This PR was squashed before being merged into the 3.4 branch (closes#22629).
Discussion
----------
[Security] Trigger a deprecation when a voter is missing the VoterInterface
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | yes
| Tests pass? | yes
| Fixed tickets | ~
| License | MIT
| Doc PR | ~
Right now it's possible to add voters to the access decision manager that do not have a `VoterInterface`.
- No Interface, no `vote()` method, and it will give a PHP error.
- No Interface, but `vote()` method, it will still work.
- If I don't implement the interface _and_ have no `vote()` method, I will get weird exception that's not meaningful: `Attempted to call an undefined method named "vote" of class "App\Voter\MyVoter".`
This PR will deprecate the ability to use voters without the interface, it will also throw a proper exception when missing the interface _and_ the `vote()` method. Why when using and not when setting? Due to the fact that the voters can be set lazily via the `IteratorArgument`. The SecurityBundle will trigger a deprecation if the interface is not implemented and an exception if there's not even a `vote()` method present (to prevent exceptions at run-time).
This should have full backwards compatibility with 3.3, but give more meaningful errors. The only behavioral difference, might be that the container will throw an exception instead of maybe succeeding in voting when 1 voter would be broken at the end of the list (based on strategy). This case however, will be detected during development and deployment, rather than run-time.
Commits
-------
9c253e1ff6 [Security] Trigger a deprecation when a voter is missing the VoterInterface
Useful if an application provides their own base Controller that
references items in the container. It also makes it simpler for that
base controller to add additional optional dependencies by only overriding
getSubscribedServices instead of having to reimplement setContainer and
use ControllerTrait.
This PR was merged into the 2.7 branch.
Discussion
----------
Keep s-maxage when expiry and validation are used in combination
| Q | A
| ------------- | ---
| Branch? | 2.7
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets |
| License | MIT
| Doc PR |
(Symfony) docs say that [expiration wins over validation](https://symfony.com/doc/current/http_cache/validation.html). So,
a) when both the master and embedded response are public with an s-maxage, the result should be public as well and use the lower s-maxage of both, *also* in the case that the embedded response carries validation headers. (The cache may use those for revalidating the embedded response once it has become stale, but that does not impact expiration-based caching of the combined response.)
b) when both the master and embedded response are public with an s-maxage, the result should be public as well and use the lower s-maxage of both, *also* in the case that the master response carries validation headers. However, those *must not* be passed on to the client: They do not apply to the combined response, but may only be used by the cache itself to revalidate the (raw) master response.
Commits
-------
09bcbc70e7 Embedding a response that combines expiration and validation, that should not defeat expiration on the combined response
This PR was squashed before being merged into the 2.7 branch (closes#23129).
Discussion
----------
Fix two edge cases in ResponseCacheStrategy
| Q | A
| ------------- | ---
| Branch? | 2.7
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets |
| License | MIT
| Doc PR |
While reviewing how `ResponseCacheStrategy` calculates the caching-related headers for responses that embed subrequests, I came across two cases that I think are currently implemented incorrectly.
a) When the main response is public and cacheable with an expiration time, but it embeds (via ESI) a controller that does not set any caching-related headers, this embedded response is more constrained. So, the resulting (combined) response must not be cacheable, especially it may not keep the s-maxage.
b) When the main response is public and cacheable with an expiration time, but it embeds (via ESI) a controller that explicitly creates a "private" response, the resulting (combined) response must be private as well.
Commits
-------
c6e8c07e4d Fix two edge cases in ResponseCacheStrategy
This PR was merged into the 3.4 branch.
Discussion
----------
[Yaml] Deprecate using the non-specific tag
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | no
| New feature? | no <!-- don't forget updating src/**/CHANGELOG.md files -->
| BC breaks? | no
| Deprecations? | yes
| Tests pass? | yes
| Fixed tickets |
| License | MIT
| Doc PR |
Values tagged with the non-specific tag must not be transformed in an integer, this tag means that they must not be evaluated (see [the spec](http://www.yaml.org/spec/1.2/spec.html#tag/non-specific/)).
I applied this change in https://github.com/symfony/symfony/pull/22762 to comply with the spec.
Commits
-------
60f5046661 [Yaml] Deprecate using the non-specific tag
* 3.2:
[SecurityBundle] Move cache of the firewall context into the request parameters
Fix Usage with anonymous classes
[Workflow] Added more keywords in the composer.json
[Cache] APCu isSupported() should return true when apc.enable_cli=Off
[PropertyAccess] Do not silence TypeErrors from client code.
This PR was squashed before being merged into the 3.2 branch (closes#22943).
Discussion
----------
[SecurityBundle] Move cache of the firewall context into the request parameters
Following [this proposal](https://github.com/symfony/symfony/pull/22605#issuecomment-301276639). Since the matching context relates to the request, this information should have been cached inside the request parameters.
| Q | A
| ------------- | ---
| Branch? | 3.2
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | #22605
| License | MIT
| Doc PR | n/a
* Avoid memory leak when handling multiple requests
* Adding the new request parameter `_firewall_context` might be considered as a breaking change. That adds a new "public" property that could be used by end developers.
