* Add eslint-plugin-promise to detect uncaught rejections
* Move alert generation for errors to actions/alert
* Add missing rejection handling for Promises
* Use catch() instead of onReject on then()
Then it will catches rejection from onFulfilled. This detection can be
disabled by `allowThen` option, though.
`navigator.share()` rejects Promise if user cancelled sharing, and it may
print it as an error on JavaScript console.
This patch ignores it and prints other errors on the console.
`Response.prototype.clone()` must be called before the response used.
This fixes an error from ServiceWorker and failing to load image when the
image is not cached.
This patch adds null item (i.e. gap) handling on below functions to avoid TypeError.
* `filterNotifications` called on user mute/block
* `deleteByStatus` called on status deletion
* Fix stack overflow on importFetchedAccounts
When the account has moved property, it should process destination
account instead of source account itself.
* Set account id instead of account object for moved property
This restores "foo has moved to" indication on account view, and
fixes `reblog` index on `accounts` object store.
Migration is wrapped by transaction, so manual `commit_db_transaction` without transaction restarting causes "there is no transaction in progress" warnings. We should use `disable_ddl_transaction!` instead, if we can omit transaction completely.
We added an index for `[account_id, reblog_of_id]`, but we already have a similar index for `reblog_of_id`. Those index will be bigger according to statuses count. For example, `reblog_of_id` index uses 800MB for 10GB statuses table.
So this patch swaps indexed columns like `[reblog_of_id, account_id]`, then it will covers both usage with single index.
Since those index creation may take a while, I've also disabled previous index creation.
* Show confirmation dialog on leaving WebUI while composing
Currently, Back button and Back hotkey can cause leaving from WebUI, as well as browser's back button. Users may hit those buttons accidentally, and their composing text will be lost.
So this prevents it by showing confirmation dialog from `onbeforeunload` event.
* Fix message and comments
We have changed how we store reblogs in the redis for bigint IDs. This process is done by 1) scan all entries in users feed, and 2) re-store reblogs by 3 write commands.
However, this operation is really slow for large instances. e.g. 1hrs on friends.nico (w/ 50k users). So I have tried below tweaks.
* It checked non-reblogs by `entry[0] == entry[1]`, but this condition won't work because `entry[0]` is String while `entry[1]` is Float. Changing `entry[0].to_i == entry[1]` seems work.
-> about 4-20x faster (feed with less reblogs will be faster)
* Write operations can be batched by pipeline
-> about 6x faster
* Wrap operation by Lua script and execute by EVALSHA command. This really reduces packets between Ruby and Redis.
-> about 3x faster
I've taken Lua script way, though doing other optimizations may be enough.
Specifically, this fixes status length calculation to be same as JS side.
BTW, since this pattern used in not only preview card fetching, we
should extract it (with twitter-regex?) and write tests I think.
We've changed un-reblogging behavior when we implement Snowflake, to insert un-reblogged status at the position reblogging status existed.
However, our API expects home timeline is ordered by status ids, and max_id/since_id filters by zset score. Due to this, un-reblogged status appears as a last item of result set, and timeline expansion may skips many statuses.
So this reverts that change...reblogged status inserted at corresponding position to its id.
The main change of this PR is removing `order by visibility` hack.
This was introduced to force using of `index_statuses_on_account_id` instead of PK index, but it seems no longer needed probably due to `index_statuses_on_account_id_id`. Removing this avoids reading all rows, so really improves first fetching of the user who has lot of statuses.
I have also changed JOIN to IN + subquery, which slightly faster in most cases.
* If OEmbed response doesn't have a required property `type`, ignore it.
e.g. `NoMethodError: undefined method 'type' for ...`
* If we failed to detect encoding, fallback to default behavior of Nokogiri.
e.g. `KeyError: key not found: :encoding`
When ancestors get loaded, we scroll to the target status (i.e. skip
ancestors). However, ancestors may get loaded before the status itself,
then it causes TypeError because `this.node` is undefined yet.
Since we don't show anything until the status gets loaded, we don't need
to scroll to the target status in this time. If we get the status itslef
later, it causes `componentDidUpdate` and scrolling correctly.
* Improve error handling on LinkCrawlWorker
* Ignore TimeoutError and InvalidURIError too
* Record errors to debug log
* Enable dead job queue on LinkCrawlWorker
Since most of acceptable errors were already ignored, only our side issue should go to dead job queue.
* Ignore all http gem errors
PostgreSQL10 has been released, but upgrading from older versions needs dump/restore. If you pull new version without those handling, db service will fail to launch.
To prevent accidentally upgrading, and as a recommended version, this patch specifies PostgreSQL and Redis version.
We added horizontal layout to preview card for wide image. However, max height of the thumbnail is still limited to 120px and it makes nearly square images to too small for that layout.
This PR increases max height as well as max width.
Currently we're using a list of MIME types for `accept` attribute on `input[type="file"]` for filter options of file picker, and actual file extensions will be infered by browsers. However, infered extensions may not include our expected items. For example, "image/jpeg" seems to be infered to
only ".jfif" extension in Firefox.
