There are some hard limits in browser Canvas implementations, that will
throw a runtime exception when crossed. Since those limits are platform
dependent, the best we can do is catch it and back off from trying to
optimize a problematic file.
For example, a 60MB PNG can be processed fine by Chrome but Firefox will
fail trying to extract the ImageData from the CanvasRenderingContext2D
with NS_ERROR_FAILURE.
Also cleans up the media-optimization-utils and add post-resize size logs
The styling between the "Create Invite" and "Share Topic" modals is
shared. The margin that was used to organize inputs in a list is not
needed for the "Share Topic" modal.
We changed (https://github.com/discourse/discourse/pull/13407) behaviour of the topic level bookmark button recently. That PR made the button be opening the edit bookmark modal when there is only one bookmark on the topic instead of just removing that bookmark as it was before.
This PR fixes the next problems that weren't taken into account in the previous PR:
1. Everything should work fine even on very big topics when a bookmarked post is unloaded from the post stream. I've added code that loads the post we need and makes everything work as expected
2. When at least one bookmark on the topic has a reminder, we should always be showing the icon with a clock on the topic level bookmark button
3. We should show correct tooltips for the topic level bookmark button
There is a big difference between regular watched words and regular
expressions and this has been confusing in the past. This notice adds
an explanation.
This commit also reorganizes the code of the test modal.
We want to submit the flag modal on pressing CTRL + ENTER and CMD + ENTER.
Here's how our modals work:
Every modal can be dismissed by pressing ESC. This behaviour can be disabled for a specific modal if we need to.
Every modal can be submitted by pressing ENTER if the cursor wasn't on a text area or a form at the moment of pressing.
Now, the flag modal is actually a one big form and pressing ENTER doesn't submit it. I've added submitting by CTRL+ENTER but at first it was interfering with the basic modal submitting by ENTER. It's a pretty tricky thing to fix because we use the keyup event for submitting by ENTER and we need to use the keydown event for submitting with modifiers (because submitting by CMD+ENTER on Macs doesn't work with keyup).
Eventually, I fixed the problem just by adding a possibility to disable default submitting on ENTER (in the same way as we already have the possibility of disabling dismissing on ESC). Then I disabled default submitting for the flag form and implemented submitting by CTRL+ENTER and CMD+ENTER. This way everything is simple and robust. I did it only for the flag modal but it'll be easy and safe to add the same behaviour to another modal.
This PR improves navigation within lightboxes that contain multiple images for both touch and non-touch devices.
Currently, if a gallery contains multiple large images, and you click on the one currently displayed, two things happen.
1. we zoom in
2. we navigate to the next image
a0bbc346cb/app/assets/javascripts/discourse/app/lib/lightbox.js (L43-L49)
So, you get taken to the next image, and it shows zoomed in, even when the intention was to zoom in on the previous image.
Magnific popup has an option to disable image-click navigation in galleries. This PR toggles that on for non-touch devices.
The result is that if you click on an image in a gallery on a non-touch device, we zoom in on that image instead of navigating to the next one.
This has no impact on arrow/keyboard navigation.
Magnific popup also has an API when images change; we reset the zoom class when that happens. So, when you navigate to the next image, it won't be zoomed in.
For touch devices, clicking on the image will navigate to the next one without zooming in. Users can pinch-zoom if they want to see more details on touch devices.
I used jQuery for this because both Magnific popup and our implementation for this are based on jQuery. No point making a few lines use vanilla for this when the rest doesn't.
Add Members could also invite new users via emails, but that was a less
known fact. Splitting the previous modal into two more accessible
modals should make this feature more discoverable.
Effectively reverts 3ddc33b07c
Makes the failure states testable; see the uncommented test.
I don't think we're re-catching these errors anyway?
_update:_
We did in a single instance in discourse-code-review but it wasn't really intentional and I fixed it in https://github.com/discourse/discourse-code-review/pull/73
* pretender wasn't catching the request because it ran after this test finished
* restore wasn't needed, we do `sinon.restore()` after each test
The error was:
```
↪ Unit | Model | user::resolvedTimezone [✔]
↪ Unit | Utility | url::routeTo with prefixUnhandled request in test environment: /forum/u/chuck.json (PUT)
Error: Unhandled request in test environment: /forum/u/chuck.json (PUT)
at Pretender.server.unhandledRequest (discourse/tests/setup-tests:173:15)
at Pretender.handleRequest (pretender:400:14)
at FakeRequest.send (pretender:169:21)
at Object.send (jquery:10100:10)
at Function.ajax (jquery:9683:15)
at performAjax (discourse/app/lib/ajax:174:19)
at eval (discourse/app/lib/ajax:183:11)
at invokeCallback (ember:63104:17)
at publish (ember:63087:9)
at eval (ember:57463:16)
[✘]
```
A minimal reproduction:
`http://localhost:3001/qunit?seed=3&testId=da76996b&testId=e52a53e7`
The exception page is shown before Ember can actually figure out what the final destination URL we're going to is.
