In jQuery file upload land, we were sending a single file through
at a time to matching upload handlers. This in turn required plugin
authors to marshal the files as they came through one by one if they
wanted to group them together to do something with them. Now that
we are using uppy, files come through in the groups they are added
in (for example dropping multiple, selecting multiple from the system
file dialogue).
This commit changes the matching upload handlers to send through
all matching files at once instead of piecemeal.
Error introduced in #14781
```
Error: Assertion Failed: You attempted to update <(unknown):ember3217>.bookmarks to "<(unknown):ember3846>", but it is being tracked by a tracking context, such as a template, computed property, or observer. In order to make sure the context updates properly, you must invalidate the property when updating it. You can mark the property as `@tracked`, or use `@ember/object#set` to do this.
```
In f6528afa01 I added parity support
for composer upload handlers to the uppy-ized composer. However the
way I assumed that it was only possible to handle a single file
upload at a time was false; it only appeared this way in the old
jQuery file upload composer because jQuery file upload sent through
files one at a time even if multiple were added at once. This caused
issues in certain plugins and themes by third parties.
This commit fixes the issue by making the uppy upload handler work
the same as the old one, by capturing all of the added files that
have matching handlers then going through them one by one and passing
them to the handler function.
For widget event handlers, we register a single listener on the `<body>`, and then notify the relavent widget (if any) when the event fires.
`touchstart` and `touchmove` events are particularly performance sensitive because they block scrolling on mobile. Therefore we want to avoid registering global non-passive listeners for these events.
This commit updates the WidgetTouchStartHook and WidgetDragHook implementations to automatically register listeners on the specific widget DOM elements when required.
This commit removes the last global scroll-blocking event handler from Discourse core. That means that mobile scrolling is now completely decoupled from our JS app. Even if the JS app is completely blocked (e.g. during rendering), scrolling will now continue to work. This should make things feel a lot smoother, especially on lower performance devices.
These were set to `passive: true` in ff72522f.
However, two consumers of this mixin (topic-navigation and site-header) do need to call `e.preventDefault()`, so we can't use passive listeners here.
That's ok, because this mixin only applies to a specific component's element, not the entire page. So having these non-passive listeners doesn't affect the vast majority of scrolling
This mixin calls the "scrolled" method of some object with no parameters, so there is no way that consumers would ever call `event.preventDefault()`. Therefore we can make the listeners passive, and improve scrolling performance on mobile.
This commit also updates the mixin to remove JQuery usage. The API is slightly modified to remove the need for an event 'name' for binding/unbinding.
The calls to `.bindScrolling` and `.unbindScrolling` in user-stream.js are removed because they are already called by the LoadMore mixin which is applied to the component.
The `bindScrolling` method claimed to offer debouncing-by-default. However, a bug in the `opts` parsing meant that debouncing was skipped if a 'name' was passed in. Therefore the only consumer actually being debounced was the LoadMore mixin. This commit fixes the opts parsing, so all consumers get the same behavior.
However, when scrolling, debounce is rarely what we want. The documentation of `bindScrolling` says "called every 100ms". In fact, debounce means that the functions were only called 'after the user **stops scrolling** for 100ms'. If you're scrolling very slowly (e.g. when using momentum-based scrolling on mobile), then this can be quite frustrating. This is why "Load more" is only triggered on topics/topic-lists when you completely stop scrolling.
Therefore, this commit also replaces the default 'debounce' with a 'throttle'. The 'throttle' is configured with `immediate = false`, so that it fires on the trailing edge, and therefore the final call will always be **after** we finish scrolling. (the default `immediate: true` would fire on the leading edge, and so the last call could be up to 100ms **before** we finish scrolling).
Registering non-passive listeners for the touchstart event can affect scroll performance on mobile devices, and now shows a warning in Chrome. Our current version of Ember unconditionally registers all event listeners, even if they're unused. It also doesn't support passive event listeners. Once we get to Ember 4.0, it lazily registers event listeners, and supports passive listeners via the `{{on` helper.
We already disable the ember `mousemove` and `touchmove` events for performance, so it makes sense to do the same for `touchstart`. We are not using `touchstart` anywhere in core, and I cannot find any official/unofficial plugins which use it. If a `touchstart` event is required, plugins/themes can always register their own listeners (preferably on a specific element, rather than the whole `document`)
We do call `event.preventDefault()` on these events. They're limited to a single element, so performance impact should be negligable. Adding `passive: false` prevents the chrome dev tools warning.
