Commits
-------
887c0e9 moved EngineInterface::stream() to a new StreamingEngineInterface to keep BC with 2.0
473741b added the possibility to change a StreamedResponse callback after its creation
8717d44 moved a test in the constructor
e44b8ba made some cosmetic changes
0038d1b [HttpFoundation] added support for streamed responses
Discussion
----------
[HttpFoundation] added support for streamed responses
To stream a Response, use the StreamedResponse class instead of the
standard Response class:
$response = new StreamedResponse(function () {
echo 'FOO';
});
$response = new StreamedResponse(function () {
echo 'FOO';
}, 200, array('Content-Type' => 'text/plain'));
As you can see, a StreamedResponse instance takes a PHP callback instead of
a string for the Response content. It's up to the developer to stream the
response content from the callback with standard PHP functions like echo.
You can also use flush() if needed.
From a controller, do something like this:
$twig = $this->get('templating');
return new StreamedResponse(function () use ($templating) {
$templating->stream('BlogBundle:Annot:streamed.html.twig');
}, 200, array('Content-Type' => 'text/html'));
If you are using the base controller, you can use the stream() method instead:
return $this->stream('BlogBundle:Annot:streamed.html.twig');
You can stream an existing file by using the PHP built-in readfile() function:
new StreamedResponse(function () use ($file) {
readfile($file);
}, 200, array('Content-Type' => 'image/png');
Read http://php.net/flush for more information about output buffering in PHP.
Note that you should do your best to move all expensive operations to
be "activated/evaluated/called" during template evaluation.
Templates
---------
If you are using Twig as a template engine, everything should work as
usual, even if are using template inheritance!
However, note that streaming is not supported for PHP templates. Support
is impossible by design (as the layout is rendered after the main content).
Exceptions
----------
Exceptions thrown during rendering will be rendered as usual except that
some content might have been rendered already.
Limitations
-----------
As the getContent() method always returns false for streamed Responses, some
event listeners won't work at all:
* Web debug toolbar is not available for such Responses (but the profiler works fine);
* ESI is not supported.
Also note that streamed responses cannot benefit from HTTP caching for obvious
reasons.
---------------------------------------------------------------------------
by Seldaek at 2011/12/21 06:34:13 -0800
Just an idea: what about exposing flush() to twig? Possibly in a way that it will not call it if the template is not streaming. That way you could always add a flush() after your </head> tag to make sure that goes out as fast as possible, but it wouldn't mess with non-streamed responses. Although it appears flush() doesn't affect output buffers, so I guess it doesn't need anything special.
When you say "ESI is not supported.", that means only the AppCache right? I don't see why this would affect Varnish, but then again as far as I know Varnish will buffer if ESI is used so the benefit of streaming there is non-existent.
---------------------------------------------------------------------------
by cordoval at 2011/12/21 08:04:21 -0800
wonder what the use case is for streaming a response, very interesting.
---------------------------------------------------------------------------
by johnkary at 2011/12/21 08:19:48 -0800
@cordoval Common use cases are present fairly well by this RailsCast video: http://railscasts.com/episodes/266-http-streaming
Essentially it allows faster fetching of web assets (JS, CSS, etc) located in the <head></head>, allowing those assets to be fetched as soon as possible before the remainder of the content body is computed and sent to the browser. The end goal is to improve page load speed.
There are other uses cases too like making large body content available quickly to the service consuming it. Think if you were monitoring a live feed of JSON data of newest Twitter comments.
---------------------------------------------------------------------------
by lsmith77 at 2011/12/21 08:54:35 -0800
How does this relate the limitations mentioned in:
http://yehudakatz.com/2010/09/07/automatic-flushing-the-rails-3-1-plan/
Am I right to understand that due to how twig works we are not really streaming the content pieces when we call render(), but instead the entire template with its layout is rendered and only then will we flush? or does it mean that the render call will work its way to the top level layout template and form then on it can send the content until it hits another block, which it then first renders before it continues to send the data?