This means that the new page is not present in the history stack, so if we attempt to use the history stack to go back, we will actually navigate back by two steps.
By instead forcing a navigation to the current URL, we achieve the goal of going "back" with no history mucking.
Unfortunately, the actual URL that was attempted is not available. Additionally, this only works for the on-screen back button and not the browser back.
Additionally, several modernizations of the exception page code were made.
This was previously broken by 59ef48c0b9 (#11425, #11424).
Centralize the logic into the exception controller, which avoids the problematic bug and makes it easy to add additional detailed 404 pages in the future.
Sometimes oneboxes contain the same link multiple times and the link
count was shown for each of them. This commit adds link count only to
the most important link, that being either a heading or the header of
the onebox.
If a user posted a URL that appeared inside a Onebox, then the user
got a duplicate link notice. This was fixed by skipping those links in
Ruby.
If a user posted a URL that was Oneboxes and contained other links that
appeared in previous posts, then the user got a duplicate link notice.
This was fixed by skipping those links in JavaScript.
The generated regular expressions did not contain \b which matched
every text that contained the word, even if it was only a substring of
a word.
For example, if "art" was a watched word a post containing word
"artist" matched.
The dismiss new keyboard shortcut (x,r) has been broken since
7a79bd7da3. A fix was done and JS
tests were added in 006d52f32b
and b01e4738ab but the test was not
quite correct and so the bottom dismiss new button was not clicked.
This also fixes an issue with our keyboard shortcut click handling.
If multiple elements matched the selector they were all clicked. Now
we just click the first match.
If you click on a bookmark in the post stream you get an Edit Bookmark modal. This does not happen if you click the topic bookmark button.
We want to open the Edit modal too if there is only one bookmark on a topic (it doesn't matter on the first post or not). The other behaviour if there are > 1 bookmarks in the topic is to prompt the user to confirm delete of all the bookmarks in the topic. This behaviour will stay as-is.
I have done some refactoring in this PR, and still, there is a place for improvement. For example, we don't call post.deleteBookmark() method when deleting several bookmarks. I just don't want to refactor too much in one PR.
The first thing we needed here was an enum rather than a boolean to determine how a directory_column was created. Now we have `automatic`, `user_field` and `plugin` directory columns.
This plugin API is assuming that the plugin has added a migration to a column to the `directory_items` table.
This was created to be initially used by discourse-solved. PR with API usage - https://github.com/discourse/discourse-solved/pull/137/
Note that this commit will also disable daily grouping for datasets with more than 30 data points. This will also smartly do the grouping by month when grouping a full year.
Adds the last updated at and by SMTP/IMAP fields to the UI, we were already storing them in the DB. Also makes sure that `imap_mailbox_name` being changed makes the last_updated_at/by field update for IMAP.
The `bootstrap.json` contains most preloaded information but some routes
provide extra information, such as invites.
This fixes the issue by having the preload request pass on the preloaded
data from the source page, which is then merged with the bootstrap's
preloaded data for the final HTML payload.
Steps to reproduce the bug:
- Create bookmarks for several posts on a topic
- Click the topic level bookmark button, it’ll open the modal that asks to confirm clearing all bookmarks from the topic
- Choose No
- Try to push the topic level bookmark button again - it won’t work
And it's fixed with this commit
This can happen when an avatar-flair component is rendered to an anonymous user on a login_required site (e.g. when they are redeeming an invite). The lack of group information was causing an error to be raised. With this commit, it now simple skips rendering the flair.
* Revert "DEV: skips three tests following cc1e73 (#13386)"
This reverts commit 2be201660a.
* FIX: Do not refresh post stream twice
This also improves the test suite and simulates a long running request
* FIX: Update local copy of raw
Next Week should mean next Monday, Next Month - the first day of the next month, and so on.
Also, we'll be using the name "Next Monday" instead of "Next Week" because it's easier to understand. No one can get confused by next Monday.
* DEV: skips two tests following cc1e73
Following the fix in cc1e73b8e4 we now refresh the whole stream which causes expected states of these tests to not exist anymore.
I'm skipping theses tests while we decide for a better fix.
We previously only showed the link to the Email section
of group settings if both SMTP and IMAP were enabled for
a site, but this is not necessary now, only SMTP can be
enabled by itself so we should show the section if SMTP
is enabled.