None of these places call `event.preventDefault()`. Therefore we can register the event listeners as 'passive', and improve scroll performance on mobile devices.
Calling `setProperty("--header-offset", newValue)` will always cause a 'Recalculate Style' event, even if the value is unchanged. On my browser, these 'Recalculate Style' events take about 6-7ms each time the `dockCheck` function is run.
This commit stores the 'previous' value in an instance variable, and only calls setProperty if the value has changed. This brings the total runtime of `dockCheck` down to about 70µs on my machine.
The theme creator endpoints return JSON with a 201 CREATED
status code. With the recent changes to bootstrap-json/index.js
for the Ember CLI proxy, these endpoints were broken because
201 was not an accepted status code. This commit simply
adds 201 to the array, but prettier forced a reformat as well!
- Remove JQuery
- Remove legacy `document.webkitHidden` support. None of our currently supported browsers need this
- Use `passive` event listeners. These allows the browser to process the events first, before passing control to us
- Add a new `unseenTime` parameter. This allows consumers to request a delay before being notified about the browser going into the background
- Add a method for removing a callback
- Fire the callback when presence changes in either direction. Previously it would only fire when the user becomes present after a period of inactivity.
- Ensure callbacks are only called once for each state change. Previously they would be called every 60s, regardless of the value
- Listen to the `visibilitychanged` and `focus` events, treating them as equivalent to user action. This will make messagebus re-activate more quickly when switching back to a stale tab
- Add test helpers
- Delete the unused `discourse/lib/page-visible` module.
- Call message-bus's onVisibilityChange API directly, rather than dispatching a fake event on the `document`
Use @here to mention all users that were allowed to topic directly or
through group, who liked topics or read the topic. Only first 10 users
will be notified.
c401d641 introduced a new translation key for auth providers, and provided new strings for core providers. However, not all plugins have added this string. This commit makes the screenreader title fallback to the regular title in those cases.
This commit adds handlers for the composer uppy mixin to allow
for cancelling individual file uploads, not just all of them
at once. This is also combined with better tracking of in progress
uploads along with their progress percentage, for UI that needs
to be able to display the progress for individual files and
also cancel individual files.
To use this, a cancel button in the UI should call a function like this:
```javascript
cancelSingleUpload(fileId) {
this.appEvents.trigger(`${this.eventPrefix}:cancel-upload`, {
fileId,
});
},
```
Additionally, the `inProgressUploads` can be shown in the UI. It is an array of objects with the file name, ID, and the progress percentage. We can add more data to this if needed down the line.
This takes the uppy chunking algorithm and combines it with some
form submission from resumable.js for parity with the current
backup controller to make local backup uploads work with uppy.
We can then use this to replace the resumable-upload component
and the resumable.js library from our codebase, once stable.
This is disabled by default, so people using local backups will not
be affected. The enable_experimental_backup_uploader site setting
must be enabled for this to work.
When rendering the markdown code blocks we replace the
offending characters in the output string with spans highlighting a textual
representation of the character, along with a title attribute with
information about why the character was highlighted.
The list of characters stripped by this fix, which are the bidirectional
characters considered relevant, are:
U+202A
U+202B
U+202C
U+202D
U+202E
U+2066
U+2067
U+2068
U+2069
In the topic lists, it's important that we apply `pointer-events: none;` to the links. 0e371d4 updated the selector used for this css.
In `templates/list/topic-list-item.hbs`, `.main-link` is applied to the same element as `.topic-list-data`, so the new selector applied correctly.
In `templates/mobile/list/topic-list-item.hbr`, `.main-link` is nested within `.topic-list-data`, so the new selector did not apply correctly.
This commit switches the selector back to simply `.main-link`, so that it works for both mobile and desktop.
Calling `window.getComputedStyle` during initialization causes the browser to pause and 'Recalculate Style'. On my machine, this adds about 7ms to boot time. Instead, we can check for the `rtl` class on the html element, which is added by the server, and doesn't require computing styles.
This fixes an issue CvX found on PR #14666 where a previous fix
overwrote a computed property.
The better fix (as is often the case with Ember) is to remove an
observer and call methods when things change ourselves.
* FEATURE: Always show advanced invite options
The UI is more simple and more efficient than how it was when the
advanced options toggle was introduced. It does not make sense to keep
it anymore.
* UX: Minor copy edits
* UX: Merge expire invite controls
There were two controls in the create invite modal. One was a static
text that displayed how much time is left until the invite expires. The
other one was a datetime selector that set the time the invite expires.
This commit merges the two controls in a single one: staff users will
continue to see the datetime selector without the static text and
regular users will only see the static text because they cannot set
when the invite expires.