---------------------------------------------------------------------------
by stof at 2011/12/21 09:02:53 -0800
@lsmith77 this is why the ``stream`` method calls ``display`` in Twig instead of ``render``. ``display`` uses echo to print the output of the template line by line (and blocks are simply method calls in the middle). Look at your compiled templates to see it (the ``doDisplay`` method)
Rendering a template with Twig simply use an output buffer around the rendering.
---------------------------------------------------------------------------
by fabpot at 2011/12/21 09:24:33 -0800
@lsmith77: We don't have the Rails problem thanks to Twig as the order of execution is the right one by default (the layout is executed first); it means that we can have the flush feature without any change to how the core works. As @stof mentioned, we are using `display`, not `render`, so we are streaming your templates for byte one.
---------------------------------------------------------------------------
by fabpot at 2011/12/21 09:36:41 -0800
@Seldaek: yes, I meant ESI with the PHP reverse proxy.
---------------------------------------------------------------------------
by fabpot at 2011/12/21 09:37:34 -0800
@Seldaek: I have `flush()` support for Twig on my todo-list. As you mentioned, It should be trivial to implement.
---------------------------------------------------------------------------
by fzaninotto at 2011/12/21 09:48:18 -0800
How do streaming responses deal with assets that must be called in the head, but are declared in the body?
---------------------------------------------------------------------------
by fabpot at 2011/12/21 09:52:12 -0800
@fzaninotto: What do you mean?
With Twig, your layout is defined with blocks ("holes"). These blocks are overridden by child templates, but evaluated as they are encountered in the layout. So, everything works as expected.
As noted in the commit message, this does not work with PHP templates for the problems mentioned in the Rails post (as the order of execution is not the right one -- the child template is first evaluated and then the layout).
---------------------------------------------------------------------------
by fzaninotto at 2011/12/21 10:07:35 -0800
I was referring to using Assetic. Not sure if this compiles to Twig the same way as javascript and stylesheet blocks placed in the head - and therefore executed in the right way.
---------------------------------------------------------------------------
by fabpot at 2011/12/21 10:34:59 -0800
@Seldaek: I've just added a `flush` tag in Twig 1.5: 1d6dfad4f5
---------------------------------------------------------------------------
by catchamonkey at 2011/12/21 13:29:22 -0800
I'm really happy you've got this into the core, it's a great feature to have! Good work.
To stream a Response, use the StreamedResponse class instead of the
standard Response class:
$response = new StreamedResponse(function () {
echo 'FOO';
});
$response = new StreamedResponse(function () {
echo 'FOO';
}, 200, array('Content-Type' => 'text/plain'));
As you can see, a StreamedResponse instance takes a PHP callback instead of
a string for the Response content. It's up to the developer to stream the
response content from the callback with standard PHP functions like echo.
You can also use flush() if needed.
From a controller, do something like this:
$twig = $this->get('templating');
return new StreamedResponse(function () use ($templating) {
$templating->stream('BlogBundle:Annot:streamed.html.twig');
}, 200, array('Content-Type' => 'text/html'));
If you are using the base controller, you can use the stream() method instead:
return $this->stream('BlogBundle:Annot:streamed.html.twig');
You can stream an existing file by using the PHP built-in readfile() function:
new StreamedResponse(function () use ($file) {
readfile($file);
}, 200, array('Content-Type' => 'image/png');
Read http://php.net/flush for more information about output buffering in PHP.
Note that you should do your best to move all expensive operations to
be "activated/evaluated/called" during template evaluation.
Templates
---------
If you are using Twig as a template engine, everything should work as
usual, even if are using template inheritance!
However, note that streaming is not supported for PHP templates. Support
is impossible by design (as the layout is rendered after the main content).
Exceptions
----------
Exceptions thrown during rendering will be rendered as usual except that
some content might have been rendered already.
Limitations
-----------
As the getContent() method always returns false for streamed Responses, some
event listeners won't work at all:
* Web debug toolbar is not available for such Responses (but the profiler works fine);
* ESI is not supported.
