Compact Language Detector v3 (CLD3) is the successor of CLD2, which was
used in the previous implementation. CLD3 includes improvements since CLD2,
and supports newer compilers. On the other hand, it has additional
requirements and cld3-ruby, the FFI of CLD3 for Ruby, is still new and may
be still inmature.
Though CLD3 is named after CLD2, it is implemented with a neural network
model, different from the old implementation, which is based on a Naïve
Bayesian classifier.
CLD3 supports newer compilers, such as GCC 6. CLD2 is not compatible with
GCC 6 because it assigns negative values to varibales typed unsigned.
(see internal/cld_generated_cjk_uni_prop_80.cc) The support for GCC 6 and
newer compilers are essential today, when some server operating system
such as Ubuntu Server 16.10 has GCC 6 by default.
On the one hand, CLD3 requires C++11 support. Environments with old
compilers such as Ubuntu Server 14.04 needs to update the system or install
a newer compiler.
CLD3 needs protocol buffers as a new dependency. However,it is not
considered problematic because major server operating systems, CentOS and
Ubuntu Server provide them.
The FFI cld3-ruby was written by me (Akihiko Odaki) for use in Mastodon.
It is still new and may be inmature, but confirmed to pass existing tests.
* Dont use raise_error by itself (avoids warning)
* Add coverage for AccountFilter
* Improve coverage and refactor for Subscription#lease_seconds
* Improve coverage and refactor for NotificationMailer
* Simplify assignment of min/max threshold on subscription
* Fix#2473 - Use sidekiq scheduler to refresh PuSH subscriptions instead of cron
Fix an issue where / in domain would raise exception in TagManager#normalize_domain
PuSH subscriptions refresh done in a round-robin way to avoid hammering a single
server's hub in sequence. Correct handling of failures/retries through Sidekiq (see
also #2613). Optimize Account#with_followers scope. Also, since subscriptions
are now delegated to Sidekiq jobs, an uncaught exception will not stop the entire
refreshing operation halfway through
Fix#2702 - Correct user agent header on outgoing http requests
* Add test for SubscribeService
* Extract #expiring_accounts into method
* Make mastodon:push:refresh no-op
* Queues are now defined in sidekiq.yml
* Queues are now in sidekiq.yml
* Likely fix#2458, fix#2031 - handle out-of-order deletes for statuses
If a delete arrives before the original status, cache that information
for 6h, and if the original status arrives in that window, ignore it
* Add test case
Fix#2196 - Respond with 201 when Salmon accepted, 400 when unverified
Fix#2629 - Correctly handle confirm_domain? for local accounts
Unify rules for extracting author acct from XML, prefer <email>, fall back
to <name> + <uri> (see also #2017, #2172)
* Replace browserify with webpack
* Add react-intl-translations-manager
* Do not minify in development, add offline-plugin for ServiceWorker background cache updates
* Adjust tests and dependencies
* Fix production deployments
* Fix tests
* More optimizations
* Improve travis cache for npm stuff
* Re-run travis
* Add back support for custom.scss as before
* Remove offline-plugin and babili
* Fix issue with Immutable.List().unshift(...values) not working as expected
* Make travis load schema instead of running all migrations in sequence
* Fix missing React import in WarningContainer. Optimize rendering performance by using ImmutablePureComponent instead of
React.PureComponent. ImmutablePureComponent uses Immutable.is() to compare props. Replace dynamic callback bindings in
<UI />
* Add react definitions to places that use JSX
* Add Procfile.dev for running rails, webpack and streaming API at the same time
* Add coverage for create with empty acct value
* Add coverage for create with webfinger failure
* Add coverage for create with webfinger providing bad values
* Add coverage for create when webfinger is good
* Add coverage for session[:remote_follow] having data
* Simplify how remote follow pulls acct from session
* Remote follow behaves more like model
* Move the discovery portions of remote follow out of controller
* Check for suspended accounts
* Naive approached to timeline filtering
* Convert allowed_languages into a db column
* Allow users to choose languages to see statuses in
* Style list items as two columns
* Add a hint to explain language filtering preference
* Add coverage for admin/confirmations controller
* Coverage for statuses controller show action
* Add coverage for admin/domain_blocks controller
* Add coverage for settings/profiles#update
This adds a test for the `Feed#get` method. While the data
transformations in `Feed#get` may seem redundant, they are important to
maintain the order from Redis. The tests I wrote will fail if someone
tries to refactor away this "redundancy" (as I tried to do in the first
iteration of this change).
In #2110, a new attachment type "unknown" was introduced for
attachments that were rejected due to a domain being blocked using
reject_media. However, the "type" field was never set to "unknown"
because a default value of "0" (image) is set for that column,
causing the `type.blank?` expression to always equal false.
This version uses type_changed? instead, causing the type to be set
to "unknown" unless a type has been explicitly set. This introduces
a small change in behaviour causing the type to be set to unknown
before paperclip calls `before_post_process`. Presumably this
behaviour is more appropriate than the current one because the
attachment type has not been determined by that point.
Included are new tests for `ProcessFeedService` and
`UpdateRemoteProfileService` which now check that remote media is
downloaded for non-blocked domains and is rejected for others.
* Add render_views in more places
* Delegate methods from account to user with allow nil true, so that admin accounts show view renders when missing a user
* Use actual account instances in authorize follow controller spec
* Add rough outline of coverage needed for public timeline
* Specs for visibility, replies, boosts
* Specs for silenced account
* Specs for local_only option
* Specs for blocks and mutes
* Add tentative spec around including other silenced account statuses
* Add with_public_visibility scope
* Add simple coverage for tag_timeline
* Tag timeline includes replies
* Replace tag.statuses with a tagged_with scope in tag timeline method
* Use with_public_visibility in tag timeline
* Extract common scope between public and tag timelines to method
* Extract local domain check to local_only scope
* Extract local_only check to starting scope method
* Move list of excluded from timeline account ids to account model
* Simplify excluded accounts list on account model
* Only join accounts when needed
* Rename method for account specific filtering
* Extract method for account exclusions
* Fix bug where silenced accounts were not including statuses from other silenced accounts
* DRY up filter application from account or no account
* timeline_scope can be private
* Add spec showing that account can find its excluded accounts ids
* Add spec which fails if local_only does not have a left outer join
* rubocop
* Instead of parsing shared status contents verbatim, make roundtrip
to purported original URL. Confirm that the "original" URL is from the
same domain as the author it claims to be from.
* Fix obvious typo, add comment
* Use URI look-up first
* Add test, update Goldfinger dependency to make less useless HTTP requests per Webfinger lookup
Matt Jankowski