This PR was merged into the 3.4 branch.
Discussion
----------
[Cache] Accessing undefined constants raises an Error in php8
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
Calling `constant()` for an undefined constant will raise an `Error` on php 8. This PR adjust the Memcached tests to this new behavior.
Commits
-------
49fd0efb12 [Cache] Accessing undefined constants raises an Error in php8
This PR was merged into the 3.4 branch.
Discussion
----------
Add php 8 to travis
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | no
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
This PR adds an extra php 8 job to travis that is allowed to fail. This will allow us to monitor the remaining work.
* I did not include APCu because loading the extension would clutter our log with warnings. We might fix this by compiling APCu from master though.
* I did not manage to compile `ext-redis` and `ext-mongodb`.
* But hey, `ext-memcached` works!
Commits
-------
08084f370d Add php 8 to travis.
This PR was merged into the 3.4 branch.
Discussion
----------
Skip Doctrine DBAL on php 8 until we have a compatible version
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
This PR skips all tests on php 8 that would trigger an autoload on DBAL's `PDOConnection`. Those tests would exit with a fatal error because of incompatible method signatures (see doctrine/dbal#3791) and until we can test against a compatible version of DBAL, there's (at least from my point of view) nothing we can do about it. Once this issue is resolved, we can revert this change.
Commits
-------
1090738264 Skip Doctrine DBAL on php 8 until we have a compatible version.
This PR was merged into the 3.4 branch.
Discussion
----------
Address deprecation of ReflectionType::getClass()
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
Calling `ReflectionType::getClass()` will trigger a deprecation warning on php 8. This PR switches to `getType()` if available.
Commits
-------
53b1677a4e Address deprecation of ReflectionType::getClass().
This PR was merged into the 3.4 branch.
Discussion
----------
[VarDumper] ReflectionFunction::isDisabled() is deprecated
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
See php/php-src#5473. Calling `ReflectionFunction::isDisabled()` is pointless on php 8 and doing so triggers a deprecation warning.
Someone who is more familiar with that component might want to have a second look on this PR. I'm not really sure if this is the right way to fix the issue.
Commits
-------
1da347e5ba [VarDumper] ReflectionFunction::isDisabled() is deprecated.
This PR was merged into the 3.4 branch.
Discussion
----------
[Validator] Catch expected ValueError
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
`mb_check_encoding()` raises a `ValueError` on php 8 if an invalid encoding was passed. The validator that was patched here is expected to fail gracefully in that situation. This PR restores that behavior under php 8.
Maybe we can reconsider this behavior for Symfony 5.2. It feels a bit odd to me because we swallow a potential misconfiguration of the validator.
Commits
-------
8f3f67f82a [Validator] Catch expected ValueError.
This PR was merged into the 3.4 branch.
Discussion
----------
[DomCrawler] Catch expected ValueError
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
`mb_convert_encoding()` raises a `ValueError` on php 8 if an unknown character set is passed. This causes trouble with the whole test suite because the temporary error handler is not deregistered in that case.
Since the previously raised warning was actively ignored by that error handler, I'm now ignoring the `ValueError` as well. Also, I've wrapped the `restore_error_handler()` call into a `finally` statement to make the whole construct a bit more robust.
Commits
-------
b1db13728b [DomCrawler] Catch expected ValueError.
This PR was merged into the 3.4 branch.
Discussion
----------
[PropertyAccess] Parse php 8 TypeErrors correctly
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
The wording of `TypeError` messages has changed in php 8. Since `PropertyAccessor` parses those messages, the logic needed some adjustments.
Commits
-------
7100c3ce1b [PropertyAccess] Parse php 8 TypeErrors correctly.
This PR was merged into the 3.4 branch.
Discussion
----------
[Debug] php 8 does not pass $context to error handlers
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
php 8 will call error handlers without the optional `$context` parameter. Thus, error handlers that make that parameter mandatory will break.
Commits
-------
593897c9e1 [Debug] php 8 does not pass $context to error handlers.
This PR was merged into the 3.4 branch.
Discussion
----------
UploadedFile minor fixes
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no <!-- please update UPGRADE-*.md and src/**/CHANGELOG.md files -->
| Tickets | none <!-- prefix each issue number with "Fix #", if any -->
| License | MIT
| Doc PR | no
According to the [constructor assignments](cb7e78c809/src/Symfony/Component/HttpFoundation/File/UploadedFile.php (L60)), properties `$originalName` and `$mimeType` are never null in `UploadedFile`, so `getClientOriginalName()` and `getClientMimeType()` always return `string`. Also `$test` does not need a default value because it's always set in the constructor.
Commits
-------
eb8d626c27 Properties $originalName and $mimeType are never null in UploadedFile
This PR was merged into the 3.4 branch.
Discussion
----------
[Intl] Fix call to ReflectionProperty::getValue() for static properties
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
When we call `ReflectionProperty::getValue()` for a static property, we don't need to pass any arguments. The code that was fixed here raised a `TypeError` on php 8 because the argument has to be an object now.
Commits
-------
d4045897d6 [Intl] Fix call to ReflectionProperty::getValue() for static properties.
This PR was merged into the 3.4 branch.
Discussion
----------
[Config] Removed implicit cast of ReflectionProperty to string
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
PHP 8 does not allow casting `ReflectionProperty` instances to string anymore.
Commits
-------
8adbadede7 [Config] Removed implicit cast of ReflectionProperty to string.
This PR was merged into the 3.4 branch.
