This PR was merged into the master branch.
Commits
-------
acfc750#2042 initial implementation of fatal error handler
Discussion
----------
Display traces for fatal errors
Bug fix: no
Feature addition: yes
Backwards compatibility break: no
Symfony2 tests pass: looks like yes
Fixes the following tickets: #2042 (partly)
License of the code: MIT
Output looks like on screen http://easycaptures.com/fs/uploaded/737/1191436899.png . I've added one line to css to prevent displaying standard xdebug trace http://easycaptures.com/fs/uploaded/737/5939488074.png
---------------------------------------------------------------------------
by Koc at 2012-11-08T21:55:41Z
So, community please advice me, how can I trigger `KernelEvents::EXCEPTION` event in `ErrorHandler` or `ExceptionHandler`? Or should I provide other event for this?
---------------------------------------------------------------------------
by stof at 2012-11-08T22:03:23Z
@Koc Don't. the exception handler is there to be the safe guard when developing, and does not depend on the kernel (which would be required to trigger the event). If you were triggering the listener again, it would mean that any exception thrown in a listener would lead to a loop.
And if it is for the fatal error handling, you simply cannot be sure the kernel is still available (and even less in a wokring state) at this point.
---------------------------------------------------------------------------
by Koc at 2012-11-08T22:06:31Z
But how can I notify logger (which will send me mail or just log this situation)?
---------------------------------------------------------------------------
by fabpot at 2012-11-09T07:33:41Z
The error handler is only registered when in debug mode in the Kernel and can be triggered very early in the handling of a request (even before we have access to the dispatcher or anything else). So, the current PR looks fine to me (apart from the typo and the lack of unit tests).
---------------------------------------------------------------------------
by Koc at 2012-11-09T09:13:03Z
> The error handler is only registered when in debug mode
Ooh! I haven't see that before. But the goal - be notified about errors by email or log-file. Like now exceptions with traces from site emails to me.
---------------------------------------------------------------------------
by fabpot at 2012-11-09T09:20:54Z
I think there are two goals. The first one being to have nice pages in the development environment when a fatal error occurs. And this PR addresses that feature quite nicely. The second can be addressed in another PR.
---------------------------------------------------------------------------
by henrikbjorn at 2012-11-14T11:50:22Z
I have some questions about the ErrorHandler. Is there a reason for it only to be registered in an debug environment (which prod is not). Would assume that if i enable the ErrorHandler in productions aswell Monolog would log thoose instead of them just vanishing?
---------------------------------------------------------------------------
by Koc at 2012-11-14T12:01:50Z
I am thinking about it too. But as Fabien says it will another PR
---------------------------------------------------------------------------
by GromNaN at 2012-11-18T10:38:09Z
You should add a memory reserve to be able to handle "Out of memory" errors.
An example is here :
513d628966/lib/Raven/ErrorHandler.php (L91)513d628966/lib/Raven/ErrorHandler.php (L62)
---------------------------------------------------------------------------
by fabpot at 2012-11-28T11:35:21Z
@Koc can you finish this PR (probably by integrating the memory reserve as explained by @GromNaN)?
---------------------------------------------------------------------------
by Koc at 2012-11-28T11:46:12Z
of course, on this weekend
---------------------------------------------------------------------------
by Koc at 2012-12-02T17:44:44Z
@fabpot done
* 2.1: (24 commits)
forced Travis to use source to workaround their not-up-to-date Composer on PHP 5.3.3
[Routing] removed irrelevant string cast in Route
Fixed typo
Make YamlFileLoader and XmlFileLoader file loading extensible
[HttpKernel] fix typo
Fixed singularization of "prices"
[Form] Removed an exception that prevented valid formats from being passed, e.g. "h" for the hour, "L" for the month etc.
[HttpKernel] fixed Client when using StreamedResponses (closes#5370)
fixed PDO session handler for Oracle (closes#5829)
[HttpFoundation] fixed PDO session handler for Oracle (closes#5829)
[Locale] removed a check that is done too early (and it is done twice anyways)
Update src/Symfony/Component/Validator/Resources/translations/validators.fa.xlf
Adding new localized strings for farsi validation.
[HttpFoundation] moved the HTTP protocol check from StreamedResponse to Response (closes#5937)
[Form] Fixed forms not to be marked invalid if their children are already marked invalid
[Form] Excluded some tests in NumberToLocalizedStringTransformerTest which fail on ICU 4.4, but work on ICU 4.8
added missing tests from previous merge
[Form] Fixed NumberToLocalizedStringTransformer to accept both comma and dot as decimal separator, if possible
Fix export-ignore on Windows
Show correct class name InputArgument in error message
...