Also note that streamed responses cannot benefit from HTTP caching for obvious
reasons.
Commits
-------
8710a13 Added example to the change log file
c9a2b49 Fixed xml encoder test script, and group `item` tags into an array
a0561e5 Replaced `item` with `*item` when parsing XML string
Discussion
----------
Replaced `item` with `*item` when parsing XML string
---------------------------------------------------------------------------
by fabpot at 2011/11/23 22:14:12 -0800
Tests do not pass:
1) Symfony\Tests\Component\Serializer\Encoder\XmlEncoderTest::testDecode
Failed asserting that two arrays are equal.
--- Expected
+++ Actual
@@ @@
'key2' => 'val'
- 'A B' => 'bar'
'Barry' => Array (...)
+ 'item' => Array (...)
)
'qux' => '1'
)
.../tests/Symfony/Tests/Component/Serializer/Encoder/XmlEncoderTest.php:173
---------------------------------------------------------------------------
by fabpot at 2011/11/24 22:57:37 -0800
I don't understand the patch anymore. I don't see any use of `*item` in the code.
---------------------------------------------------------------------------
by excelwebzone at 2011/11/24 23:04:07 -0800
I run some testing and you can't use '*item' XML parser reject it. So I modified it to convert it to an array.. Look at the test script change
---------------------------------------------------------------------------
by fabpot at 2011/11/24 23:13:30 -0800
So, you probably need to change the CHANGELOG as well? You should add an example which shows a before/after example.
---------------------------------------------------------------------------
by excelwebzone at 2011/11/24 23:15:51 -0800
Yes, forgot to change that..
---------------------------------------------------------------------------
by fabpot at 2011/11/25 01:27:42 -0800
ping @Seldaek, @lsmith77
---------------------------------------------------------------------------
by Seldaek at 2011/11/25 04:16:43 -0800
There are other meta-names available in the XmlEncoder, @-something for attributes, then there is something happening with a # but I'm not quite sure what. I'm just saying, maybe *item isn't the best name, if it introduces a third metacharacter. Apart from that I'm fine with it.
---------------------------------------------------------------------------
by excelwebzone at 2011/11/25 08:45:31 -0800
Maybe we can rename it to `wildcard` instead
---------------------------------------------------------------------------
by excelwebzone at 2011/11/25 15:12:09 -0800
Any chance we can push this throw?
---------------------------------------------------------------------------
by lsmith77 at 2011/11/27 04:06:25 -0800
here is the old PR #2682
@Seldaek: i think your comment was made for an older version of the patch.
overall I am fine with the change, the Serializer component takes a fairly simple approach. it is also not designed to really produce XML or JSON cleanly from the same data. it will really only be able to output a clean API for one or the other with the same data structure.
---------------------------------------------------------------------------
by excelwebzone at 2011/12/01 06:25:24 -0800
@fabpot can we merge this change
Commits
-------
413756c [BC break][SecurityBundle] Changed the way to register factories
Discussion
----------
[BC break][SecurityBundle] Changed the way to register factories
As discussed in #2454, this changes the way to register the factories to let each bundles register the factories it provides.
Commits
-------
d2195cc Fixed phpdoc and updated the changelog
9e41ff4 [SecurityBundle] Added a validation rule
b107a3f [SecurityBundle] Refactored the configuration
633f0e9 [DoctrineBundle] Moved the entity provider service to DoctrineBundle
74732dc [SecurityBundle] Added a way to extend the providers section of the config
Discussion
----------
[WIP][SecurityBundle] Added a way to extend the providers section of the config
Bug fix: no
Feature addition: yes
BC break: <del>no (for now)</del> yes
Tests pass: yes
This adds a way to extend the ``providers`` section of the security config so that other bundles can hook their stuff into it. An example is available in DoctrineBundle which is now responsible to handle the entity provider (<del>needs some cleanup as the service definition is still in SecurityBundle currently</del>). This will allow PropelBundle to provide a ``propel:`` provider for instance.
