfaces belong to the main namespace, and then the subfolder would only contain two classes.
Considering maintainability for Twig, I think that this PR in fact increases it. TwigExtension before always had to check the whole type hierarchy, while now the code in AbstractRendererEngine makes sure that this process is speeded up.
Before:
```
load _some_entity_field_label:
- check _some_entity_field_label
- check entity_label
- check choice_label
- check form_label
load _some_other_entity_field_label
- check _some_other_entity_field_label
- check entity_label
- check choice_label
- check form_label
a.s.o.
```
After:
```
load _some_entity_field_label:
- check _some_entity_field_label
- check entity_label (hits the cache if entity_label was checked before)
- check choice_label (hits the cache if choice_label was checked before)
- check form_label
load _some_other_entity_field_label
- check _some_other_entity_field_label
- check entity_label (now definitely hits the cache)
a.s.o.
```
Since many fields share the same ancestors in the inheritance tree, this definitely improves performance.
As can also be deducted here, custom block names such as `_some_entity_field_label` are now a major drawback. There is nothing we can cache for them, so they need to be checked for every individual block that we load. Removing this feature surprisingly gains no performance for Twig (I need to investigate why at some point), but it speeds up rendering for **250ms** using the PHP engine on [this example form](advancedform.gpserver.dk/app_dev.php/taxclasses/1), dropping the rendering time from 1.25 to 1 sec on my local machine. I'm not sure what we should do here.
---------------------------------------------------------------------------
by stof at 2012-07-17T07:21:31Z
@bschussek could it be possible to have an implementation checking the custom block and another one skipping it ? This way, the user could disable this feature when he does not need it.
---------------------------------------------------------------------------
by bschussek at 2012-07-17T07:38:34Z
@stof It would be possible to add a switch to `FormRenderer` that controls whether custom blocks are checked or not.
If this switch is disabled by default, we break BC. If this switch is enabled by default, it will be pretty useless. People will start designing away for custom blocks, and once they want to improve performance, they can't turn off the switch anymore because it would require too many changes.
---------------------------------------------------------------------------
by stof at 2012-07-17T08:08:38Z
@fabpot what do you think about it ?
---------------------------------------------------------------------------
by bschussek at 2012-07-17T08:41:43Z
Another option that just came to mind is to remove inheritance checks for anything but _widget and _row. I.e., if we render `entity_widget`, check
```
_id_widget
entity_widget
choice_widget
form_widget
```
But if we render `entity_label`, only check
```
_id_label
form_label
```
This improves PHP Templating for **170ms** and Twig for **20ms**. We gain another **150ms** for PHP Templating and **~15ms** for Twig if we also restrict custom fields (_id_widget) to the _widget and _row suffixes (it's really hard to tweak the renderer for Twig.. I think a lot of its performance bottlenecks lie in Twig itself).
Do you have any data on how often blocks other than _widget and _row are customized for specific types/IDs?
---------------------------------------------------------------------------
by stof at 2012-07-17T09:47:38Z
Well, I think most of the time other blocks are not even customized based on the type :)
---------------------------------------------------------------------------
by Tobion at 2012-07-17T14:32:39Z
From my experience rendering the form components individually is easier and more flexible than customizing by ID or type.
But there are still use cases for customizing like library-like bundles (e.g. Bootstrap).
Commits
-------
23d8735 Added NativeFileSessionHandler to classes to compile .
12d6ae7 Removed FileSessionHandler from FrameworkExtension to stop compiling a file that does not exist.
Discussion
----------
[FrameworkBundle] Removed FileSessionHandler from FrameworkExtension to stop compiling a file that does not exist.
PR #4899 removed FileSessionHandler which caused a class not found error from FrameworkBundle after the cache was created. This PR will fix it.
Bug fix: yes
Feature addition: no
Backwards compatibility break: no
Symfony2 tests pass: yes
Todo: -
License of the code: MIT
---------------------------------------------------------------------------
by stof at 2012-07-13T19:12:51Z
you should add the NativeSessionHandler class in the list instead as it replaces it
---------------------------------------------------------------------------
by tystr at 2012-07-13T19:14:29Z
+1
---------------------------------------------------------------------------
by zachbadgett at 2012-07-13T19:15:55Z
Done
Commits
-------
cd7835d [Form] Cached the form type hierarchy in order to improve performance
2ca753b [Form] Fixed choice list hashing in DoctrineType
2bf4d6c [Form] Fixed FormFactory not to set "data" option if not explicitely given
7149d26 [Form] Removed invalid PHPDoc text
Discussion
----------
[Form] WIP Improved performance of form building
Bug fix: no
Feature addition: no
Backwards compatibility break: no
Symfony2 tests pass: yes
Fixes the following tickets: -
Todo: **Update the Silex extension**
This PR is work in progress and up for discussion. It increases the performance of FormFactory::createForm() on a specific, heavy-weight form from **0.848** to **0.580** seconds.