Conflicts:
.travis.yml
src/Symfony/Component/Form/Extension/Core/DataTransformer/NumberToLocalizedStringTransformer.php
This PR was merged into the master branch.
Commits
-------
b27b749 made usage of Composer autoloader for subtree-split unit tests
Discussion
----------
made usage of Composer autoloader for subtree-split unit tests
This PR also normalizes the way components are tested.
---------------------------------------------------------------------------
by stof at 2012-11-09T23:14:22Z
👍
This PR was squashed before being merged into the master branch (closes#5381).
Commits
-------
0f3126f Added lockExists to Store interface, fixed locking bugs, added tests.
Discussion
----------
Added lockExists to Store interface, fixed locking bugs, added tests.
While working on Drupal's HttpCache implementation, I discovered that the base HttpCache class does an is_file to check for a lock, which assumes a file-based cache is being used. This seems like a mistake since the rest of the Store interface is easily swappable. I added a lockExists method so that this is properly abstracted.
I also noticed there were no tests for the change I made, so I added some very basic locking tests. While adding those I found that the existing lock method is a bit broken. This line here:
```php
<?php
if (false !== $lock = @fopen($path = $this->getPath($this->getCacheKey($request).'.lck'), 'x')) {
```
will return false if the file couldn't be written for any reason, but the rest of the method assumes that if $lock == false, the lock exists already. So if the file couldnt be written due to the parent directory not existing, $path will be returned as if it exists, which is clearly not the desired behavior.
I changed this to return false if the file couldnt be written and doesn't exist, $path if it exists, and true if the lock was created. It still doesn't feel great to have bool|string return values, but that's the best I could come up with atm. I also added a check for the parent directory that creates it if it doesn't exist. The new tests fail without it.
I also broke out that code a bit as it was very difficult to read.
---------------------------------------------------------------------------
by henrikbjorn at 2012-08-30T09:11:16Z
Symfony have a editorconfig file which set the correct indentation settings. http://editorconfig.org/
---------------------------------------------------------------------------
by msonnabaum at 2012-08-30T13:00:20Z
Updated based on stof's feedback.
---------------------------------------------------------------------------
by msonnabaum at 2012-08-30T13:21:40Z
Fixed based on code style feedback.
---------------------------------------------------------------------------
by jonathaningram at 2012-09-05T12:29:47Z
@msonnabaum, this seems to be distantly related to my recent PR too: #5376.
---------------------------------------------------------------------------
by stof at 2012-10-13T20:35:55Z
@fabpot anything left to merge this ?
---------------------------------------------------------------------------
by catch56 at 2012-10-23T16:42:10Z
This looks great to me, Couldn't find anything to complain about.
The ContainerAwareTraceableEventDispatcher class was tied to both the
Symfony container and the HttpKernel profiler. It made it non reusable
in another context.
The new TraceableEventDispatcher only keeps the HttpKernel profiler
integration and is able to wrap any other event dispatcher. It makes it
reusable in frameworks using the Symfony HttpKernel component like
Silex.
The only drawback is that we don't have access to the listener
priorities in the collected data anymore (but the listeners are still
ordered correctly). The change is still worth it I think.
This PR was merged into the master branch.
Commits
-------
7fe44da Whitespace corrections
6d30f20 Switched to using a method to get original class name that did not require string parsing
3c8d607 Changed test to use a longer form, complete check of the contents of the trace
de77c88 Whitespace correction
03a7bb9 Added a unit test to verify incomplete classes do not cause flatten exception to throw
e562418 Added a bit to convert incomplete objects in the error message
Discussion
----------
[HttpKernel] Added a bit to convert incomplete objects in the error message
Bug fix: yes
Feature addition: no
Backwards compatibility break: no
Symfony2 tests pass: yes*
\* no - errors in MongoDbSessionHandlerTest attempting to access private property, however running just src/Symfony/Component/HttpKernel does pass.
Objects of class __PHP_Incomplete_Class are only sometimes an object.
```
$object = unserialize('O:14:"BogusTestClass":0:{}');
$object instanceof __PHP_Incomplete_Class; // true
is_object($object); // false
gettype($object); // "object"
```
Since it is "not an object", the flatter attempts to turn it into a string, it triggers:
```
__PHP_Incomplete_Class could not be converted to string.
```
Which then hides the root error message.
---------------------------------------------------------------------------
by pborreli at 2012-09-18T16:16:33Z
have you seen that http://stackoverflow.com/questions/965611/forcing-access-to-php-incomplete-class-object-properties looks like you can still access the object even if it's a __PHP_Incomplete_Class with foreach
---------------------------------------------------------------------------
by rrehbeindoi at 2012-09-18T16:38:38Z
Thank you for the tip re: foreach.