Rendering an empty flair element with the css `background-image: url();` causes the browser to attempt an image request against the current document URL. Making duplicate requests for the document URL can cause some unusual race conditions, especially related to cookies. If this user-avatar-flair element was present on the site homepage (e.g. if categories+latest is the homepage), then it can prevent the signup flow from working correctly.
This commit updates the user-avatar-flair component to be a transparent wrapper around the avatar-flair component. If the user has no flair, no avatar-flair element will be rendered. This avoids the `background-image: url();` situation, and fixes the auth flow.
This commit also removes the duplicate avatar flair rendering from the `latest-topic-list-item` component. This wasn't particularly obvious, since the duplicate flairs were being rendered directly on top of each other.
The problem was happening in component integration tests on the rendering stage, sometimes the rendering would never finish.
Using time moments in the future when faking time solves the problem. Unfortunately, I don't know why exactly it helps. It was just a lucky guess after some hours I spent trying to figure out what's going on. But I've done a lot of testings, so looks like it really works. I'll be monitoring builds for some time after merging this anyway.
Unit tests seem to work alright with moments in the past. And we don't fake time in acceptance tests at the moment but I guess they would very likely be flaky with time moments from the past since they also do rendering.
I'm actually thinking of moving all fake time moments to the future (including moments in unit tests) to decrease the chances of flakiness. But I don't want to do everything in one PR, because I can accidentally introduce new flakiness.
A pretty easy way of picking time moments in the future for tests is to use the 2100 year. It has the same calendar as 2021. If a day is Monday in 2021 it's Monday in 2100 too.
Before this fix if your forum was set up with a subfolder and you
clicked on a link to a different subfolder it would not work. For
example:
subfolder: /cool
link is: /about-us
Previously it would try to resolve /about-us as /cool/about-us. With
this fix it redirects to /about-us correctly.
Editing a post that was just posted caused it to be reloaded and made a
request to the server. This had an additional side effect where the
model instances used by post stream and composer would be different and
changes did not propagate correctly.
Previously due to "rowheader" role we would read out topic titles twice.
This adjusts it so we apply the heading role only to the topic link.
In turn this makes navigation through topic lists more accurate (h) only
lands you on topic links. It also reduces the amount of duplicate reading
NVDA does.
Before:
Topic title link new topic link support link b481 link 19h link 2 button...
After:
Topic title link
This reduces noise, up and down once you land on a topic link can give you
more context.
* UX: Improvements to reorder categories UX
Before, moving a category from, for example, position 25 to position 0 would result in switching the positions of the two categories at those positions.
Category A at position 0 would move to position 25, and Category B at position 25 would move to position 0.
Instead of switching positions, the reorder categories function should retain the order of categories except for the one being moved.
So, Category B at position 25 would still move to position 0, but Category A is merely bumped down to position 1.
This improves the UX because if a user *really* wants to switch the two categories, it results in one extra step. However in the other (what I think is normal) case, it saves the 24 other switches the user has to make to get Category A back to position 1 (you can imagine the user having to click the up arrow button repeatedly to return Category A to the top of the page). Now, imagine trying to do this with a site with 100s of categories. Yikes!
The UX improvement described above is what this commit accomplishes by redesigning the `move()` method of the reorder-categories controller. It adds some overhead to adjust the positions of all categories in between the origin and target positions, but in testing this is not noticible to the user. It's better for the computer to do extra work than the user.
* UX: Allow decimal input in reorder-categories for more precise positioning.
A common UX pattern when reordering a list of items is to allow a user to specify a target position as a decimal between two valid integer positions. The user is indicating they want the target list item to move in between the list items at the positions on either side of the target position.
For example, say there are three categories Category A at position 0, Category B at position 1, and Category C at position 3.
To move Category C in between Categories A and B, a user can now simply update Category C's position to 0.5.
The `ember_jquery` bundle contains production builds of Ember and jQuery
which doesn't work with tests. This commits introduces a new
`theme_qunit_vendor` bundle which is copy of the `vendor` bundle but
doesn't contain `ember_jquery`.
This commit is a partial revert of
409c8585e4
Previously, the `transformed.blah` shortcut could only be used in top-level hbs statements like {{transformed.blah}}. When attempting to use it in a sub-expression like `{{concat "hello" transformed.world}}`, it would raise a "transformed is not defined" error.
This commit updates the shortcut logic to make `transformed.blah` and `attrs.blah` work consistently in all hbs expressions.