* UX: Remove invite link
It should only be visible after the invite was created.
Time spent in the 'find module with suffix' portion of our `customResolve` function were adding up to around 100ms-150ms when booting the app. This time is spread over 150+ calls, so it's not immediately obvious in flamegraphs.
This commit implements a (reversed) [Trie](https://en.wikipedia.org/wiki/Trie) which enables fast suffix-based lookups on a list of strings.
In my tests, this requires < 5ms to initialize, and brings the cumulative 'find module with suffix' time down to `< 5ms`. This corresponds to a ~100ms improvement in LCP metrics in my browser.
The only behavior change is to remove support for module filenames which are **not** dasherized. I haven't found any core/theme/plugin modules which are not dasherized in their filenames.
This commit refactors the direct external upload routes (get presigned
put, complete external, create/abort/complete multipart) into a
helper which is then included in both BackupController and the
UploadController. This is done so UploadController doesn't need
strange backup logic added to it, and so each controller implementing
this helper can do their own validation/error handling nicely.
This is a follow up to e4350bb966
* FIX: register customOptions as select kit filter
We are allowing plugins to define custom filters which are added to CUSTOM_USER_SEARCH_OPTIONS const. However, we need to have static placeholder for custom filters, so those props will be passed, and we can use it later.
* fix
* FEATURE: display warning when sharing a topic in a restricted category
If a topic belongs to a category that is not readable by everyone, display a text warning of "Only visible to members of groups: [group_a], [group_b]"
* DEV: Adding a new category means we need to bump this value
* DEV: pass category to showModal
- ensures arrow up/down doesn’t also apply to textarea while autocomplete is opened
- ensures esc is closing autocomplete and also not closing composer while autocomplete is opened
- Removes jquery
- Removes a not unregistered listener and uses component event
- Removes external-url class as it was only valid in one case of the dropdown
- Uses @action
- Tagless
- Other minor changes
After permanently deleting the first post of a topic the user was
sometimes stuck on the page because of an infinite loop. This problem
happened more often in Firefox.
This patch takes the small component we had for sticky avatars and adds
it into our core code base.
A small refactor has been made to have a `StickyAvatars` dedicated class.
When uploading emoji with the new uppy upload mixin, we were
not sending the name of the emoji in the payload, or more
accurately uppy was already using the file name as the name
value and we were not overriding it from data. This commit
changes the behaviour for single files uploaded via the uppy
upload mixin, by merging the file's meta object with this.data
from the parent component.
Also removes "appEventsCache". (and reduces the reported test memory usage by ~33%)
There's no longer any need to remove appEvent listeners in application-instance initializers' `teardown`, as app instances are recreated before each test (in both legacy and ember cli envs)
Intersection observer callback can be called after the component gets destroyed:
```
Assertion Failed: calling set on destroyed object: <@ember/component:ember6019>.docked = false
at assert (ember:37774:17)
at _set2 (ember:17304:46)
at Class.set (ember:29529:29)
at Class._intersectionHandler (discourse/app/components/topic-progress:135:16)
at Backburner._run (ember:56389:25)
at Backburner._join (ember:56365:21)
at Backburner.join (ember:56082:19)
at join (ember:42874:28)
at IntersectionObserver.eval (ember:42978:19)
```
This api allows to add a dropdown at the bottom of a topic, note that this API is mobile only for now.
Also included in the commit:
- various doc fixes
- adding tests for both buttons and dropdowns APIs
- uses thrown instead of @ember/error to ensure execution is halted when incorrect parameters are given
This PR introduces a new `enable_experimental_backup_uploads` site setting (default false and hidden), which when enabled alongside `enable_direct_s3_uploads` will allow for direct S3 multipart uploads of backup .tar.gz files.
To make multipart external uploads work with both the S3BackupStore and the S3Store, I've had to move several methods out of S3Store and into S3Helper, including:
* presigned_url
* create_multipart
* abort_multipart
* complete_multipart
* presign_multipart_part
* list_multipart_parts
Then, S3Store and S3BackupStore either delegate directly to S3Helper or have their own special methods to call S3Helper for these methods. FileStore.temporary_upload_path has also removed its dependence on upload_path, and can now be used interchangeably between the stores. A similar change was made in the frontend as well, moving the multipart related JS code out of ComposerUppyUpload and into a mixin of its own, so it can also be used by UppyUploadMixin.
Some changes to ExternalUploadManager had to be made here as well. The backup direct uploads do not need an Upload record made for them in the database, so they can be moved to their final S3 resting place when completing the multipart upload.
