The old method updated only existing records, without considering that
new tags might have been created or some tags might not exist anymore.
This was usually not a problem because the stats were also updated by
other code paths.
However, the ensure consistency job should be more solid and help when
other code paths fail or after importing data.
Also, update category tag stats too should happen when updating other
category stats as well.
When uploading images via direct to S3 upload, we were
assuming that we could not pre-emptively check the file
size because the client may do preprocessing to reduce
the size, and UploadCreator could also further reduce the
size.
This, however, is not true of gifs, so we would have an
issue where you upload a gif > the max_image_size_kb
setting and had to wait until the upload completed for
this error to show.
Now, instead, when we direct upload gifs to S3, we check
the size straight away and present a file size error to
the user rather than making them wait. This will increase
meme efficiency by approximately 1000%.
### What is this change?
The lounge category was replaced with the general category in https://github.com/discourse/discourse/pull/18097.
However, there are still a few references to the lounge category in code. In particular, `Category#seeded?` is erroring out in production looking for `SiteSetting.lounge_category_id`.
In #21498, we split `BaseStore#download` into a "safe" version which returns nil on errors, and an "unsafe" version which raises an exception, which was the old behaviour of `#download`.
This change updates call sites that used the old `#download`, which raised exceptions, to use the new `#download!` to preserve behaviour (and silence deprecation warnings.)
It also silences the deprecation warning in tests.
What is the problem?
This is a follow up to 4cca7de22d. In the
commit, CSS was used to disable the collapsing of sections in the header
dropdown navigation menu when the `navigation_menu` site setting is set
to `header dropdown`. However, using CSS is not the correct approach as
the underlying code is still marking the section as collapsable which
means that the sections will still be displayed as collapsed with no way
to "uncollapse" if the local store has already marked the section as
collapsed.
What is the fix?
This commit removes the usage of CSS to hide the collapsabe button and
instead correctly marks the section as not collapsable in the code.
What is the problem?
In `SvgSpriteController#search` and `SvgSpriteController#icon_picker_search`, the controller actions
was using the `RailsMultisite::ConnectionManagement.with_hostname` API
but `params[:hostname]` was always `nil` because the routes does not
have a `:hostname` param component and the client does not ever pass the
`:hostname` param when making the request. When `RailsMultisite::ConnectionManagement.with_hostname` is
used with a `nil` argument, it ends up connecting to the default
multisite database. Usually this would be bad because we're allowing a
site in a multisite setup to connect to another site but thankfully no
private data is being leaked here.
What is the fix?
Since `SvgSpriteController#search` and `SvgSpriteController#icon_picker_search` are login required route,
there is no need for us to switch database connections. The fix here is
to simply remove the use of `RailsMultisite::ConnectionManagement.with_hostname`.
Context of the problem
When viewing the topic list for either the personal inbox or the group
PM inbox, we store a cache of the topic list if the user has loaded more
topics in the topic list. This cache is used to improve the experience
for users so that navigating to a topic and then back would not make
them lose their "last read" position in the topic list. Without this
cache, users will have to start from the top of the topic list each time
they navigate back after reading a topic.
What is the problem?
After archiving a PM, the user is redirected to either the personal
inbox or the group PM inbox. The problem is that if a topic list cache
exists, we will render the topic list using the cache. However, this
means that the archived PM will still appear in the list leading to
confusion for our users.
What is the fix?
To fix this, we will simply clear the topic list cache after a user
archives a topic.
It wasn't possible (at least in any reasonable way) to pass params like `tags`. Also removes the export and inlines the function as that was used only to test the function and the test is gone.
This test was passing, but the environment it was testing was incorrect.
The `image-controls` markdown rule allowlists several svgs when previewing.
But since `previewing: true` is only set on the parent `ComposerEditor`
component, the test in `DEditor` wasn't aware of that, so the output was
ignoring the `previewing` option.