Co-authored-by: Jordan Vidrine <jordan@jordanvidrine.com>
We don't want to show the draft checkmark in the composer when drafts are saved, as it’s a little bit distracting to see it keeps appearing and disappearing. Only in the case of error does it need to show anything, we will be showing a "drafts offline" warning as we did it before.
An important detail is that the warning was appearing and disappearing all the time too. Now, the warning won’t be flashing while a user is typing, it’ll be disappearing only when the draft was eventually saved.
I made a change in https://github.com/discourse/discourse/pull/13083/files to suppress re-throwing the error from popupAjaxError if isTesting() but that causes issues in other places instead. If I remove it I get this error in the group email test I added, so I am removing that test here too.
* DEV: replace swipe events to use translate rather than left/right
translate is better for animations. also use native css animations for opening
and closing.
* a11y: respect prefers reduced motion on mobile timeline
* DEV: reduce jquery usage
* DEV: add tests for menu swipe events
test is run in 50% zoom/transform which means offsets and x of touch events need to be halved
Refactor test window to use a transform rather than non-standard zoom property
Co-authored-by: Penar Musaraj <pmusaraj@gmail.com>
A followup to e3b0abc and a replacement PR for #13298.
Fixes long topic titles wrapping to a separate line in the dropdown search results.
Also replaces divs that were incorrectly nested inside spans.
* FIX: Quoting Oneboxed content should exclude formatting
When a post is quoted that includes Oneboxed content, we should not include the formatting generated by the Onebox. Rather, we should attempt to collapse the link referenced by the Onebox to a single line text link.
* DEV: fix tests
Tag-chooser component expects an array of blocked tags, but was passed
a string instead. That made tag-chooser to not allow any tags that were
a substring of the current one.
In Ember CLI addons get put into the vendor bundle, as opposed to their
own bundle like we're doing in the Rails app. We never use pretty-text
without our vendor bundle so this should have no difference on
performance.
We need to keep the pretty-text bundle for server side cooking.
Rather than returning the size of the currently rendered image in the composer window (which is dependent on browser settings such as window size and zoom level), return the actual dimensions of the image file itself.
(Also see commit abac614492 which was an earlier attempt to fix this by excluding Oneboxed images entirely. That was reverted as the CSS selector didn’t work on all browsers.)
If we don't escape periods, they are interpreted as wildcards and it
becomes impossible to visit profiles of other users whose usernames
match. E.g., if your username was `a.c` and attempted to visit `abc`'s
profile, you would be incorrectly redirected to your own profile.
* FIX: Ignore `allowlistgeneric` Onebox image sizes
The size of an image contained within the preview pane of a Composer window may vary depending on the configuration of the browser displaying the Composer (e.g., dimension of browser window, zoom level, etc.).
Presently, the dimensions of the images from the browser creating the post containing the Onebox will be used to render the Onebox to anyone who views the post. It is safer to let the backend figure out the dimensions of the images. Therefore, exclude `.onebox.allowlistedgeneric` images from the list of `image_sizes` sent to the backend.
* DEV: Replace jQuery selector with pure JS
* DEV: remove more jQuery
After editing a post, it is refreshed by two ways. One of them is
triggered by the client side which will route the client to the edited
post and force a reload this way. The other way is via Message Bus.
This commit ignores both of the ways and tries to update the post
immediately and then refresh the post stream.
It was not clear that replace watched words can be used to replace text
with URLs. This introduces a new watched word type that makes it easier
to understand.
I merged this PR in yesterday, finally thinking this was done https://github.com/discourse/discourse/pull/12958 but then a wild performance regression occurred. These are the problem methods:
1aa20bd681/app/serializers/topic_tracking_state_serializer.rb (L13-L21)
Turns out date comparison is super expensive on the backend _as well as_ the frontend.
The fix was to just move the `treat_as_new_topic_start_date` into the SQL query rather than using the slower `UserOption#treat_as_new_topic_start_date` method in ruby. After this change, 1% of the total time is spent with the `created_in_new_period` comparison instead of ~20%.
----
History:
Original PR which had to be reverted **https://github.com/discourse/discourse/pull/12555**. See the description there for what this PR is achieving, plus below.
The issue with the original PR is addressed in 92ef54f402
If you went to the `x unread` link for a tag Chrome would freeze up and possibly crash, or eventually unfreeze after nearly 10 mins. Other routes for unread/new were similarly slow. From profiling the issue was the `sync` function of `topic-tracking-state.js`, which calls down to `isNew` which in turn calls `moment`, a change I had made in the PR above. The time it takes locally with ~1400 topics in the tracking state is 2.3 seconds.
To solve this issue, I have moved these calculations for "created in new period" and "unread not too old" into the tracking state serializer.