This changeset is not perfect; it introduces some special cases in UploadController to handle backups that was previously in BackupController, because UploadController is where the multipart routes are located. A subsequent pull request will pull these routes into a module or some other sharing pattern, along with hooks, so the backup controller and the upload controller (and any future controllers that may need them) can include these routes in a nicer way.
PERF: Update like count in visible posts without an extra GET per like
Currently when a user is reading a topic and some post in it receive a
like from another user, the Ember app will be notified via MessageBus
and issue a GET to `/posts/{id}` to get the new like count. This worked
fine for us until today, but it can easily create a self-inflicted DDoS
when a topic with a large number of visitors gets a large number of
likes, since we will issue `visitors * likes` GET requests requests.
This patch optimizes this flow, by sending the new like count down in
the MessageBus notification, removing any need for the extra request.
It shouldn't cause any drift on the count because we send down the full
count instead of the difference too.
Possible follow-ups could include handling like removal.
`d-modal-body.js` was setting the text of a `modal-alert` element to `""`, but not removing any classes on that element. Changing this to call `_clearFlash()` ensures that a variety of styling classes are also removed from the element, which prevents empty alert elements being included on any subsequent modals that are displayed.
Several other controllers have also been modified to change the class of the error from `alert-error` to `error. The `alert-` is unnecessary, as it is added by `_flash(msg)` within `d-modal-body.js`.
Since db7be947df the functionality
to not save a bookmark when pressing ESC on the modal has been
broken, because a new initiatedByESC event was introduced there
in d-modal instead of clicking on the modal close button. This
commit fixes the issue by adding an initiatedByESC property
to onClose inside application.js, and checking for that OR
initiatedByCloseButton in the bookmark modal.
This commit, while changing the watched word uploader to use
uppy, also fixes a minor bug with the UppyUploadMixin where
the file input's value was not cleared after reset, which
prevented subsequent file uploads. The composer mixin already
has this fix.
* FIX: ensure required_tag_group_name is null if no value present
If the array was present but empty `required_tag_group_name` would be set to undefined, which would then be removed from the payload of the remote request.
Addming the length check ensures the value is set to null, which is sent as an empty value (which the backend sees, and can remove it and persist the change on the Category object).
This commit changes the emoji uploader to use the UppyUploadMixin,
and makes some minor changes to the emoji uploader (tightening the
copy for drag and drop and adding a percentage for the upload).
Since no other uppy upload mixin components have needed to upload
multiple files so far, this necessitated adding a tracker for the
in progress uploads so we know when to reset the uploader once all
uploads are complete.
At the moment, the emoji uploader cannot be used for direct S3 uploads
because the admin emoji controller creates other records and does other
magic with the emojis. At some point we need to factor this kind of thing
into the ExternalUploadManager.transform! action to complete external
uploads.
We are no longer able to display the image returned by Instagram directly within a Discourse site (either in the composer, or within a cooked post within a topic), so:
- Display an image placeholder in the composer preview
- A cooked post should use an iframe to display the Instagram 'embed' content
When sending a full search request to backend (i.e. when hitting Enter),
the debouncer needs to be cancelled, otherwise it will get invoked and
trigger a second search request to the backend.
When deleting a for_topic bookmark, we were calling
bookmark.attachedTo() for the bookmarks:changed event,
but the bookmark was not always a Bookmark model instance,
so sometimes that call would error. Now we make sure that
the bookmarks in the topic.bookmarks JS array are all
bookmark model instances, and added a test to cover this
deleting for_topic bookmark case as well.
A follow up PR should investigate why `proposal-logical-assignment-operators` is not getting used here (test file?) but this should be enough to get things running.
I was previously relying on `this.isDestroying` returning `true` during `willDestroyElement`. This was an incorrect assumption.
This commit refactors the logic into an explicit `cleanup` function, and also adds some cleanup for empty keys in the `subscribedProxy` array
The flow goes from:
- getting current user object
- creating a POJO using some of the current user keys
- passing this POJO around, which end up being used in message bus
- the processing fn associated ens up doing User.create on this object will both create a User object, but also inject store in it, store is holding a reference to currentUser Object and...
BOOM, we have an object holding a reference to the same object, which JSON.stringify used in prepareBody of pretender doesn't like.
This commit adds uploadHandler support to composer uploads using
uppy. The only things we have that are using this are discourse-brightcove and
discourse-video, which both pop modal windows to handle the file upload and
completely leave out all the composer-type flows. This implementation simply
follows the existing one, where if a single file is uploaded and there
is a matching upload handler we take control away from uppy and hand
it off to the upload handler.
