We need a nice way to only return some hashtag data
sources based on various site settings. This commit
adds an enabled? method that every hashtag data source
must implement. If this returns false the data source
will not be used at all for hashtag lookups or search.
Simplified query based on SiteSettings to join only relevant user_options rows.
In addition, index was added to 'watched_precedence_over_muted` column in `user_options` table to speed up query
* FEATURE: Inline topic summary. Cached version accessible to everyone.
Anons and non-members of the `custom_summarization_allowed_groups_map` groups can see cached summaries for any accessible topic. After the first 12 hours and if the posts to summarize have changed, allowed users clicking on the button will automatically re-generate it.
* Ensure chat summaries work and prevent model hallucinations when there are no messages.
In previous PR https://github.com/discourse/discourse/pull/22340 bug was introduced. Notifications were blocked when, even if topic was watched directly. New query is taking TopicUser into consideration.
In addition, in user interface, when `watched_precedence_over_muted` is not set, then value from SiteSetting should be displayed.
Recently, site setting watched_precedence_over_muted was introduced - https://github.com/discourse/discourse/pull/22252
In this PR, we are allowing users to override it. The option is only displayed when the user has watched categories and muted tags, or vice versa.
Recently, SQL query returning users who have muted category or tag were introduced, and it is causing performance issues.
It is much more effective to first get IDs of users who have CategoryUser/TagUsers related to specific topic and then in second query get relevant users.
Updates the interface for implementing summarization strategies and adds a cache layer to summarize topics once.
The cache stores the final summary and each chunk used to build it, which will be useful when we have to extend or rebuild it.
New setting which allow admin to define behavior when topic is in watched category and muted topic and vice versa.
If watched_precedence_over_muted setting is true, that topic is still visible in list of topics and notification is created.
If watched_precedence_over_muted setting is false, that topic is not still visible in list of topics and notification is skipped as well.
This PR splits up the preference that controls the count vs dot and destination of sidebar links, which is really hard to understand, into 2 simpler checkboxes:
The new preferences/checkboxes are off by default, but there are database migrations to switch the old preference to the new ones so that existing users don't have to update their preferences to keep their preferred behavior of sidebar links when this changed is rolled out.
Internal topic: t/103529.
This method is a huge footgun in production, since it calls
the Redis KEYS command. From the Redis documentation at
https://redis.io/commands/keys/:
> Warning: consider KEYS as a command that should only be used in
production environments with extreme care. It may ruin performance when
it is executed against large databases. This command is intended for
debugging and special operations, such as changing your keyspace layout.
Don't use KEYS in your regular application code.
Since we were only using `delete_prefixed` in specs (now that we
removed the usage in production in 24ec06ff85)
we can remove this and instead rely on `use_redis_snapshotting` on the
particular tests that need this kind of clearing functionality.
Communities can use sidebar or header dropdown, therefore navigation menu is a better name settings in 2 places:
- Old user sidebar preferences;
- Site setting about default tags and categories.
When we introduced unicode support in the regular expressions used in watched words (9a27803) we didn't realize the cost adding the `u` flag would be.
Turns out, it's pretty bad when you have lots of regular expressions to test. A customer had slightly less than 200 watched words, and it would freeze the browser for about 2s on the first check of those regular expressions (roughly 10ms per regular expression).
This commit introduces a new field (`word`) to the serialized watched words which is then converted to a very fast and cheap regular expression on the client-side. We use that regexp to quicly check whether a matcher is even worth trying so that we don't incure the cost of compiling the expensive unicode regexp.
This commit also busts the `WordWatcher` cache since we added a new field to be serialized.
One nice side effect of using `matchAll` instead of a `while / exec` loop is that the likeliness of having a bad regexp matching infinitely is vastly reduced 🙌
At the moment, PMs to groups with default notification level set to
`watching_first_post` do not generate "emailable" notifications. This happens
because, topic user notification level which is indirectly derived
from the group's default notification level is set to `tracking` if the
group's notification level happens to be `watching_first_post`.
This leads to a `group_message_summary` notification being created
instead of a `private_message` notification which results in no email
alerts being sent when a topic is created.
As this `watching_first_post` --> `tracking` switcheroo appears to be
intentional instead being a bug, this change extends `PostAlerter`'s
`notify_pm_users` method to create a `private_message` notification for
first posts created in a `watching_first_post` group even if the topic
user notification level is set to `tracking`
* DEV: Implement staff logs for user columns edits
* deleted extra space in staff logger detail string, deleted string when no changes are made, added basic test coverage for EditDirectoryColumnsController
* fixed change made to #self.staff_actions un UserHistory
* implemented a method that builds the details, previous_values and new_values in a dynamic way
* removed details of changes
* refactored small merge
This commit introduces a new `within_user_updater_transaction` event that's triggered inside the transaction that saves user updates in `UserUpdater`. Plugins can hook into the transaction using the event to include custom changes in the transaction. Callbacks for this event receive 2 arguments:
1. the user being saved
2. the changed attributes that are passed to `UserUpdater`.
