Follow-up to #13452, fixing broken `uri.nil?` test.
Also remove the separate check for `uri` presence, as that would result
in a “Please review 2 errors below” while only one would be listed.
* Fix “Email changed” notification sometimes having wrong e-mail
Fixes#6778
The root of the issue is that `send_devise_notification` was called before
the changes were properly commited to the database, causing the mailer to
pick previous values if running too early.
Devise's documentation provides guidance on how to handle that[1][2], however,
I have found it to not be working, as the following happens, in that order:
- `send_devise_notification` is called for the `email_changed` notification.
In that case, `changed?` is false and `saved_changes?` is true, so
if we use the former, we have the same issue.
- the `after_commit` hook is called
- `send_devise_notification` is called for the `confirmation_instructions`
notification.
In that case, `changed?` is still false, and `saved_changes?` still true,
so if we use the latter, that second notification email is simply not
going to be sent (as we would be queuing the notification *after*
executing the after_commit hook).
This is because it may be called from either an `after_update` or
`after_commit` hook, the difference not being a call to `save` but the
transaction actually being committed to the database. This may arguably
be a bug in Devise, or Devise's notification.
The proposed workaround is inspired by Devise's documentation but checks
whether a transaction is open to make the call whether to immediately
send the notification or defer it to the `after_commit` hook.
[1]: https://www.rubydoc.info/github/plataformatec/devise/Devise%2FModels%2FAuthenticatable:send_devise_notification
[2]: https://github.com/heartcombo/devise/blob/406915cb781e38255a30ad2a0609e33952b9ec50/lib/devise/models/authenticatable.rb#L133-L194
* Fix cases when sending notifications without changing the model
* Defer sending if and only if in transaction including current record
* Fix “Show more” not switching to “Show less” on public pages
Fixes#13169
* Fix initial text of CW button on public pages when CW are unfolded by default
* Fix PostgreSQL load when linking in announcements
Fixes#13245 by caching status lookups
Since statuses are supposed to be known already and we only
need their URLs and a few other things, caching them should
be fine.
Since it's only used by announcements so far, there won't
be much statuses to cache.
* Perform status lookup when saving announcements, not when rendering them
* Change EntityCache#status to fetch URLs instead of looking into the database
* Move announcement link lookup to publishing worker
* Address issues pointed out during review
* Fix 404 and 410 API errors being silently discarded in WebUI
Fixes#13278
* Return more appropriate error when user replies to a deleted toot
* Please CodeClimate
* Fix 404/410 errors on fetching account timelines & identity proofs
* Refactor error handling
* Move error message string to statuses.errors
Mastodon enforces the “sensitive” flag on media attachments whenever a toot
is posted with a Content Warning. However, it does so *after* potentially
converting the Content Warning to toot text (when there is no toot text),
which leads to inconsistent and surprising behavior for API clients.
This commit fixes this inconsistency.
Only look up private toots from database if the request failed because of 401,
403 or 404 errors, as those may indicate a private toot, rather than something
that isn't a toot or cannot be processed.
* Change ActivityPub follower/following collections to not link first page
* Add support for hiding followers and following of remote users
* Switch to using a single `hide_collections` column
* Address code style remarks
Fixes#13136
When a user's canonical acct domain is different from its id's domain
(WEB_DOMAIN ≠ LOCAL_DOMAIN), two webfinger queries are required to find the
canonical domain from the URI. However, we skip webfinger queries when
updating only the key of a remote user, which led to the creation of a
duplicate account, using the URI's domain instead of the canonical acct: one.
`request.format` is not a symbol but a `Mime::Type`, so the condition actually
never matched, and a session was created even for those requests, preventing
caching.
If the “Why do you want to join?” textarea is left empty and the entered params
do not validate, the textarea isn't shown again, unlike other fields.
This commit fixes that by populating an empty `UserInviteRequest` when needed.
* Change meaning of /api/v1/announcements/:id/dismiss to mark an announcement as read
* Change how unread announcements are counted in UI
* Add unread marker to announcements and mark announcements as unread as they are displayed
* Fixups
* Use bundle install in deployment mode for Continuous Integration to catch Gemfile.lock issues
* Fix deprecation warnings
* Fix Gemfile.lock
* More changes to catch Gemfile.lock issues
* Return last_status_at as date, not datetime
* Fix relative timestamp for dates when delay is inferior to 1 day
* Also fix public directory
* Fix error when last_status_at isn't set
* Move announcements above scroll container; add button to temporarily hide them
* Remove interface for dismissing announcements
* Display number of unread announcements
* Count unread announcements accurately
* Fix size of announcement box not fitting the currently displayed announcement
* Fix announcement box background color to match button color
- update http gem to avoid errors
- update blurhash gem to avoid shared object loading error
- update goldfinger gem so the http gem could be updated
- update json gem to avoid warnings
* Fix invalid votes from the API being accepted
Fixes#12556
- Ensure `choice` is an integer instead of silently converting to 0
- Ensure `choice` corresponds to an actual choice of the poll
* Please CodeClimate
* Fix wrong grouping in Twitter valid_url regex
* Add support for xmpp URIs
Fixes#9776
The difficult part is autolinking, because Twitter-text's extractor does
some pretty ad-hoc stuff to find things that “look like” URLs, and XMPP
URIs do not really match the assumptions of that lib, so it doesn't sound
wise to try to shoehorn it into the existing regex.