This moves the test one level higher, to `ComposerEditor`, and because
now `previewing: true` is correctly used, it updates the test to show
that the svg element is present, but an `onload` attributes is stripped.
We have been struggling lately finding site settings due to 30 setting limit
This was introduced for performance reasons a while back but is no longer as
needed given that ember is faster.
Additionally searching is hard, so allow people to use fuzzy search against
setting name.
What is the problem?
Previously the `sections` getter was initializing duplicate `lib/sidebar/(community-)section` instances every time it was evaluated. This change in identity was causing Ember's `{{#each` helper to totally rerender every section whenever the getter was evaluated.
What is the fix?
This commit refactors things to lean on Ember's components for state/lifecycle management. The `{{#each` loop is done over the source data, which is guaranteed to only change identity when there is a real config change. Individual section components are initialized for each section, and are responsible for constructing and tearing down their own `lib/sidebar/(community-)section` instances.
This commit also updates `lib/sidebar/(community-)section` to support service injection rather than passing service references around.
Co-authored-by: David Taylor <david@taylorhq.com>
We use the `:empty` css selector on `#modal-alert`, so we need to strip any whitespace from the contents to ensure the selector functions correctly. Followup to ad431ab03a
* FIX: Displaying the wrong number of minimum tags in the composer
When the minimum number of tags set for the category is larger than the minimum number of tags
set in the category tag-groups, the composer was displaying the wrong value.
This commit fixes the value displayed in the composer to show the max value between the required
for the category and the tag-groups set for the category.
This bug was reported on Meta in https://meta.discourse.org/t/tags-from-multiple-tag-groups-required-only-suggest-select-at-least-one-tag/263817
* FIX: Limiting tags in categories not working as expected
When a category was restricted to a tag group A, which was set to only allow
one tag from the group per topic, selecting a tag belonging only to A returned
other tags from A that also belonged to other group/s (if any).
Example:
Tag group A: alpha, beta, gamma, epsilon, delta
Tag group B: alpha, beta, gamma
Both tag groups set to only allow one tag from the group per topic.
If Category 1 was set to only allow tags from the tag group A, and the first tag
selected was epsilon, then, because they also belonged to tag group B, the tags
alpha, beta, and gamma were still returned as valid options when they should not be.
This commit ensures that once a tag from a tag group that restricts its tags to
one per topic is selected, no other tag from this group is returned.
This bug was reported on Meta in https://meta.discourse.org/t/limiting-tags-to-categories-not-working-as-expected/263143.
* FIX: Moving topics does not prompt to add required tag for new category
When a topic moved from a category to another, the tag requirements
of the new category were not being checked.
This allowed a topic to be created and moved to a category:
- that limited the tags to a tag group, with the topic containing tags
not allowed.
- that required N tags from a tag group, with the topic not containing
the required tags.
This bug was reported on Meta in https://meta.discourse.org/t/moving-tagged-topics-does-not-prompt-to-add-required-tag-for-new-category/264138.
* FIX: Editing topics with tag groups from parents allows incorrect tagging
When there was a combination between parent tags defined in a tag group
set to allow only one tag from the group per topic, and other tag groups
relying on this restriction to combine the children tag types with the
parent tag, editing a topic could allow the user to insert an invalid
combination of these tags.
Example:
Automakers tag group: landhover, toyota
- group set to limit one tag from the group per topic
Toyota models group: land-cruiser, hilux, corolla
Landhover models group: evoque, defender, discovery
If a topic was initially set up with the tags toyota, land-cruiser it was
possible to edit it by removing the tag toyota and adding the tag landhover
and other landhover model tags like evoque for example.
In this case, the topic would end up with the tags toyota, land-cruiser,
landhover, evoque because Discourse will automatically insert the
missing parent tag toyota when it detects the tag land-cruiser.
This combination of tags would violate the restriction specified in
the Automakers tag group resulting in an invalid combination of tags.
This commit enforces that the "one tag from the group per topic"
restriction is verified before updating the topic tags and also
make sure the verification checks the compatibility of parent tags that
would be automatically inserted.