There's also new modifier in this commit called `users_controller_update_user_params` to allow plugins to allowlist custom params in the `UsersController` which eventually end up getting passed as attributes to the `UserUpdater` and the new `within_user_updater_transaction` event where they can be used to perform additional updates using the custom params.
-----
New API is used in https://github.com/discourse/discourse-mailinglist-integration/pull/1.
This commit makes some fundamental changes to how hashtag cooking and
icon generation works in the new experimental hashtag autocomplete mode.
Previously we cooked the appropriate SVG icon with the cooked hashtag,
though this has proved inflexible especially for theming purposes.
Instead, we now cook a data-ID attribute with the hashtag and add a new
span as an icon placeholder. This is replaced on the client side with an
icon (or a square span in the case of categories) on the client side via
the decorateCooked API for posts and chat messages.
This client side logic uses the generated hashtag, category, and channel
CSS classes added in a previous commit.
This is missing changes to the sidebar to use the new generated CSS
classes and also colors and the split square for categories in the
hashtag autocomplete menu -- I will tackle this in a separate PR so it
is clearer.
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.
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.
The value field of ThemeField is only used when viewing a diff in the staff action logs and local theme editing. value is being serialized into the theme index as well, which is not used. It's a huge amount of JSON that we can cut by removing it.
This also breaks up the various theme serializers into separate classes so they autoload properly (or at least restart the server on edit)
This commit fixes the following scenario:
1. The user is searching for hashtags in chat, where the subcategory
type is not highest-ranked in priority order.
2. There can, but doesn't have to be, a higher-ranked matching chat
channel that has the same slug as the subcategory.
3. Since it is not the highest-ranked type, the subcategory, which
normally has a ref of parent:child, has its ref changed to
child::category, which does not work
This was happening because whenever a hashtag type was not highest
ranked, if _any_ other hashtag results conflicted slugs, we would
append the ::type suffix. Now, we only append this suffix if a
higher-ranked type conflicts with the hashtag, and we use the current ref
to build the new typed ref to preserve this parent:child format as well,
it's more accurate.
When user.last_seen was less than push_notification_time_window_mins we
where delaying the notification for the whole
push_notification_time_window_mins PLUS the time the user was away from.
Originally reported in https://meta.discourse.org/t/-/259688
During search indexing we "stuff" the index with additional keywords for
entities that look like domain names.
This allows searches for `cnn` to find URLs for `www.cnn.com`
The search stuffing attempted to keep indexes aligned at the correct positions
by remapping the indexed terms. However under certain edge cases a single
word can stem into 2 different lexemes. If this happened we had an off by
one which caused the entire indexing to fail.
We work around this edge case (and carry incorrect index positions) for cases
like this. It is unlikely to impact search quality at all given index position
makes almost no difference in the search algorithm.
That column is obsolete since we added the `granted_title_badge_id` column in 2019 (56d3e29a69). Having both columns can lead to inconsistencies (mostly due to old data from before 2019).
For example, `BadgeGranter.revoke_ungranted_titles!` doesn't work correctly if `badge_granted_title` is `false` while `granted_title_badge_id` points to the badge that is used as title.
Similar spirit to e195e6f614,
this moves the Bookmarkable registration to DiscoursePluginRegistry
so plugins which are not enabled do not register additional
bookmarkable classes.
We currently apply type: :link watched words to custom user fields. This makes the user card pretty ugly because we don't allow html / links there. Additionally, the admin UI also does not say that we apply this to custom user fields, but only words in posts.
So this PR is to remove the replacement of link-type watch words for custom user fields.
- Reduce duplication of terms in post index from unlimited to 6. This will
result in reduced index size and reduced weighting for posts containing
a huge amount of duplicate terms. (Eg: a post containing "sam sam sam sam
sam sam sam sam", will index as "sam sam sam sam sam sam", only including
the word up to 6 times.) This corrects a flaw where title weighting could
be ignored.
- Prioritize exact matches of words in titles. Our search always performs
a prefix match. However we want to give special weight to exact title matches
meaning that a search for "sum" will find topics such as "the sum of us" vs
"summer in spring".
- Pick up fixes to our search algorithm which are missing from old indexes.
Specifically pick up the fix that indexes URLs properly. (`https://happy.com`
was stemmed to `happi` in keywords and then was not searchable)
see also:
https://meta.discourse.org/t/refinements-to-search-being-tested-on-meta/254158
Indexing will take a while and work in batches, in the background.