This is why I used a specific regex (very close, although slightly more
permissive than the RFC), and a specific scan function (a simplified version
of the generalized one from Twitter).
* Remove leading “xmpp:” from auto-linked text
Fixes#12795
It was already possible to create domain blocks for TLDs, but those
weren't enforced, nor editable. This commit changes it so that they
are enforced and editable.
* Hide blocked, muted, and blocked-by users from toot favourite lists
* Hide blocked, muted, and blocked-by users from toot reblog lists
* Hide blocked, muted, and blocked-by users from followers/following (API)
* Fix tests
* Hide blocked, muted, and blocked-by users from followers/following on public pages
Fixes#12690
The `strong_migrations` update from ba2eac8824
introduced a check for `change_column_null` specific to Postgres. This rejects
old migrations.
This commit just wraps old migrations with `safety_assured` to bypass this
check. Alternatives would have been to:
- Disable that check entirely (a possibility added in that same
`strong_migrations` version) for Mastodon, but it makes sense to write new
migrations without such a strong lock.
- Rewrite the old migrations to do it in a way that do not require an exclusive
lock. I thought fixing those old migrations for performance wasn't worth the
pain. Also, if I understand correctly, the next version of
`strong_migrations` is going to include a helper to do that. We could update
those migrations at that point.
Fixes#12607
`will-change: transform` apparently makes manual scrolling impossible on
Firefox/Windows. While this should probably be considered a Firefox bug,
`will-change: transform` seem like a very aggressive performance hint that
may possibly make the browser consume more resources than needed, especially
in multiple-column mode.
This was originally added to improve scrolling performances on mobile, but
I think this isn't necessary anymore, because of the two following reasons:
- `contain: paint` (which is implied by `contain: strict`, which we apply
whenever the browser supports grids) should have similar effects
- in single-column mode, the scrolling container is the root element, which
I believe is optimized in at least Chromium
Keep in mind that I have not been able to make in-depth benchmarks, and
especially not been able to try on mobile, so performances should probably
be investigated further…
* Remove unused “salmon_url” string
* Remove more unused translatable strings
The following strings all used to be on the admin account page
but aren't used anymore: profile_url, outbox_url, followers_url, feed_url
* Fix comments in statuses_cli
The comment for the part of query keeping statuses which have replies was
incorrect, explaining the part of the query kept favourited statuses instead
* Keep statuses favourited by local users in tootctl statuses remove
* Do delete old replies that are not direct interactions with local users
* Skip reblogs of local statuses instead of all reblogs
* Optimize some queries
* Keep old statuses with recent reblogs
* Add option to clear toots from followed accounts too
* Remove “protocol” argument and return value, as only ActivityPub is supported
* Remove FetchRemoteAccountService, only use ActivityPub::FetchRemoteAccountService
* Fix tests
* Revert "Fix ignoring whole status because of one invalid hashtag (#11621)"
This reverts commit dff46b260b.
* Fix statuses being rejected because of invalid hashtag names
* Add spec for invalid hashtag names in statuses
* Add test for featured tags controller
* Fix poll options not being selectable via keyboard
Fixes#12384
* Improve styling of poll option checkboxes/radio buttons
* Use more appropriate ARIA roles for poll options
* Allow switching between single and multiple choice from keyboard
* Coding style
* Avoid using .bind()
* Add follow_request notification type
The notification type already existed in the backend but was never pushed
to the front-end. This also means translation strings were also available
for the backend, from the notification mailer.
Unlike other notification types, these are off by default, to match what
I remember of Gargron's view on the topic: that follow requests should not
clutter notifications and should instead be reviewed at the user's own
leisure in the dedicated column.
Since follow requests have their own column, I've deemed it unnecessary to
add a specific tab for them in the notification quick filter.
* Show follow request link in single-column if there are pending requests, even if account isn't locked
* Push follow requests from notifications to the follow_requests list
* Offer to accept or reject follow request from the notification
* Redesign follow request notification
Before this patch, if remote poll options have leading or trailing spaces,
the information stored locally won't match them, causing federated voting to
fail.
ThibGandGitHub