After the changes, the user will receive an error similar to:
The tags land-cruiser, landhover cannot be used simultaneously.
Please include only one of them.
Moving all control of 'hidden' into Ember will resolve issues we're seeing with Ember fighting against manual DOM manipulation (both vanilla JS and JQuery).
Looking up `controller:modal` from components is not ideal. However, the next step in the refactoring is to create a modal 'service' which will be able to injected into components cleanly.
Having these things configured at the invocation of showModal is a strange API, and means that any changes to the modal require updating the call sites. It makes much more sense for these to be defined as part of the modal's own template. This was already supported for many of the properties. This commit adds support for the `modalClass` and `titleAriaElementId` config to be passed to DModalBody.
For now there is no deprecation message. Support for passing these things to `showModal` will be dropped as part of an upcoming conversion of modals from controllers to components.
Before this commit chat was applying a fixed height on everything under the `/chat` route. It's only really needed on the channel page with the composer at the bottom of the page.
This commits makes the following changes:
- moves height limitation from `#main-outlet-wrapper` to `.chat-channel`
- makes browse channel page and members list page full height and rely on main document scrollbar
- adds height computation for draft header and direct message creator block to ensure the height is correct when creating a draft channel
- makes chat index full height to rely on the browser scrollbar. As a result the <kbd> + </kbd> button used on mobile to create a direct message as been moved out of `<ChannelsList>` into the chat index template
- sidebar height was relying on chat setting a max height, as a result the height computation of sidebar has been changed to work correctly, especially with an opened keyboard on mobile or ipad
Watched words were converted to regular expressions containing \W, which
handled only ASCII characters. Using [^[:word]] instead ensures that
UTF-8 characters are also handled correctly.
What is the problem?
The main problem here is that we were incorrectly registering the same `onStateChange` callback with `TopicTrackingState`
each time a user reads a post. When a user reads a post, the state in `TopicTrackingState` is updated and it triggers all
the `onStateChange` callbacks which have been registered. In the `CommunitySection` class, we register a callback which
would then call the `onTopicTrackingStateChange` method for each link in the class. For the `EverythingSectionLink` class,
this would lookup the state in `TopicTrackingState` to get a new count of unread/new topics and update the `totalUnread` and
`totalNew` properties which are tracked. For some reason that I have yet to figure out, updating the either of the tracked properties
would result in Ember rerendering the entire `{{#each this.sections as |section|}}` in `component/sidebar/user/custom-sections.hbs`
template. Note that `this.sections` refers to a `@cached` getter in the `SidebarUserCustomSections` class. The problem is that
the `sections` getter is initializing a new bunch of sidebar sections related classes without calling the teardown function.
As a result, we end up registering new `onStateChange` callbacks in `TopicTrackingState` in `CommunitySection` without
removing the old ones. Over time, the number of callbacks build up and we end up slowing down the application. While we do
not know the reason why defining a getter for the `sections` is causing the entire block to re-render, I realized that
it is dangerous to use a getter for `sections` here since we have very little control on when the cached is broken.
Instead, I moved the `sections` getter to a tracked property instead where the property is updated via `appEvents`. With
this change, updating the tracked properties in `EverythingSectionLink` is no longer triggering a complete re-render of the
said block above. We also now call `teardown` on the section objects that has been initialised before updating the `sections`
property.
An extensibility point we support server side is setting meta_data
(topic / post custom fields) with the composer payload.
Previous to this change even though we had a lot of setup code we never
actually sent the payload.
This ensures that on create we send meta_data.
- Update welcome topic copy
- Edit the welcome topic automatically when the title or description changes
- Remove “Create your Welcome Topic” banner/CTA
- Add "edit welcome topic" user tip
a373bf2 updated the behavior of replace-emoji so that the input is treated as unsafe-by-default. fancy_title is already escaped, so we need to mark it as html-safe to avoid it being double-escaped.