Commits
-------
b3203cb8ab [SecurityBundle] Move cache of the firewall context into the request parameters
This PR was squashed before being merged into the 3.3 branch (closes#23088).
Discussion
----------
[FrameworkBundle] Dont set pre-defined esi/ssi services
| Q | A
| ------------- | ---
| Branch? | 3.3
| Bug fix? | yes
| New feature? | no
| BC breaks? | not sure
| Deprecations? | no
| Tests pass? | yes/no
| Fixed tickets | #23080
| License | MIT
| Doc PR | symfony/symfony-docs#... <!--highly recommended for new features-->
It fixes the issue, but im not sure what's expected if you dont use http cache (solely enabled ssi/esi in config). Before the services were initialized, now they are synthetic as http cache sets them, but thats optional =/
Commits
-------
8c26aab0fe [FrameworkBundle] Dont set pre-defined esi/ssi services
This PR was squashed before being merged into the 2.7 branch (closes#23057).
Discussion
----------
[Translation][FrameworkBundle] Fix resource loading order inconsistency reported in #23034
| Q | A
| ------------- | ---
| Branch? | 2.7
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | #23034
| License | MIT
| Doc PR |
Fixes the bug reported in #23034:
When mixing `addResource()` calls and providing the `resource_files` option, the order in which resources are loaded depends on the `kernel.debug` setting and whether a cache is used.
In particular, when several loaders provide translations for the same message, the one that "wins" may change between development and production mode.
Commits
-------
2a9e65dea9 [Translation][FrameworkBundle] Fix resource loading order inconsistency reported in #23034
This PR was merged into the 3.4 branch.
Discussion
----------
Consistent error handling in remember me services
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | yes
| BC breaks? | yes
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | -
| License | MIT
RememberMeServices lacked consistent error handling so far making it impossible for implementors to e.g. maintain sufficiently detailed audit logs for remember me errors. Since remember me is a very sensitive area in any application, detailed logging is crucial.
The change proposed allows `loginFail` to optionally take the exception object as a second parameter and uses said exception consistently internally by calling `loginFail` instead of `cancelCookie`.
Commits
-------
eda1888f71 Consistent error handling in remember me services
This PR was squashed before being merged into the 2.7 branch (closes#23092).
Discussion
----------
[Filesystem] added workaround in Filesystem::rename for PHP bug
[Filesystem] added workaround in Filesystem::rename for https://bugs.php.net/bug.php?id=54097
Standard PHP rename() of dirs across devices/mounted filesystems produces confusing copy error & throws IOException in Filesystem::rename. I got it during console cache:clear in the Docker environment. This PR possible fixes https://github.com/symfony/symfony/issues/19851 and other environment related issues.
Workaround is on \rename() fails try to Filesystem::mirror & Filesystem::remove if $origin is directory
| Q | A
| ------------- | ---
| Branch? | 2.7
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets |
| License | MIT
| Doc PR |
Commits
-------
3ccbc479da [Filesystem] added workaround in Filesystem::rename for PHP bug
This PR was squashed before being merged into the 2.7 branch (closes#23123).
Discussion
----------
Add tests for ResponseCacheStrategy to document some more edge cases
| Q | A
| ------------- | ---
| Branch? | 2.7
| Bug fix? | no
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets |
| License | MIT
| Doc PR |
Adds some test cases for possible combinations of master/subrequest responses to better document behaviour in edge cases. Should now cover the entire `ResponseCacheStrategy`.
I hope 2.7 is the right target branch because having more tests for all releases should be a good thing™️.
Commits
-------
69e84633dd Add tests for ResponseCacheStrategy to document some more edge cases
This PR was merged into the 3.3 branch.
Discussion
----------
[HttpFoundation] add back support for legacy constant values
| Q | A
| ------------- | ---
| Branch? | 3.3
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets |
| License | MIT
| Doc PR |
Due to the data type change of the `Request::HEADER_` constants in Symfony 3.3 #23067 introduced a small BC break if someone used the old constant values statically instead of referring to the constants themselves.
Commits
-------
fddd754c0a add back support for legacy constant values
This PR was merged into the 2.7 branch.
Discussion
----------
[HttpFoundation] fix for Support for new 7.1 session options
| Q | A
| ------------- | ---
| Branch? | 2.7
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | #21606
| License | MIT
| Doc PR | n/a
Commits
-------
71c1b6f5bffixes#21606
This PR was merged into the 3.4 branch.
Discussion
----------
[Serializer] DateTimeNormalizer: allow to provide timezone
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | no
| New feature? | yes
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | N/A
| License | MIT
| Doc PR | N/A
My own use-case was for denormalization of a csv file provided by a third-party. The datetime format inside does not contain any timezone information, and won't change, but it's established to be UTC (or at least consistent).
