This PR was squashed before being merged into the master branch (closes#7559).
Discussion
----------
[HttpFoundation] [HttpKernel] Internal sub-requests should have X-Forwarded-For header providing real client IP
This is a better alternative to fix issue highlighted in #7554 and #7557.
| Q | A
| ------------- | ---
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | #7554, #7557
| License | MIT
When dealing with inline fragment renderer, it emulates an internal request by overriding the REMOTE_ADDR on Request. This is true, since conceptually request came from local server.
The problem that this introduces is that overriding the server value, it turns into an impossible state to retrieve the real client ip, only returning the local server IP (which is hardcoded to 127.0.0.1).
This patch takes the same approach as a Varnish call (it behaves the exact same way, reusing all code built for handling client ip handling on sub-requests), populating the X-Forwarded-For header and also making getClientIp smarter by removing possible local IP addresses from being considered as the client IP address.
Commits
-------
773e109 [HttpFoundation] [HttpKernel] Internal sub-requests should have X-Forwarded-For header providing real client IP
* 2.2:
[Config] #7644 add tests for passing number looking attributes as strings
[HttpFoundation][BrowserKit] fixed path when converting a cookie to a string
[BrowserKit] removed dead code
[HttpFoundation] fixed empty domain= in Cookie::__toString()
fixed detection of secure cookies received over https
[2.2] Pass ESI header to subrequests
[Translation] removed an uneeded class property
[Translation] removed unneeded getter/setter
[Translator] added additional conversion for encodings other than utf-8
fixed source messages to accept pluralized messages [Validator][translation][japanese] add messages for new validator
fix a DI circular reference recognition bug
[HttpFoundation] fixed the creation of sub-requests under some circumstances for IIS
Conflicts:
src/Symfony/Component/HttpFoundation/Tests/CookieTest.php
This PR was merged into the master branch.
Discussion
----------
[2.3] [WIP] Synchronized services...
| Q | A
| ------------- | ---
| Bug fix? | no
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | #5300, #6756
| License | MIT
| Doc PR | symfony/symfony-docs#2343
Todo:
- [x] update documentation
- [x] find a better name than contagious (synchronized)?
refs #6932, refs #5012
This PR is a proof of concept that tries to find a solution for some problems we have with scopes and services depending on scoped services (mostly the request service in Symfony).
Basically, whenever you want to inject the Request into a service, you have two possibilities:
* put your own service into the request scope (a new service will be created whenever a sub-request is run, and the service is not available outside the request scope);
* set the request service reference as non-strict (your service is always available but the request you have depends on when the service is created the first time).
This PR addresses this issue by allowing to use the second option but you service still always has the right Request service (see below for a longer explanation on how it works).
There is another issue that this PR fixes: edge cases and weird behaviors. There are several bug reports about some weird behaviors, and most of the time, this is related to the sub-requests. That's because the Request is injected into several Symfony objects without being updated correctly when leaving the request scope. Let me explain that: when a listener for instance needs the Request object, it can listen to the `kernel.request` event and store the request somewhere. So, whenever you enter a sub-request, the listener will get the new one. But when the sub-request ends, the listener has no way to know that it needs to reset the request to the master one. In practice, that's not really an issue, but let me show you an example of this issue in practice:
* You have a controller that is called with the English locale;
* The controller (probably via a template) renders a sub-request that uses the French locale;
* After the rendering, and from the controller, you try to generate a URL. Which locale the router will use? Yes, the French locale, which is wrong.
To fix these issues, this PR introduces a new notion in the DIC: synchronized services. When a service is marked as synchronized, all method calls involving this service will be called each time this service is set. When in a scope, methods are also called to restore the previous version of the service when the scope leaves.
If you have a look at the router or the locale listener, you will see that there is now a `setRequest` method that will called whenever the request service changes (because the `Container::set()` method is called or because the service is changed by a scope change).
Commits
-------
17269e1 [DependencyInjection] fixed management of scoped services with an invalid behavior set to null
bb83b3e [HttpKernel] added a safeguard for when a fragment is rendered outside the context of a master request
5d7b835 [FrameworkBundle] added some functional tests
ff9d688 fixed Request management for FragmentHandler
1b98ad3 fixed Request management for LocaleListener
a7b2b7e fixed Request management for RequestListener
0892135 [HttpKernel] ensured that the Request is null when outside of the Request scope
2ffcfb9 [FrameworkBundle] made the Request service synchronized
ec1e7ca [DependencyInjection] added a way to automatically update scoped services
HttpContentRenderer has been renamed to FragmentHandler.
The RendererStrategy subnamespace has been renamed to Fragment.
The strategy classes now have Fragment in their names.
ProxyRouterListener has been renamed to FragmentListener
The router_proxy configuration entry has been renamed to fragments.
