We previously used the tobscure/m.attrs.bidi github repo, but that repo was recently taken offline. We decided to integrate it as a util instead of publishing it as a separate package since we seem to be the only project using it, and adopting it into a new project requires barneycarroll/mattr, which does not seem to be used anywhere.
The code added here was taken from https://github.com/askvortsov1/m.attrs.bidi, a fork (without changes) of the tobscure repo. Support for alternative module systems and ways of registering bidi were removed, and the file was formatted in compliance with our prettier config.
Historically, extensions using subscribers has caused problems because subscribers were constructed/applied at extension boot. This caused some classes (e.g. UrlGenerator) to be resolved early, breaking parts of Flarum. For this reason, subscriber support wasn't included in the initial version of the Event extender.
However, updating extensions has shown that there is a legitimate use case for subscribers in organizing clean code; for instance, core's own `DiscussionMetadataUpdater`.
This commit introduces support for subscribers, but only applies them after the app has booted, which avoids the early resolution issues. Since event listeners/subscribers are only intended to be used with domain events, which would never be dispatched during app boot, the late activation of subscribers should not cause issue.
Under InnoDB, database entries created in transactions are not processed by fulltext indexes until the transaction is committed. To work around this, cases that test fulltext search have been split off into a separate class that adds and removes seed discussions/posts outside of transactions during setUp/tearDown.
Some tests need to change settings, but since MemoryCacheSettingsRepository caches settings in-memory, those changes aren't reflected. The new `purgeSettingsCache` removes it from the container, eliminating that cache.
For UserTest, we also need to regenerate the display name driver, since that's set statically on boot, before we'll get a change to clear the settings cache.
Before transactions, each test class would need to explicitly state starting state for permissions, which made the initial permission configuration somewhat arbitrary. Now, we might as well use the initial state of the default installation.
One of the User show_test tests has been commented out until
Previously, the `prepareDatabase` method would directly modify the database, booting the app in the process. This would prevent any extenders from being applied, since `->extend()` has no effect once the app is booted.
Since the new implementation of `prepareDatabase` simply registers seed data to be applied during app boot, the workaround of sticking this seed data into `prepDb` is no longer necessary, and seed data common to all test cases in a class can be provided in `setUp`.
When needed, app boot is explicitly triggered in individual test cases by calling `$this->app()`.
Making the $user argument nullable prevents this unnecessary exception, and doesn't introduce any issues since we check that $user exists as part of the method.
Fixes https://github.com/flarum/core/issues/1959
These transform lines are known to cause issues on iOS, and were added to hack around chrome issues that have since been fixed upstream.
Due to a commit by @fabpot in october, the mimetypes symfony class
now re-orders the shortened mimetypes that are returned when looking
up based on header mimetype. Our validator uses the first key, pops
the prefix off and then matches against our hardcoded array.
I've added a constraint to symfony/mime ^5.2.0 which ships with this change.
This constraint is fully compatible with our current lineup. In addition
I changed the hardcoded array to use the first entry from symfony mime types
now `jpg` instead of `jpeg`.
- Anchor scroll when inserting post placeholders
- Indicate that pages are loading at start of `loadPage`, which allows `onscroll` to not request that multiple pages be loaded at the same time
These changes are particularly applicable to firefox, where previously, dozens of posts could be skipped at a time if scroll up was held while at the top of the viewport.
Policy application has also been refactored, so that policies return one of `allow`, `deny`, `forceAllow`, `forceDeny`. The result of a set of policies is no longer the first non-null result, but rather the highest priority result (forceDeny > forceAllow > deny > allow, so if a single forceDeny is present, that beats out all other returned results). This removes order in which extensions boot as a factor.
