* Never return broken notifications #409
Since notifications are stored in Redis, we can't expect relational
integrity: deleting a user will *not* delete notifications that
mention it.
But if we return notifications with missing bits (a `follow` without a
`user`, for example), the frontend will get very confused and throw an
exception while trying to render them.
This change makes sure we never expose those broken notifications. For
uniformity, I've applied the same logic to notes and roles mentioned
in notifications, even if nobody reported breakage in those cases.
Tested by creating a few types of notifications with a `notifierId`,
then deleting their user.
(cherry picked from commit 421f8d49e5d7a8dc3a798cc54716c767df8be3cb)
* Update Changelog
* Update CHANGELOG.md
* enhance: 通知がミュートを考慮するようにする
* enhance: 通知が凍結も考慮するようにする
* fix: notifierIdがない通知が消えてしまう問題
* Add tests (通知がミュートを考慮しているかどうか)
* fix: notifierIdがない通知が消えてしまう問題 (grouped)
* Remove unused import
* Fix: typo
* Revert "enhance: 通知が凍結も考慮するようにする"
This reverts commit b1e57e571dfd9a7d8b2430294473c2053cc3ea33.
* Revert API handling
* Remove unused imports
* enhance: Check if notifierId is valid in NotificationEntityService
* 通知作成時にpackしてnullになったらあとの処理をやめる
* Remove duplication of valid notifier check
* add filter notification is not null
* Revert "Remove duplication of valid notifier check"
This reverts commit 239a6952f717add53d52c3e701e7362eb1987645.
* Improve performance
* Fix packGrouped
* Refactor: 判定部分を共通化
* Fix condition
* use isNotNull
* Update CHANGELOG.md
* filterの改善
* Refactor: DONT REPEAT YOURSELF
Note: GroupedNotificationはNotificationの拡張なのでその例外だけ書けば基本的に共通の処理になり複雑な個別の処理は増えにくいと思われる
* Add groupedNotificationTypes
* Update misskey-js typedef
* Refactor: less sql calls
* refactor
* clean up
* filter notes to mark as read
* packed noteがmapなのでそちらを使う
* if (notesToRead.size > 0)
* if (notes.length === 0) return;
* fix
* Revert "if (notes.length === 0) return;"
This reverts commit 22e2324f9633bddba50769ef838bc5ddb4564c88.
* 🎨
* console.error
* err
* remove try-catch
* 不要なジェネリクスを除去
* Revert (既読処理をpack内で行うものを元に戻す)
* Clean
* Update packages/backend/src/core/entities/NotificationEntityService.ts
* Update packages/backend/src/core/entities/NotificationEntityService.ts
* Update packages/backend/src/core/entities/NotificationEntityService.ts
* Update packages/backend/src/core/entities/NotificationEntityService.ts
* Update packages/backend/src/core/NotificationService.ts
* Clean
---------
Co-authored-by: dakkar <dakkar@thenautilus.net>
Co-authored-by: kakkokari-gtyih <daisho7308+f@gmail.com>
Co-authored-by: かっこかり <67428053+kakkokari-gtyih@users.noreply.github.com>
Co-authored-by: tamaina <tamaina@hotmail.co.jp>
Co-authored-by: syuilo <4439005+syuilo@users.noreply.github.com>
* refactor: use IdentifiableError instead of NoteCreateService.ContainsProhibitedWordsError
* fix: notes with prohibited words are reprocessed with delay
* docs(changelog): 禁止キーワードを含むノートがDelayed Queueに追加されて再処理される問題
* lint: fix lint errors
* fix: rethrowするべきなのにrethrowし忘れていたのを修正
* ignore `instance.actor` when checking if there are local users
We've seen this happen a few times:
* there was some AP software at $some_domain
* it gets replaced by Misskey
* before the first user can be created, an AP activity comes in
* Misskey resolves the activity
* to do this, it creates the `instance.actor` to sign its request
* now there *is* a local user, so the `meta` endpoint returns
`requireSetup:false`
* the admin is very confused
This commit factors out the check, and doesn't count the
`instance.actor` as a real user.
* autogen bits
`/users/:user`, `/@:user`, `/notes/:note` return different responses
depending on the request's `Accept:` header. If we don't consistently
return a `Vary: Accept` header, browsers and caching proxies will get
confused, and return AP representations when HTML was requested, or
vice versa.
Co-authored-by: dakkar <dakkar@thenautilus.net>
Co-authored-by: syuilo <Syuilotan@yahoo.co.jp>