There is no need to html-safe the result of replace-emoji - it's already done as part of the helper.
### What is the problem?
It is possible to pass an arbitrary value to the limit parameter in `TagsController#search`, and have it flow through `DiscourseTagging.filter_allowed_tags` where it will raise an error deep in the database driver. MiniSql ensures there's no injection happening, but that ultimately results in an invalid query.
### How does this fix it?
This change checks more strictly that the parameter can be cleanly converted to an integer by replacing the loose `#to_i` conversion semantics with the stronger `Kernel#Integer` ones.
**Example:**
```ruby
"1; SELECT 1".to_i
#=> 1
Integer("1; SELECT 1")
#=> ArgumentError
```
As part of the change, I also went ahead to disallow a limit of "0", as that doesn't seem to be a useful option. Previously only negative limits were disallowed.
### Background
When SSRF detection fails, the exception bubbles all the way up, causing a log alert. This isn't actionable, and should instead be ignored. The existing `rescue` does already ignore network errors, but fails to account for SSRF exceptions coming from `FinalDestination`.
### What is this change?
This PR does two things.
---
Firstly, it introduces a common root exception class, `FinalDestination::SSRFError` for SSRF errors. This serves two functions: 1) it makes it easier to rescue both errors at once, which is generally what one wants to do and 2) prevents having to dig deep into the class hierarchy for the constant.
This change is fully backwards compatible thanks to how inheritance and exception handling works.
---
Secondly, it rescues this new exception in `UserAvatar.import_url_for_user`, which is causing sporadic errors to be logged in production. After this SSRF errors are handled the same as network errors.
This fixes a bug in the create invite API where if you passed in an
integer for the group_ids field it would fail to add the user to the
specified group.
The issues fixed:
1. Previously all static pages (e.g. login-required landing page, /tos, /privacy, forgot-password) were wrapped in the faq-read-tracking component
2. All these pages shared one controller with methods that were relevant to one route
3. There were two route-generating functions: `static-route-builder` and `build-static-route` 🤣
4. They were using the deprecated `renderTemplate()` API
5. A slight misuse of Ember API (`controllerFor()`)
6. Small mark-faq-read related bugs
added site toggle functionality through site settings
added tests to implemented feature
Introduced suggested correction
renamed find_new_topic method and deleted click_new_topic_button method
Currently the /new-category url can be accessed by moderators, regardless of whether the Site Setting for moderators_manage_categories_and_groups is true or false.
On top of this, non authorized users can also access this page but shows errors (no 404 loaded).
Since the 404 redirect happens within Ember, we need to allow the site setting value to be accessed within JS.
After this change all non admin users will see a 404 for this route, the exception being moderators if the moderators_manage_categories_and_groups setting has a value of true.
/t/73360
The problem
When selecting text and clicking the "Edit" button that pops up, this opens up the Fast Edit dialog.
The fast edit feature doesn't work well with non standard characters (non-ascii). If the user selects a string of text that contains non-ascii characters, sometimes they won't save. It is non-obvious to the user why this is happening. This issue occurs more frequently when editing content that is written in non-english languages, as fast-edit doesn't work well with non-ascii characters. We currently do a global replace on a couple of the more obvious quotation marks when the fast edit dialog attempts to save, but there are too many edge cases for foreign language content.
The solution
We can fix this issue by using a catch-all approach for non-ascii characters before the user clicks the edit button to bring up the fast edit dialog. Then we can fallback to the full composer to edit their text, which has much better support for non-ascii characters.
What does this regex do?
The regex used matches any character that is not within the ASCII range of 0x00 to 0x7F, which includes all control characters and non-ASCII characters.
This regex pattern can be used to match any character that is not a standard ASCII character, such as accented characters, non-Latin characters, and special symbols.
We were giving topics with repeated words extra weight in search index.
This meant that it was trivial to stuff words into title to dominate in search
given we search for exact title matches first.
The following tweak means that:
`invite invited invites`
and
`invite some stuff`
Both rank the same for title searching.