* 2.1:
Added Base64 encoding, decoding to MongoDBProfilerStorage
Fix duplicated code and a field name
refactor src/Symfony/Component/Translation/Loader/MoFileLoader.php
fixed typo
Update src/Symfony/Component/Validator/Resources/translations/validators.pl.xlf
fixed issue #5596 (Broken DOM with the profiler's toolbar set in position top)
[Form] Fixed the testsuite for PHPUnit 3.6 as travis still uses it
added dirs generated by build-data.php in locale component to .gitignore
[Process] Fixed bug introduced by 7bafc69f38.
[Process][Tests] Prove process fail (Add more test case)
[Process][Tests] Prove process fail
[HttpFoundation] Fixed the tests
[DomCrawler] Added test for supported encodings by mbstring
[Config] Fixed preserving keys in associative arrays
[Console] Fixed return value for Command::run
[Locale] Fixed tests
[Console] Fix some input tests
[Filesystem] Fixed tests on Windows
[Config] Fixed tests on Windows
Because the name of the kernel is calculated in the constructor,
any child class that had overriden the kernel name, will be
ignored.
By setting the kernel name in the child class, we can avoid having
to execute the regex to calculate the name upon every construction
of a Kernel.
A test (and a kernel fixture) is added to prove that the override
works correctly.
Note: the Kernel API has not been touched, so there should be no
issues with BC.
Commits
-------
6de6806 [Component][HttpKernel] fixed testGetRootDir() on Win
Discussion
----------
[Component][HttpKernel] fixed testGetRootDir() on Win
Bug fix: yes
Feature addition: no
Backwards compatibility break: no
Symfony2 tests pass: [![Build Status](https://secure.travis-ci.org/gajdaw/symfony.png?branch=kernel_test_fix)](http://travis-ci.org/gajdaw/symfony)
Fixes the following tickets: -
Todo: -
License of the code: MIT
Documentation PR: -
Method `getRootDir()` returns path containing slashes. On Windows machine `__DIR__` and `DIRECTORY_SEPARATOR` are backslashes. To pass the test on win machine we have to translate `\` into `/`.
Commits
-------
88caf3a [HttpKernel] removed the storage of the current locale in the session
Discussion
----------
[HttpKernel] removed the storage of the current locale in the session
Before this commit, the current locale was stored in the session (if one
was already started). That way, for the next requests, even if the
request locale attribute was not set, the locale was "restored".
But this is a really bad practice as it means that the same URL can have
a different content depending on the previous requests. It would have
been better if the Vary header was set but the locale can be different
from the value coming from the Accept-Language anyway.
This is a BC break but fortunately, you can restore the 2.0 behavior by
creating a simple event listener that contains the logic removed by this
commit.
---------------------------------------------------------------------------
by travisbot at 2012-07-01T06:56:48Z
This pull request [fails](http://travis-ci.org/symfony/symfony/builds/1748659) (merged 009e30f0 into 2e356c1a).
---------------------------------------------------------------------------
by schmittjoh at 2012-07-01T08:15:46Z
How about using a cookie instead? It would remove the BC break, and also be possible to use a vary header?
---------------------------------------------------------------------------
by fabpot at 2012-07-01T09:13:44Z
The goal is to make Symfony as stateless as possible; introducing a cookie would defeat this goal.
---------------------------------------------------------------------------
by drak at 2012-07-01T09:19:37Z
@fabpot - thank you for bringing this to attention. I was meaning to do it a long time ago. The requested language is entirely a per request issue and must always be so. URLs must only ever return one content, and not multiple (e.g. different languages). The correct way to behave is to detect the language based on URL and failing that where a language is not requested, to look at the preferred language from the browser request and if available it can be redirected to that resource (e.g. /fr). This is what we do in Zikula. We have a further session based setting for "preferred language" which if set will override the browser default.
In summary:
1. If the language is specified in the GET request, return that language always. E.g. domain.com/fr/foo should return a French version of foo
2. If no language is specified in the GET request: first check the session for a preferred language, otherwise check the browser string for the preferred language and then if necessary, redirect to that resource. We have a setting which additionally say "always have language in URL, and don't put language code in URL for default language"
This means what in Zikula we only ever have one URL per language version of a page, but it still allows for users to set their preferred language which is taken in to account mainly when they visit the homepage (but in fact any page without a specific language in the request).
---------------------------------------------------------------------------
by drak at 2012-07-01T09:38:06Z
+1 on this PR. Basically the request locale should be in the Request object and calculated according to the applications preferences.
---------------------------------------------------------------------------
by schmittjoh at 2012-07-01T12:38:25Z
I agree that content must be detected based on the request, but I strongly disagree with relying entirely on the URL.