Before, the FormFactory had to traverse the hierarchy and calculate the default options of each FormType everytime a form was created of that type.
Now, FormTypes are wrapped within instances of a new class `ResolvedFormType`, which caches the parent type, the type's extensions and its default options.
The updated responsibilities: `FormFactory` is a registry and proxy for `ResolvedFormType` objects, `FormType` specifies how a form can be built on a specific layer of the type hierarchy (e.g. "form", or "date", etc.) and `ResolvedFormType` *does the actual building* across all layers of the hierarchy (by delegating to the parent type, which delegates to its parent type etc.).
---------------------------------------------------------------------------
by schmittjoh at 2012-07-12T18:25:40Z
Maybe ResolvedFormType
---------------------------------------------------------------------------
by jmather at 2012-07-13T02:56:38Z
I really like ResolvedFormType. That's the naming method I took for my tag parser that handes the same conceptual issue.
---------------------------------------------------------------------------
by axelarge at 2012-07-13T05:25:00Z
ResolvedFormType sounds very clear.
This change is great and I desperately hope to see more of this kind
---------------------------------------------------------------------------
by Baachi at 2012-07-13T06:41:26Z
Yes `ResolvedFormType` sounds good :) 👍
---------------------------------------------------------------------------
by fabpot at 2012-07-13T07:11:33Z
I like `ResolvedFormType` as well.
---------------------------------------------------------------------------
by henrikbjorn at 2012-07-13T07:46:48Z
👍 `ResolvedFormType` :shipit:
---------------------------------------------------------------------------
by stof at 2012-07-13T18:01:51Z
This looks good to me
Commits
-------
c061c30 Router#resolveString should return null instead of empty string when $value is null
a1d1a02 Null default value route regression
Discussion
----------
[Router] Null default value route regression
Bug fix: yes
Feature addition: no
Backwards compatibility break: no
Symfony2 tests pass: yes
Fixes the following tickets: #4823
Todo:
License of the code: MIT
Documentation PR:
---
The commit by @vicb 0555913fbb introduces a regression in the handling of default values that are null, while there are requirements still set for this value.
This is a failing test case and fix for the issue.
---------------------------------------------------------------------------
by merk at 2012-07-10T04:24:40Z
Now contains a fix, tests now pass.
Commits
-------
0555913 [FrameworkBundle] Allow using kernel parameters in routes
Discussion
----------
[FrameworkBundle] Allow using kernel parameters in routes
Kernel parameters can now be used at any position in patterns, defaults and requirements.
Relates to: #3316, #3276
**Differences from 3316:**
- The substitution is now done in the `Router`,
- 3316 uses `$container->getParameterBag()` which is not part of the `ContainerInterface`. The way it been solved in this PR is that some code have been duplicated inside the `Router`, see `resolveString()`.
**BC break:**
Before this PR, nonexistent parameters would have be silently ignored (ie `%idontexist%` would not have been replaced). After this PR, they will throw an exception. However you can escape the value (ie `%%idontexist%%` will be accepted and replaced by `%idontexist%`).
_This behavior is not mandatory and can be reverted if needed. However this keeps the router more consistent with the DI_.
Any feedback ? @helmer @Koc
---------------------------------------------------------------------------
by Seldaek at 2012-07-04T12:40:08Z
👍 for consistency.
---------------------------------------------------------------------------
by helmer at 2012-07-04T13:07:11Z
+1 a much better solution to the problem than mine, closing #3316
---------------------------------------------------------------------------
by Tobion at 2012-07-04T13:21:59Z
How about escaping kernel params with `\%idontexist%` as I suggested it for route params in 4563?
---------------------------------------------------------------------------
by stof at 2012-07-04T13:28:55Z
@Tobion this would not be consistent with the way DI parameters are escaped elsewhere
---------------------------------------------------------------------------
by Koc at 2012-07-04T14:24:25Z
Looks good for me, thanx.