* FIX: do not notify admins on suppressed categories
Avoid notifying admins on categories where they are not explicitly members
in cases where SiteSetting.suppress_secured_categories_from_admin is
enabled.
This helps keep notification stream clean and avoids admins mistakenly
being invited to discussions that should be suppressed
The #pluck_first freedom patch, first introduced by @danielwaterworth has served us well, and is used widely throughout both core and plugins. It seems to have been a common enough use case that Rails 6 introduced it's own method #pick with the exact same implementation. This allows us to retire the freedom patch and switch over to the built-in ActiveRecord method.
There is no replacement for #pluck_first!, but a quick search shows we are using this in a very limited capacity, and in some cases incorrectly (by assuming a nil return rather than an exception), which can quite easily be replaced with #pick plus some extra handling.
The algorithm failed to find the correct category by slug when there are multiple sub-sub-categories with the same child-category name and the first child doesn't have the correct grandchild.
So, searching for "child / grandchild" worked in the following case, it found (3):
- (1) parent 1
- (2) child
- (3) grandchild
- (4) parent 2
- (5) child
- (6) grandchild
But it failed to find the grandchild in the following case:
- (1) parent 1
- (2) child
- (4) parent 2
- (5) child
- (6) grandchild
And this also fixes a flaky spec by forcing categories to always order by by `parent_category_id` and `id`.
This makes it possible to partly revert 60990aab55
If a post contains domain with a word that stems to a non prefix single
words will not match it.
For example: in happy.com, `happy` stems to `happi`. Thus searches for happy
will not find URLs with it included.
This bloats the index a tiny bit, but impact is limited.
Will require a full reindex of search to take effect.
When we are done refining search we can consider a full version bump.
Previous regex did not allow for cases where a lexeme contains a : (colon)
This can happen when parsing URLs. New algorithm allows for this.
Test was amended to more clearly call out index problems
* FEATURE: allow restricting duplication in search index
This introduces the site setting `max_duplicate_search_index_terms`.
Using this number we limit the amount of duplication in our search index.
This allows us to more correctly weight title searches, so bloated posts
don't unfairly bump to the top of search results.
This feature is completely disabled by default and behind a site setting
We will experiment with it first. Note entire search index must be rebuilt
for it to take effect.
---------
Co-authored-by: Alan Guo Xiang Tan <gxtan1990@gmail.com>
Using a shared channel with per-message permissions means that every client is updated with the channel's 'last_id', even if there are no messages available to them. Per-user channel names avoid this problem - the last_id will only be incremented when there is a message for the given user.
There was an issue where if hashtag-cooked HTML was sent
to the ExcerptParser without the keep_svg option, we would
end up with empty </use> and </svg> tags on the parts of the
excerpt where the hashtag was, in this case when a post
push notification was sent.
Fixed this, and also added a way to only display a plaintext
version of the hashtag for cases like this via PrettyText#excerpt.
Currently, `Tag#topic_count` is a count of all regular topics regardless of whether the topic is in a read restricted category or not. As a result, any users can technically poll a sensitive tag to determine if a new topic is created in a category which the user has not excess to. We classify this as a minor leak in sensitive information.
The following changes are introduced in this commit:
1. Introduce `Tag#public_topic_count` which only count topics which have been tagged with a given tag in public categories.
2. Rename `Tag#topic_count` to `Tag#staff_topic_count` which counts the same way as `Tag#topic_count`. In other words, it counts all topics tagged with a given tag regardless of the category the topic is in. The rename is also done so that we indicate that this column contains sensitive information.
3. Change all previous spots which relied on `Topic#topic_count` to rely on `Tag.topic_column_count(guardian)` which will return the right "topic count" column to use based on the current scope.
4. Introduce `SiteSetting.include_secure_categories_in_tag_counts` site setting to allow site administrators to always display the tag topics count using `Tag#staff_topic_count` instead.
* Firefox now finally returns PerformanceMeasure from performance.measure
* Some TODOs were really more NOTE or FIXME material or no longer relevant
* retain_hours is not needed in ExternalUploadsManager, it doesn't seem like anywhere in the UI sends this as a param for uploads
* https://github.com/discourse/discourse/pull/18413 was merged so we can remove JS test workaround for settings
Martin Brennan
Krzysztof Kotlarek
Roman Rizzi
Jarek Radosz
Osama Sayegh
Mark VanLandingham
Régis Hanol
Selase Krakani
Juan David Martínez Cubillos
Joffrey JAFFEUX
Natalie Tay
Bianca Nenciu
Sam
Jan Cernik
Rafael dos Santos Silva
Gerhard Schlager
Loïc Guitaut
Ted Johansson
David Taylor
Daniel Waterworth
Alan Guo Xiang Tan
Isaac Janzen