In order to keep BC with the existing configuration for the in-memory and the chain providers, I had to allow using a prototyped node instead of forcing using an array node with childrens. This introduces some issues:
- impossible to validate easily that a provider uses only one setup as prototyped node always have a default value (the empty array)
- the ``getFixableKey`` method is needed in the interface to support the XML format by pluralizing the name.
Here is my non-BC proposal for the configuration to clean this:
```yaml
security:
providers:
first:
memory: # BC break here by adding a level before the users
users:
joe: { password: foobar, roles: ROLE_USER }
john: { password: foobarbaz, roles: ROLE_USER }
second:
entity: # this one is BC
class: Acme\DemoBundle\Entity\User
third:
id: my_custom_provider # also BC
fourth:
chain: # BC break by adding a level before the providers
providers: [first, second, third]
```
What do you think about it ? Do we need to keep the BC in the config of the bundle or no ?
Btw note that the way to register the factories used by the firewall section should be refactored using the new way to provide extension points in the extensions (as done here) instead of relying on the end user to register factories, which would probably mean a BC break anyway.
---------------------------------------------------------------------------
by lsmith77 at 2011/10/23 09:19:23 -0700
i don't think we should keep BC. the security config is complex as is .. having BC stuff in there will just make it even harder and confusing.
---------------------------------------------------------------------------
by willdurand at 2011/10/23 09:41:25 -0700
Is the security component tagged with `@api` ?
So basically, we just have to create a factory (`ModelFactory` for instance) and to register it in the `security` extension, right ? Seems quite simple to extend and much better than the hardcoded version…
Why did you call the method to pluralize a key `getFixableKey` ?
---------------------------------------------------------------------------
by beberlei at 2011/10/23 14:48:26 -0700
Changing security config will introduce risk for users. We should avoid that
---------------------------------------------------------------------------
by stof at 2011/10/23 15:34:47 -0700
@beberlei as the config is validated, it will simply give them an exception during the loading of the config if they don't update their config.
---------------------------------------------------------------------------
by stof at 2011/10/24 01:01:42 -0700
@schmittjoh @fabpot Could you give your mind about it ?
---------------------------------------------------------------------------
by stof at 2011/10/31 17:08:12 -0700
@fabpot @schmittjoh ping
---------------------------------------------------------------------------
by stof at 2011/11/11 14:08:18 -0800
I updated the PR by implementing my proposal as the latest IRC meeting agreed that we don't need to keep the BC for this change. This allows to add the validation rule now.
---------------------------------------------------------------------------
by stof at 2011/11/16 11:16:06 -0800
@fabpot ping
---------------------------------------------------------------------------
by fabpot at 2011/11/16 22:29:05 -0800
@stof: Before merging, you must also add information about how to upgrade in the CHANGELOG-2.1.md file.
---------------------------------------------------------------------------
by stof at 2011/11/17 00:01:23 -0800
@fabpot done
The Firewall is now executed after the Router. This was needed to have access
to the locale and other request attributes that are set by the Router. This
change implies that all Firewall specific URLs have proper (empty) routes like
`/login_check` and `/logout`.
That allows projects that only use HttpFoundation and not HttpKernel to be able to
enforce the HTTP specification "rules".
$request = Request::createFromGlobals();
$response = new Response();
// do whatever you want with the Respons
// enforce HTTP spec
$response->prepare($request);
$response->send();
Within Symfony2, the prepare method is automatically called by the ResponseListener.
The locale management does not require sessions anymore.
In the Symfony2 spirit, the locale should be part of your URLs. If this is the case
(via the special _locale request attribute), Symfony will store it in the request
(getLocale()).
This feature is now also configurable/replaceable at will as everything is now managed
by the new LocaleListener event listener.
How to upgrade:
The default locale configuration has been moved from session to the main configuration:
Before:
framework:
session:
default_locale: en
After:
framework:
default_locale: en
Whenever you want to get the current locale, call getLocale() on the request (was on the
session before).