So by providing the new `datetime_timezone` option, the returned instance of `\DateTime(Interface)` will properly be set with the expected timezone. (In case the format already supports the time offset, the provided timezone is ignored in favor of the one parsed by the `\DateTime` object)
Regarding normalization, the expected behavior of this feature is to consistently return the same time offset.
Commits
-------
c10a780afb [Serializer] DateTimeNormalizer: allow to provide timezone
This PR was merged into the 2.7 branch.
Discussion
----------
[VarDumper] fixes
| Q | A
| ------------- | ---
| Branch? | 2.7
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | #23115 and #23151 (hopefully)
| License | MIT
| Doc PR | -
it looks like stream_context_get_params can return false (looking at php-src and #23151)
and doing a foreach on SplObjectStorage changes its internal iterator state.
Commits
-------
d7238c9d96 [VarDumper] fixes
This PR was merged into the 3.4 branch.
Discussion
----------
[DI] Reference instead of inline for array-params
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | no
| New feature? | yes
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | -
| License | MIT
| Doc PR | -
- [x] Tests to be written.
This PR is part of my "container on a diet" quest.
When big array parameters are resolved, they create data duplication in the dumped container. This is even worse when the same big array parameters are used several times.
Even though OPcache stores static arrays in shared memory (php7), it does not deduplicate them (it does for strings.)
Instead of inlining arrays, this PR leverages the `$this->parameters` property when possible.
Commits
-------
7c3d0c7a46 [DI] Reference instead of inline for array-params
This PR was merged into the 3.2 branch.
Discussion
----------
[PropertyAccess] Do not silence TypeErrors from client code.
| Q | A
| ------------- | ---
| Branch? | 3.2
| Bug fix? | yes
| New feature? | no <!-- don't forget updating src/**/CHANGELOG.md files -->
| BC breaks? | no
| Deprecations? | no <!-- don't forget updating UPGRADE-*.md files -->
| Tests pass? | yes
| Fixed tickets |
| License | MIT
| Doc PR |
Fixes TypeError silencing in `setValue()` when said error is thrown inside setter/adder/etc.
An example is given in the included test, but more real-life story is botched accessors for a many-to-one association on a Doctrine entity:
```php
class B {
function setA(A $a) { ... } // forgotten "= null" here
}
class A {
function removeB(B $b) {
if ($this->bs->contains($b)) {
$this->bs->removeElement($b);
$b->setA(null); // TypeError thrown
}
return $this;
}
}
```
No error is shown to the user, even though removing doesn't work.
This bug is not present in 2.7 & 2.8.
Commits
-------
45b961de2e [PropertyAccess] Do not silence TypeErrors from client code.
This PR was squashed before being merged into the 3.4 branch (closes#23154).
Discussion
----------
[WebProfilerBundle] Sticky ajax window
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | no
| New feature? | yes
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | #23125
| License | MIT
| Doc PR | -
This toggles the ajax toolbar block being sticky on click. I find it quite useful in heavy ajax apps :)
Not sure the state needs to be persisted in local storage or so, could be done :) however for our app all the navigating happens via react router, hence no real need for us to persist it between requests.
Commits
-------
b92929b0be [WebProfilerBundle] Sticky ajax window
This PR was merged into the 3.2 branch.
Discussion
----------
[PropertyAccess] Fix Usage with anonymous classes
| Q | A
| ------------- | ---
| Branch? | 3.2
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | #23136
| License | MIT
Replace forbidden characters in the the class names of Anonymous Classes in form of
"class@anonymous /symfony/src/Symfony/Component/PropertyAccess/Tests/PropertyAccessorTest.php0x7f3f5f267ad5"
Wrapped in eval to avoid PHP parsing errors < 7 and using `rawurlenceode` for perf reasons
Thanks @nicolas-grekas for the help and patience. Let me know if anything is missing.
Commits
-------
3f7fd432df Fix Usage with anonymous classes
This PR was merged into the 3.3 branch.
Discussion
----------
[Config] Fix ** GlobResource on Windows
| Q | A
| ------------- | ---
| Branch? | 3.3
| Bug fix? | yes
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | #23103
| License | MIT
| Doc PR | -
We cannot tell Finder to use RecursiveDirectoryIterator::UNIX_PATHS so we have to fix paths on Windows.
Commits
-------
44955bea53 [Config] Fix ** GlobResource on Windows
This PR was merged into the 3.3 branch.
Discussion
----------
Fix AutowiringTypesTest transient tests
| Q | A
| ------------- | ---
| Branch? | 3.3
| Bug fix? | no
| New feature? | no
| BC breaks? | no
| Deprecations? | no
| Tests pass? | yes
| Fixed tickets | -
| License | MIT
| Doc PR | -
In 3.3, we have a race condition because FrameworkBundle and SecurityBundle use the same test folder for their respective AutowiringTypesTest.
Commits
-------
408e56e404 Fix AutowiringTypesTest transient tests
* 3.3:
[TwigBundle] Add Doctrine Cache to dev dependencies
[Yaml] Fix linting yaml with constants as keys
[Routing] Revert the change in [#b42018] with respect to Routing/Route.php
