Refactor Frontend + Asset code
- Use Laravel's Filesystem component for asset IO, meaning theoretically
assets should be storable on S3 etc.
- More reliable checking for asset recompilation when debug mode is on,
so you don't have to constantly delete the compiled assets to force
a recompile. Should also fix issues with locale JS files being
recompiled with the same name and cached.
- Remove JavaScript minification, because it will be done by Webpack
(exception is for the TextFormatter JS).
- Add support for JS sourcemaps.
- Separate frontend view and assets completely. This is an important
distinction because frontend assets are compiled independent of a
request, whereas putting together a view depends on a request.
- Bind frontend view/asset factory instances to the container (in
service providers) rather than subclassing. Asset and content
populators can be added to these factories – these are simply objects
that populate the asset compilers or the view with information.
- Add RouteHandlerFactory functions that make it easy to hook up a
frontend controller with a frontend instance ± some content.
- Remove the need for "nojs"
- Fix cache:clear command
- Recompile assets when settings/enabled extensions change
* Run extenders exported by extensions
* Add some basic extenders
* Patch Mithril as the very first thing so extension code can run safely
* Load the payload into the app before booting extensions
* Setup default routes before booting extensions
In the future we may have multiple Formatters, so by moving the config
callback to its own instance method we can leave the constructor
available to specify which formatter (like Assets and Routes). This
format also allows for other methods to be added.
Additionally, this adds logic to automatically flush the Formatter cache
whenever the extension is enabled or disabled.
* Replace gulp with webpack and npm scripts for JS compilation
* Set up Travis CI to commit compiled JS
* Restructure `js` directory; only one instance of npm, forum/admin are "submodules"
* Refactor JS initializers into Application subclasses
* Maintain partial compatibility API (importing from absolute paths) for extensions
* Remove minification responsibility from PHP asset compiler
* Restructure `less` directory
Now we have two extenders:
- `Extend\LanguagePack` is the "convention over configuration" loader
for complete language packs.
- `Extend\Locales` can be used to load files (by locale) from a given
directory - useful for extensions that bring along their own locales
in multiple different languages.
Refs #851.
Turns out Container::call() does not work with invokable classes.
Thus, we need to wrap callables in a custom extender class to
support injecting any resolvable type-hint automatically.
Refs #851.
This simplifies the API and gives extension developers more
flexibility, for a) maintaining backwards compatibility, and
b) doing advanced stuff that extenders do not allow.
Note that only extenders are guaranteed to work across
different versions of Flarum (once the API surface is stable).
See the discussion in https://github.com/flarum/core/pull/1335.
This makes it more consistent with other existing extenders,
while also making registration of multiple routes more
comfortable for extension developers, and likely slightly
more performant. :-)
Loading the activated extensions now means retrieving an array of
extenders (classes that implement a certain type of extension of a core
feature in Flarum).
For now, the only existing extender is the Compat extender which is used
to handle old-style bootstrappers that simply return a closure that
receives all of its dependencies via auto injection.
In the future, extensions will be able to return an array of extender
instances from their bootstrapper instead. These extender classes will
be implemented in the next step.
- Use contextual namespaces within Flarum\Core
- Clean up and docblock everything
- Refactor Activity/Notification blueprint stuff
- Refactor Formatter stuff
- Refactor Search stuff
- Upgrade to JSON-API 1.0
- Removed “addedPosts” and “removedPosts” relationships from discussion
API. This was used for adding/removing event posts after renaming a
discussion etc. Instead we should make an additional request to get all
new posts
Todo:
- Fix Extenders and extensions
- Get rid of repository interfaces
- Fix other bugs I’ve inevitably introduced
A good start I think, but still some work to do. If we go ahead with
https://github.com/flarum/core/issues/132#issuecomment-117507974 (which
I am in favour of), we can extract the entity-related stuff into some
smaller service providers (e.g. discussion repo, an event listener,
permissions, and gambits stuff could all go in
Flarum\Core\Discussions\DiscussionsServiceProvider).
Type hinting User should take place in the callbacks. Theoretically
these traits could be used for another project now, where something
else has permissions (like a Sheep class, or a number)
Ditched the idea of having language packs as extensions. Reasoning:
1. Because we use machine keys for translations (rather than English
keys), extensions need to be able to define default translations. If
English translations are to be included in extensions and not in a
language pack extension, then it doesn’t make sense to have other
languages as language pack extensions. Inconsistency → complexity.
2. Translations should maintain version parity with their respective
extensions. There’s no way to do this if extension translations are
external to the extension.
Instead, localisation will be a core effort, as well as a per-extension
effort. Translators will be encouraged to send PRs to core + extensions.
In core, each locale has a directory containing three files:
- translations.yml
- config.js: contains pluralisation logic for the JS app, as well as
moment.js localisation if necessary
- config.php: contains pluralisation logic for the PHP app
Extensions can use the Flarum\Extend\Locale extender to add/override
translations/config to a locale.
Asset compilation has been completely refactored with a better
architecture. Translations + config.js are compiled and cached for the
currently active locale.
Originally the user activity feed was implemented using UNIONs. I was
looking at make an API to add activity “sources”, or extra UNION
queries (select from posts, mentions, etc.) but quickly realised that
this is too slow and there’s no way to make it scale.
So I’ve implemented an API which is very similar to how notifications
work (see previous commit). The `activity` table is an aggregation of
stuff that happens, and it’s kept in sync by an ActivitySyncer which is
used whenever a post it created/edited/deleted, a user is
mentioned/unmentioned, etc.
Again, the API is very simple (see Core\Activity\PostedActivity +
Core\Handlers\Events\UserActivitySyncer)
It turns out that the idea of “sending” a notification is flawed. (What
happens if the notification subject is deleted shortly after? The
notified user would end up with a dud notification which would be
confusing. What about if a post is edited to mention an extra user? If
you sent out notifications, the users who’ve already been mentioned
would get a duplicate notification.)
Instead, I’ve introduced the idea of notification “syncing”. Whenever a
change is made to a piece of data (e.g. a post is created, edited, or
deleted), you make a common notification and “sync” it to a set of
users. The users who’ve received this notification before won’t get it
again. It will be sent out to new users, and hidden from users who’ve
received it before but are no longer recipients (e.g. users who’ve been
“unmentioned” in a post).
To keep track of this, we use the existing notifications database
table, with an added `is_deleted` column. The syncing/diffing is
handled all behind the scenes; the API is extremely simple (see
Core\Notifications\DiscussionRenamedNotification +
Core\Events\Handlers\DiscussionRenamedNotifier)
- Automatically serialise the attribute
- Apply Permissible grant callbacks
Need to consider splitting the $permission property into two arguments
(currently have to explode by ‘.’)