Trying to get this kind of thing working within uppy would require a few
changes because they have no way to restrict uploaders to certain file types
and with the way their uploaders are run it doesn't look like it would be easy
to add this either, so I don't think this is worth the work unless at some
point in the future we plan to have more upload handler integrations.
I also fixed an issue with `cleanUpComposerUploadHandler` which is used
in tests to reset the state of `uploadHandlers` in the composer. This
was doing `uploadHandlers = []` to clear that array, but that creates
a brand new array so anything else referencing the original array will
lose that reference. Better to set `uploadHandlers.length = 0` to
clear it. This was breaking the tests I added to see if upload handlers
were working.
We were previously showing the "n new or updated topics" alert on
category routes like `/c/category-slug/ID/none` on every new/unread
topic update. This PR looks up the category by ID, which should be more
precise.
Previously, `loadLibs` was called inside the `optimize` function of
the media-optimization-worker, which meant that it could be hit
multiple times causing load errors (as seen in b69c2f7311)
This commit moves that call to a specific message handler (the `install` message)
for the service worker, and refactors the service for the media-optimization-worker
to wait for this installation to complete before continuing with processing
image optimizations.
This way, we know for sure based on promises and worker messages
that the worker is installed and has all required libraries
loaded before we continue on with attempting any processing. The
change made in b69c2f7311 is no
longer needed with this commit.
This new app event will fire whenever a bookmark is created,
edited, or deleted for a post or topic, and replaces these old
app events which had inconsistent APIs:
* page:bookmark-post-toggled
* topic:bookmark-toggled
When the event is triggered, the arguments are in this order:
1. bookmark - The bookmark record created or changed. Will be null
if the bookmark was deleted.
2. target - Object with target (post or topic) and targetId (post ID
or topic ID)
When inviting a group to a topic, there may be members of
the group already in the topic as topic allowed users. These
can be safely removed from the topic, because they are implicitly
allowed in the topic based on their group membership.
Also, this prevents issues with group SMTP emails, which rely
on the topic_allowed_users of the topic to send to and cc's
for emails, and if there are members of the group as topic_allowed_users
then that complicates things and causes odd behaviour.
We also ensure that the OP of the topic is not removed from
the topic_allowed_users when a group they belong to is added,
as it will make it harder to add them back later.
* DEV: allow composer option to skip jumping to a post on save
* DEV: refactor js safe access in jump logic
Co-authored-by: Jarek Radosz <jradosz@gmail.com>
This happens because the state of `canLoadMore` is not cleared as the
refresh occurs, which is enough to make the page think a footer should
be displayed.
No tests here because it's tricky to test refreshing and none of our
existing acceptance tests seem to.
This removes all custom controllers and redis/messagebus logic from discourse-presence, and replaces it with core's new PresenceChannel system.
All functionality should be retained. This implementation should scale much better to large numbers of users, reduce the number of HTTP requests made by clients, and reduce the volume of messages on the MessageBus.
For more information on PresenceChannel, see 31db8352
We don't support any browser needing this for very long: https://caniuse.com/?search=selectionStart
I'm keeping some protection so It doesn’t crash but ultimately `element.selectionStart` should be enough.
Im not removing this in the commit, but the `caret_position.js` file seems barely used.
We had code to open the bookmark modal in two places -- the bookmark
list and also from within a topic. This caused the two code paths to
drift, as in the bookmark list we were not passing in the forTopic or
autoDeletePreferences data into the modal, and we were also not refreshing
the bookmark list when the bookmark was deleted from within the modal.
This commit moves the modal opening code into an importable
function from the controllers/bookmark module, and all callers
have to do is pass it an instance of Bookmark and also options
for what to do for the following:
* onAfterSave
* onAfterDelete
* onCloseWithoutSaving
Joffrey JAFFEUX
Jarek Radosz
Martin Brennan
David Taylor
Penar Musaraj
Robin Ward
Natalie Tay
Jordan Vidrine
Alan Guo Xiang Tan
Osama Sayegh
Kerry Liu
Bianca Nenciu
Vinoth Kannan
Kris
jbrw
Andrei Prigorshnev
Dan Ungureanu
Krzysztof Kotlarek
Mark VanLandingham
Loïc Guitaut
Arpit Jalan
Rafael dos Santos Silva
Natalie Tay
Joe
Jean
Roman Rizzi
janzenisaac
Jeff Wong