Because extensions can have dependencies injected, a RouteCollection could potentially be instantiated, and thus the ConfigureRoutes event would be called before extensions have had a chance to subscribe to it. Instead, we instantiate the RouteCollection on demand, but only populate it when the application boots.
- Use Symfony's Session component to work with sessions, instead of a custom database model. Separate the concept of access tokens from sessions once again.
- Extract common session/remember cookie logic into SessionAuthenticator and Rememberer classes.
- Extract AuthenticateUserTrait into a new AuthenticationResponseFactory class.
- Fix forgot password process.
- Use cookies + CSRF token for API authentication in the default client. This mitigates potential XSS attacks by making the token unavailable to JavaScript. The Authorization header is still supported, but not used by default.
- Make sensitive/destructive actions (editing a user, permanently deleting anything, visiting the admin CP) require the user to re-enter their password if they haven't entered it in the last 30 minutes.
- Refactor and clean up the authentication middleware.
- Add an `onhide` hook to the Modal component. (+1 squashed commit)
Specifically, the crash would occur when the first notification had a subject without a discussion relationship (e.g. the Subscriptions extension's newPost notification, where the subject itself was a discussion). Instead of simply eager loading the nested subject.discussion relationship, we load discussions manually instead.
This means that the "Start a Discussion" button will still be enabled if the user is not allowed to start globally, but only in certain tags.
Also add some other stuff to the changelog.
closes#640
This affects version numbers, extensions and paths, which might be
skeleton-specific. This commit moves those hardcoded values out of
the classes and instead injects them through the constructor. This
way, all prerequisites can be configured in the service provider.
Even when no language packs are enabled, a forum-en-xxx.js (or whatever the default locale is) file is still generated because other extensions may contain translations. But when enabling the English language pack, since no locales are registered with the LocaleManager, that file doesn't get flushed and therefore doesn't get regenerated with the English translations. This fix always registers the default locale with the LocaleManager so that's not the case.
Since we'll be encouraging (requiring?) people to include an "en" translation in extensions they want to put on the Marketplace, we should have a further fallback to "en" if it can't find translations in the forum's default language. That way if people only use extensions from the Marketplace, they'll hardly ever see any key names.
Thanks to @dcsjapan for the suggestion.
The default XHR error handler produce an alert which is appropriate to the response status code. It can be overridden per-request (by specifying the `errorHandler` option) so that the alert can be suppressed or displayed in a different position (e.g. inside a modal).
ref #118
If the version in the settings table mismatches the code version, then we return a 503 error for all requests coming through index.php and api.php, while admin.php serves up a form prompting for the database password which will run outstanding migrations.
We now use Symfony's Translation component. Yay! We get more powerful pluralisation and better a fallback mechanism. Will want to implement the caching mechanism at some point too. The API is replicated in JavaScript, which could definitely use some testing.
Validators have been refactored so that they are decoupled from models completely (i.e. they simply validate arrays of user input). Language packs should include Laravel's validation messages.
ref #267
Core assets are copied into the root/assets directory on installation.
The contents of an "assets" directory within any extension is copied into root/assets/extensions/{name}/ whenever the extension is enabled, and deleted whenever the extension is uninstalled.
Still needs to be refactored
This means that the expensive minification process will only be run for a file if it hasn't before. Greatly speeds up extension enabling/disabling.
Also:
- Don't check file last modification times in production for a bit of extra perf.
- Only flush CSS when theme settings are changed. This speeds up the page reload a bit.
Falls back to a less effective minification library if ClosureCompilerService errors or is unavailable. Minification takes a while (20 seconds or so), but it only happens when assets are modified. Still, this means enabling/disabling extensions is taking far too long. Possible solutions:
- Don't minify initially; set a process running in the background to do minification, and server unminified assets in the meantime.
- Refactor compiler to send each JS file to CCS individually, only if that particular file has been modified.
flarum/gulp has also been updated to no longer support uglification.
closes#582
- Reorganised all namespaces and class names for consistency and structure. Following PSR bylaws (Abstract prefix, Interface/Trait suffix).
- Move models into root of Core, because writing `use Flarum\Core\Discussion` is nice. Namespace the rest by type. (Namespacing by entity was too arbitrary.)
- Moved some non-domain stuff out of Core: Database, Formatter, Settings.
- Renamed config table and all references to "settings" for consistency.
- Remove Core class and add url()/isInstalled()/inDebugMode() as instance methods of Foundation\Application.
- Cleanup, docblocking, etc.