Titles are short and punchy, duplicating words should not give special
weight.
Requires a full reindex to take effect.
This commit fixes an issue where the Likes Received notification
count in the user digest email was not affected by the
since/last_seen date for the user, which meant that no matter
how long it had been since the user visited the count was
always constant.
Now instead for the Likes Received count, we only count the
unread notifications of that type since the user was last
seen.
This improves keyboard navigation in and out of select-kit components.
The improvements include:
- `Tab` will now dismiss the dropdown once the active element is outside
the select-kit element
- pressing `Escape` will not bubble, this is most noticeable in the
composer, pressing `Esc` there now when a dropdown is expanded will not
dismiss the composer
- `Shift+Tab` will also dismiss the dropdown once focus is outside it
The problem
The fast edit feature doesn't work well with non standard characters (non-ascii). If the user selects a string of text that contains non-ascii characters, then the edit won't save.
The solution
The best solution is to catch those non-ascii characters before the user clicks the edit button to bring up the fast edit dialog. Then we can fallback to the full composer to edit their text, which has much better support for non-ascii characters.
What does this regex do?
The regex used to catch this is [^\x00-\x7F], which matches any character that is not within the ASCII range of 0x00 to 0x7F, which includes all control characters and non-ASCII characters.
This regex pattern can be used to match any character that is not a standard ASCII character, such as accented characters, non-Latin characters, and special symbols.
The current limit (250 characters) is too low, as we have some
translations used for our badge descriptions that result in a
description length of 264 characters.
To be on the safe side, the limit is now set to 500 characters.
Ember's implicit injections feature is removed in Ember 4.x. We want to give ourselves more time to migrate to explicit injections, so this commit re-implements our implicit injections as extensions to the base framework classes.
Incremental migration to newer patterns can be achieved using the `@disableImplicitInjections` class decorator (available from `discourse/lib/implicit-injections').
This resolves and unsilences the `implicit-injections` deprecation.
A category's slug can be encoded when
`SiteSetting.slug_generation_method` has been set to "encoded". As a
result, we have to support non ASCII characters as well.
Change mechanism handling `more` button for sidebar.
Before it was using HTML details tag.
To make tests more reliable, we are switching to use ember runloop.
What is the problem?
The TopicTrackingState is a service on the client side that is used to store
state of topics which is new or has unread posts for a given user. The state
is updated via various means and the one in concern here is whenever we load
a new topic list from the server. When a topic list is loaded from the server,
we sync this new topic list with the states in TopicTrackingState. There is also
a hard limit on the number of states that is stored by TopicTrackingState for
performance reasons and the limit is currently set to 4000. It was noticed that
once this limit has been reached, syncing a topic list with TopicTrackingState can
result in the registered state change callbacks to be called unnecessarily. This
is because during `TopicTrackingState#sync` we call `TopicTrackingState#removeTopic`
if the topic in question is neither new or unread to a user. However, `TopicTrackingState#removeTopic`
would call `TopicTrackingState#_afterStateChange` even if nothing was removed.
What is the fix?
This commit fixes the problem by checking that `TopicTrackingState#_afterStateChange` is only
called in `TopicTrackingState#removeTopic` when a topic is actually removed.
This amends it so our cached counting reliant specs run in synchronize mode
When running async there are situations where data is left over in the table
after a transactional test. This means that repeat runs of the test suite
fail.
- Ensure changing timezones are reflected immediately in the date-time-input (the computed property was missing a dependent key)
- Ensure date-input doesn't lose timezone information (calling `toDate()` causes moment timestamps to lose timezone information)
This was created to resolve issues in the discourse-calendar plugin (https://github.com/discourse/discourse-calendar/pull/399)
In the expand table event handler, we currently rely on `event.target`
to select the table being expanded. Sometimes, the target is the svg icon
wrapped inside the button instead of the button itself. This throws
things off.
This change uses `currentTarget` which refers to the button
element even if the event originated from svg icon.