To ensure common file extensions are in the list, this PR adds file extensions in addition to MIME types. Also having items in both format is encouraged by HTML5 spec.
https://www.w3.org/TR/html5/forms.html#file-upload-state-(type=file)
We had returned `nil` for that case, but this raises an error instead, as a wrong usage of the method.
This method is currently only used in ActivitySerializer.
`Status#reply?` may returns true even if the thread is missing.
e.g. the replied status was deleted or couldn't be fetched.
Then it raises NoMethodError on various AP json serialization.
This issue won't happen on Atom serialization because it checks thread
existence using `StreamEntry#threaded?` instead.
* Fix behavior while the button is invisible
e.g. pointer cursor, couldn't open contextmenu
* Avoid rendering the button to remove blank space if no more items are available or no items are rendered
This PR adds section for protocol specific information, then always show
both of OStatus and ActivityPub. Specifically, this will help admins to
check PuSH subscription status and unsubscribe manually, even `protocol`
has been changed.
This also includes below changes:
* Add `overflow: hidden` to prevent float leaking
* Add missing fields for ActivityPub
Resolved:
* Lot of redundant renders while mouse moving
* Scroll jumping when timeline loaded
* Scroll position isn't kept when statuses below the scrollTop was deleted then new status arrived
Unresolved:
* Scroll position isn't kept when statuses over the scrollTop was deleted then new status arrived
-> It needs to know which statuses are over the scrollTop
* New status indicator should be active when new statuses arrived while mouse moved recently
-> It needs a) update indicator in ScrollableList, or b) set scrollTop status while mouse moving
To reflect status posting immediately, we've inserted the status into timelines directly. However, status insertion changes "latest status", and it means next timeline refresh only fetches statuses since the inserted status. This behavior is very bad for disconnected timeline and mobile views.
After this patch, it refreshes timeline for disconnected timelines, instead of direct insertion.
In before, the method uses stream_entry id as status id, so replied status was wrongly selected.
This PR uses StatusFinder which was introduced with `Api::Web::EmbedsController`.
* Raise an error for remote url in StatusFinder
Previous implementation had allowed remote url with status id which also exists on local.
Then that bug leads /api/web/embed to return wrong embed url.
* Fix oembed_controller_spec
Currently, private / direct posts via OStatus from AP compatible instance will be dropped due to failing to fetch AP version.
So this fallbacks to OStatus handling:
* when failed to fetch ActivityPub version
* when status is neither :public nor :unlisted
SerializarbleResource#as_json serializes to Symbol keyed Hash, but current
implementation of LinkedDataSignature expects String keyed Hash.
So it generates broken payload.
* Process Create / Announce activity in FetchRemoteStatusService
* Use activity URL in ActivityPub for reblogs
* Redirect to the original status on StatusesController#show
* Ignore empty response in ActivityPub::FetchRemoteStatusService
This fixes `NoMethodError: undefined method `[]' for nil:NilClass` error.
* Check json.nil? in JsonLdHelper#supported_context?
* Fallback to OStatus in FetchAtomService
* Skip activity+json link if that activity is Person without inbox
* If unsupported activity was detected and all other URLs failed, retry with ActivityPub-less Accept header
* Allow mention to OStatus account in ActivityPub
* Don't update profile with inbox-less Person object
* Add protocol field
* Switch protocol specific information according to active protocol
* Hide PuSH subscription related buttons if ActivityPub is active
`height: 100%` in `align-self: stretch` flexboxes doesn't work on Safari < 11.
https://bugs.webkit.org/show_bug.cgi?id=137730
This workaround uses flexbox instead of `height: 100%` to stretch height.
"defaults post privacy to followers-only" only means...
* default value of `visibility` param on post API
* default value for web UI privacy setting (i.e. it will be overridden if they once updated)
...so, many users won't see an effect of it.
Render function for BundleContainer must not be methods.
React doesn't know dependency of the method, so they won't rerender on property updates.
In this case, when you close modal and open another modal immediately,
old modal will be open instead of new one.
Since es5-ext used `new Function("...")`, it caused CSP violation unless "unsafe-eval" included. So this patch updates it to the version which fixes it.
Note that this package is used in polyfills, so loaded only if needed. I've encountered this issue on iOS9.
cf. medikoo/es5-ext@d3864493
This fixes a warning on status unmounting (e.g. deletion).
This also resets IntersectionObserverWrapper on disconnect to avoid `unobserve()` calls
which has bug in Edge.
Setting ASSET_HOST to `http://0.0.0.0:8080` makes urls in manifest.json to
be invalid, e.g. `http://0.0.0.0:8080/packs/application.js`.
Anyway, we don't need set this on build:development because assets would
be delivered from same origin in development (and w/o dev-server).
Webpack seems to fail to import `react-intl/locale-data/*.js` if those
files has been proceed by babel, and this also breaks applying our translation.
Note that this won't be a problem on English locale, because react-intl
includes it as default and works fine without manually added locale-data.
Also this issue seems to only occurs on production build, but I'm not sure
about reason.
(This patch has been merged as bugfix and reverted, but still valuable as
improvement)
Previously, we've attached IntersectionObserver twice for boosted statuses:
wrapper Status and wrapped Status. but wrapped Status don't need to manage
intersection and visibility by itself, because it's a part of wrapper Status.
unaristandEugen Rochko