- Improvements to HTTP architecture
- API and forum/admin Actions are now actually all the same thing (simple PSR-7 Request handlers), renamed to Controllers.
- Upgrade to tobscure/json-api 0.2 branch.
- Where possible, moved generic functionality to tobscure/json-api (e.g. pagination links). I'm quite happy with the backend balance now re: #262
- Improvements to other architecture
- Use Illuminate's Auth\Access\Gate interface/implementation instead of our old Locked trait. We still use events to actually determine the permissions though. Our Policy classes are actually glorified event subscribers.
- Extract model validation into Core\Validator classes.
- Make post visibility permission stuff much more efficient and DRY.
- Renamed Flarum\Event classes for consistency. ref #246
- `Configure` prefix for events dedicated to configuring an object.
- `Get` prefix for events whose listeners should return something.
- `Prepare` prefix when a variable is passed by reference so it can be modified.
- `Scope` prefix when a query builder is passed.
- Miscellaneous improvements/bug-fixes. I'm easily distracted!
- Increase default height of post composer.
- Improve post stream redraw flickering in Safari by keying loading post placeholders with their IDs. ref #451
- Use a PHP JavaScript minification library for minifying TextFormatter's JavaScript, instead of ClosureCompilerService (can't rely on external service!)
- Use UrlGenerator properly in various places. closes#123
- Make Api\Client return Response object. closes#128
- Allow extensions to specify custom icon images.
- Allow external API/admin URLs to be optionally specified in config.php. If the value or "url" is an array, we look for the corresponding path inside. Otherwise, we append the path to the base URL, using the corresponding value in "paths" if present. closes#244
Spent quite a while looking into the best solution here and ended up going with three separate classes. Thanks to @Luceos for the PR that got this rolling (#518). My reasoning is:
- The task of routing and URL generation is independent for each section of the app. Take Flarum\Api\Users\IndexAction for example. I don't want to generate a URL to a Flarum route... I specifically want to generate a URL to an API route. So there should be a class with that specific responsibility.
- In fact, each URL generator is slightly different, because we need to add a certain prefix to the start (e.g. /api)
- This also allows us to get rid of the "flarum.api" prefix on each route's name.
- It's still DRY, because they all extend a base class.
At the same time, I could see no reason this needed to be "interfaced", so all of the classes are concrete.
Goes a long way to fixing #123 - still just a few places left remaining with hardcoded URLs.
Previously, clicking the "mark all notifications as read" button would individually mark each of the visible notifications as read. Since we now always show a badge with the number of unread notifications, we need to make sure that all notifications (not just the visible ones) can be marked as read. Otherwise it would be possible to get stuck with an unread badge there.
This commit adds a new API endpoint which marks *all* of a user's notifications as read. The JSON-API spec doesn't cover this kind of thing (updating all instances of a certain resource type), so I'm a bit unsure regarding what the endpoint should actually be. For now I've gone with POST /notifications/read, but I'm open to suggestions.
ref #500
This is useful for very simple extensions like language packs, because it means no Composer/namespacing and thus bootstrap.php doesn't have to be changed at all.
Also add an API to let extensions define additional default route
options.
Allowing default routes with parameters (e.g. /d/123) is very difficult
because of the way Mithril routing works, and it doesn't have a
convincing use-case to justify the trouble. So I've removed the custom
input altogether.
closes#427
Core migrations are under the Flarum\Migrations\Core namespace.
Extension migrations must be under the
Flarum\Migrations\{ExtensionName} namespace.
closes#422
Some providers (e.g. Twitter) don't expose user email addresses, so it
turns out we can't use that as the sole form of identification/account
matching.
This commit introduces a new `auth_tokens` table which stores arbitrary
attributes during the sign up process. For example, when Twitter is
authenticated, a new auth token containing the user's Twitter ID will
be created. When sign up is completed with this token, that Twitter ID
will be set as an attribute on the user's account.
Anti-spam extensions may automatically hide the first post in a
discussion, and thus we had to implement smarter permissions so
discussions with zero posts wouldn't be visible to users other than the
author/mods. This change allows those hidden posts to be restored again.
Failure is silent for now... The default LESS will compile without the
invalid customisations. Not sure if we should log an error somewhere
and display it on the admin page?
closes#400
- Allow users to see their own posts, even if they have been hidden by
someone else
- Don't require hiding a post to be necessarily attributed to a user
- Hide discussions with zero posts, unless the user can edit posts, or
they are the discussion author
We can't do this using the ValidatesBeforeSave trait because the
password has been hashed by then. Instead, we must validate the
original password as it comes in.