@fabpot, if you think about it using a cookie would still be stateless. There would be no state whatsoever, the detection would be entirely based on the request. Whether the language information is transmitted in the URL or as part of request headers is for the developer to decide eventually, at least IMO. My suggestion would just provide a default which is more BC.
---------------------------------------------------------------------------
by drak at 2012-07-01T20:08:50Z
@schmittjoh it's not entirely from the URL, there are browser preferences and also user defaults ca nalso available but the latter is slightly higher level. IMO it's not really Symfony's job here, it's application level specific. We have a pretty good working example of that in Zikula. Anyone can easily implement your own requirements with a listener.
What is absolutely clear however is it is wrong for one URL to deliver more than one version of any content.
---------------------------------------------------------------------------
by schmittjoh at 2012-07-01T21:16:52Z
I'm 100% for this change. My suggestion would just be more BC while still keeping Symfony2 stateless. Of course, it can be easily implemented in userland if we do not care about BC here.
Regarding different URLs per content, I do not think that this is our decision to make. Generally, developers should be able to make whatever content negotation they see fit. Whether they rely solely on the URL, or also take other request headers into account should not be limited by Symfony2.
---------------------------------------------------------------------------
by fabpot at 2012-07-02T10:37:26Z
I've added a paragraph in the UPGRADE file with a listener example that can be used to keep BC.
Before this commit, the current locale was stored in the session (if one
was already started). That way, for the next requests, even if the
request locale attribute was not set, the locale was "restored".
But this is a really bad practice as it means that the same URL can have
a different content depending on the previous requests. It would have
been better if the Vary header was set but the locale can be different
from the value coming from the Accept-Language anyway.
This is a BC break but fortunately, you can restore the 2.0 behavior by
creating a simple event listener that contains the logic removed by this
commit.
Commits
-------
7464dcd added phpdoc
c413e7b [Routing] remove RequestContextAwareInterface from RequestMatcherInterface
921be34 [Routing] fix phpdoc
Discussion
----------
[Routing] RequestMatcherInterface doesn't need context
Matchers that implement RequestMatcherInterface should match a Request, thus they don't need the request context.
---------------------------------------------------------------------------
by travisbot at 2012-06-14T21:39:48Z
This pull request [fails](http://travis-ci.org/symfony/symfony/builds/1624496) (merged f5ff1fe0 into 7c91ee57).
---------------------------------------------------------------------------
by schmittjoh at 2012-06-15T13:32:59Z
I think it makes sense to remove the RequestContext from the RequestMatcher.
---------------------------------------------------------------------------
by travisbot at 2012-06-15T15:54:28Z
This pull request [passes](http://travis-ci.org/symfony/symfony/builds/1628931) (merged 7464dcd2 into f881d282).
---------------------------------------------------------------------------
by Tobion at 2012-06-26T12:32:06Z
Anything missing?
- Missing dependency in one file
- Move helper class out of KernelTest because it implemented an
interface that depends on another component (thus would crash the
testsuite if invoked)
Commits
-------
c195957 [Components] Tests/Autoloading fixes
Discussion
----------
Fix components
See #4141
----
This PR:
* configures each component to use composer to manage "dev" dependencies instead of env variables;
* adds phpunit configuration file on Filesystem component;
* fixes READMEs.
It's mergeable without any problems, but I would recommend to wait a fix in Composer in order to use `self.version` in `require`/`require-dev` sections.
Note: I kept `suggest` sections because it makes sense but this PR doesn't aim to provide useful explanations for each entry. It could be another PR, not that one.
---------------------------------------------------------------------------
by willdurand at 2012-04-30T20:43:13Z
@fabpot I reviewed each component, one by one. Now `phpunit` always works, even if tests are skipped. A simple `composer install --dev` allows to run the complete test suite. Each commit is well separated from the others. I guess, everything is ok now.
---------------------------------------------------------------------------
by Tobion at 2012-04-30T20:47:00Z
Please squash, as it makes no sense to have the same commit for each component.
---------------------------------------------------------------------------
by fabpot at 2012-05-01T14:26:11Z
Can you squash your commits before I merge? Thanks.
---------------------------------------------------------------------------
by willdurand at 2012-05-01T14:29:38Z
done
---------------------------------------------------------------------------
by fabpot at 2012-05-01T15:48:25Z
It does not seem that the commits are squashed.
---------------------------------------------------------------------------
by willdurand at 2012-05-01T15:54:08Z
done
* Switched to Composer to manage "dev" dependencies
* Fixed READMEs
* Excluded vendor in phpunit.xml.dist files
* Fixed message in bootstrap.php files
* Added autoloader for the component itself