Discussion
----------
[Debug] Undefined variables raise a warning in php 8
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
`ErrorHandlerTest` executes code with undefined variables to test how the error handler handles the triggered errors. In php 8.0, the severity and error message for this class of errors has been changed. I've adjusted the test accordingly.
Commits
-------
1d20b514f2 [Debug] Undefined variables raise a warning in php 8.
This PR was merged into the 3.4 branch.
Discussion
----------
[HttpKernel] Prevent calling method_exists() with non-string values
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
On php 8, `method_exists()` will raise a `TypeError` if the first argument is neither an object or string. We do have a test case that calls the `ControllerResolver` with integer `1` as `$controller`. This test should work as expected again on php 8.
Commits
-------
1bbfdcbb8d [HttpKernel] Prevent calling method_exists() with non-string values.
This PR was merged into the 3.4 branch.
Discussion
----------
[Debug] Skip test that would trigger a fatal error on php 8
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | #36872
| License | MIT
| Doc PR | N/A
This PR skips a test of `DebugClassLoader`. The test uses incompatible method signatures in class inheritance to provoke a php warning that should be handled by the debug class loader. On php 8 however, this error is not recoverable anymore, so the tested logic will be obsolete there.
Commits
-------
573d0dd493 [Debug] Skip test that would trigger a fatal error on php 8.
This PR was merged into the 3.4 branch.
Discussion
----------
[PhpUnitBridge] fix installing under PHP >= 8
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets | -
| License | MIT
| Doc PR | -
[As experimented on Twig](b952011f95), using `simple-phpunit` is not simple enough when testing with PHP 8.
This PR fixes the issue so that we could remove these lines in Twig.
On 3.4 since supporting new versions of PHP is a bugfix according to our policies.
Commits
-------
5aa25ceb41 [PhpUnitBridge] fix installing under PHP >= 8
This PR was squashed before being merged into the 3.4 branch.
Discussion
----------
[Validator] Add missing translations of nn locale
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes/no
| New feature? | no
| Deprecations? | no
| Tickets | Fix#30178
| License | MIT
Added missing translations to validator with locale nn
Commits
-------
040d01e53b [Validator] Add missing translations of nn locale
This PR was squashed before being merged into the 3.4 branch.
Discussion
----------
[HttpKernel] Fix that the `Store` would not save responses with the X-Content-Digest header present
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no
| Deprecations? | no
| Tickets |
| License | MIT
| Doc PR |
Responses fetched from upstream sources might have a `X-Content-Digest` header, for example if the Symfony Cache is used upstream. This currently prevents the `Store` from saving such responses. In general, the value of this header should not be trusted.
As I consider this header an implementation detail of the `Store`, the fix tries to be local to that class; we should not rely on the `HttpCache` or other classes to remove untrustworthy headers for us.
This fixes the issue that when using the `HttpCache` in combination with the Symfony HttpClient, responses that have also been cached upstream in an instance of `HttpCache` are not cached locally. It adds the overhead of re-computing the content digest every time the `HttpCache` successfully re-validated a response.
Commits
-------
d8964fb8b7 [HttpKernel] Fix that the `Store` would not save responses with the X-Content-Digest header present
This PR was squashed before being merged into the 3.4 branch.
Discussion
----------
[Intl] bump icu 67.1
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | no
| New feature? | no
| Deprecations? | no
| Tickets | Fix #... <!-- prefix each issue number with "Fix #", if any -->
| License | MIT
| Doc PR | symfony/symfony-docs#... <!-- required for new features -->
mainly some new locales+scripts (see 3a3a9ba)
Commits
-------
29eb271184 [Intl] bump icu 67.1
This PR was merged into the 3.4 branch.
Discussion
----------
[HttpKernel] Fix error logger when stderr is redirected to /dev/null
| Q | A
| ------------- | ---
| Branch? | 3.4
| Bug fix? | yes
| New feature? | no <!-- please update src/**/CHANGELOG.md files -->
| Deprecations? | no <!-- please update UPGRADE-*.md and src/**/CHANGELOG.md files -->
| Tickets | n/a <!-- prefix each issue number with "Fix #", if any -->
| License | MIT
| Doc PR | n/a
The HttpKernel Logger is meant to be used as a last resort logging mechanism when no logger has been explicitly configured (Monolog is not a dependency for instance).
For small apps, that can be more than enough.
But under some circumstances, it does not work. When you are using PHP-FPM, `stderr` is ignored by default (`catch_workers_output` is `false`) and so, logs are ignored as well. There is no issue with the official PHP Docker image as the setting has been explicitly set to `true`. Not an issue with Symfony CLI as well, as we also change the setting. Not a problem either with the PHP built-in server as it does not use PHP FPM anyway.
But, in many other places, where the setting has its default value, logs are lost (as you can imagine, it happened to me). As this feature is meant to be a fallback, I think it should always work, or at least, we need to make everything possible to make it work out of the box; that's why I've considered it a bug and hence a PR on 3.4.
This PR changes the default value for the output to `null`, which uses `error_log()` instead of `stderr` to log errors. Why is it better? The output of `error_log()` is controllable by the `error_logs` PHP ini setting and it is well understood by everyone (the default configuration should always work well); so it should work in most/more cases.
The other change (to be discussed) is to also log messages at the `ERROR` level and not just the `CRITICAL` ones.
/cc @dunglas
Commits
-------
5f829bdaeb [HttpKernel] Fix error logger when stderr is redirected to /dev/null (FPM)
Nicolas Grekas
Alexander M. Turek
Fabien Potencier
Giuseppe Campanelli
Matthias Pigulla
Roland Franssen
vudaltsov