When I was looking at the profiler I also noticed this issue which was just compounding the problem. Every time we modify topic tracking state we recalculate the sidebar tracking/everything/tag counts. However this calls `forEachTracked` and `countTags` which can be quite expensive as they go through the whole tracking state (and were also calling the removed moment functions).
I added some logs and this was being called 30 times when navigating to a new /unread route because `sync` is being called from `build-topic-route` (one for each topic loaded due to pagination). So I just added a debounce here and it makes things even faster.
Finally, I changed topic tracking state to use a Map so our counts of the state keys is faster (Maps have .size whereas objects you have to do Object.keys(obj) which is O(n).)
<!-- NOTE: All pull requests should have tests (rspec in Ruby, qunit in JavaScript). If your code does not include test coverage, please include an explanation of why it was omitted. -->
Refactors `TrustLevel` and moves translations from server to client
Additional changes:
* "staff" and "admin" wasn't translatable in site settings
* it replaces a concatenated string with a translation
* uses translation for trust levels in users_by_trust_level report
* adds a DB migration to rename keys of translation overrides affected by this commit
In Ember CLI, the vendor bundler includes Ember/jQuery, so this brings
our app closer to that configuration.
We have a couple pages (Reset Password / Confirm New Email) where we need
`ember_jquery` without vendor so the file still exists for those cases.
Previous to this change we would switch off MessageBus updating after 20
minutes.
This ensures that when the user becomes present again we turn on long polling.
Without long polling updates can be delayed for minutes.
The widget should accept the disabled option.
In that case, CSS class "disabled".
In addition, after click dropdown will not be shown.
Also, the option to disable a specific value in a dropdown is included
…and just prioritize the current one, instead of hiding other categories.
Context: when you open the composer by clicking "New Topic" button when in a category, or by clicking "New Topic" in the share-popup, the category selector shows only the current category and its children (and "Uncategorized"). You can still find other categories, but you have to search by name.
This PR changes that, so you now can see all the categories in the dropdown, and those that are relevant (again: current, children and uncategorized) are displayed before all other categories.
tldr: don't make choosing other categories harder - make choosing relevant ones easier.
1. It defaults to `cache: true` already
2. Setting it to `false` for non-GET request doesn't do anything
3. We were correcting `cache: false` GET requests to use `cache: true`
…so setting it to anything at all, for any type of request doesn't make sense (anymore)
When the client received a new notification, it prioritized only PM
notifications instead of maintaining the priority order. Later, the
check for missing notification deleted all notifications that were
in the wrong order because it could not match the IDs.
The correct order puts high_priority AND unread notifications first.
Low priority or read notifications (including high priority, but read
notifications) come after.
Original PR which had to be reverted **https://github.com/discourse/discourse/pull/12555**. See the description there for what this PR is achieving, plus below.
The issue with the original PR is addressed in 92ef54f402
If you went to the `x unread` link for a tag Chrome would freeze up and possibly crash, or eventually unfreeze after nearly 10 mins. Other routes for unread/new were similarly slow. From profiling the issue was the `sync` function of `topic-tracking-state.js`, which calls down to `isNew` which in turn calls `moment`, a change I had made in the PR above. The time it takes locally with ~1400 topics in the tracking state is 2.3 seconds.
To solve this issue, I have moved these calculations for "created in new period" and "unread not too old" into the tracking state serializer.
When I was looking at the profiler I also noticed this issue which was just compounding the problem. Every time we modify topic tracking state we recalculate the sidebar tracking/everything/tag counts. However this calls `forEachTracked` and `countTags` which can be quite expensive as they go through the whole tracking state (and were also calling the removed moment functions).
I added some logs and this was being called 30 times when navigating to a new /unread route because `sync` is being called from `build-topic-route` (one for each topic loaded due to pagination). So I just added a debounce here and it makes things even faster.
Finally, I changed topic tracking state to use a Map so our counts of the state keys is faster (Maps have .size whereas objects you have to do Object.keys(obj) which is O(n).)
Jarek Radosz
Rafael dos Santos Silva
Mark VanLandingham
Dan Ungureanu
Arpit Jalan
Andrei Prigorshnev
Martin Brennan
Alan Guo Xiang Tan
Roman Rizzi
Bianca Nenciu
awesomerobot
Penar Musaraj
Robin Ward
Jordan Vidrine
Joe
Joffrey JAFFEUX
Gerhard Schlager
Kane York
David Taylor
Vinoth Kannan
Sam
Jeff Wong
Grayden
Osama Sayegh
jbrw
Régis Hanol
Meghna
Krzysztof Kotlarek
Faizaan Gagan