In continuous reading modes the pages are always rendered regardless of if they are visible or not. This is the case to prevent layout shifts when prepending pages while scrolling up.
However, this means that a pages "load priority" and its "should load" flag changes everytime the current page changes, which causes unnecessary rerenders.
The "load priority" and "should load" flag do not matter once a page has been loaded and therefore should stay consistent once loaded.
This caused the reader to "jump to random pages" because it caused the image to get unrendered and rendered again, leading to layout shifts.
This issue existed already for a long time but could only rarely be observed because of the hacky "cache check" logic.
The changes from 9ed05f3fc0 caused this issue to be consistent in case the new service worker could not get installed due to not meeting the required installation prerequisites. If that was the case, the cache check always returned false causing the above-described behavior.
fixes#929
The server is not requesting these images through an extension; thus, it is not a problem to send them directly to the server since there is no problem of getting the server "stuck" due to extension rate limits.
Improves the image request logic by making it possible to check if an image is cached.
In case an image is cached, it does not need to be put into the image queue because it won't be sent to the server.
When scrolling really fast, the delta value could increase, which leads to incorrect trackpad detection, because the delta value is not consistent over consecutive scroll events
In case the connection got lost and the client reconnected again, there is a possibility that the auth token is expired. This will lead to the subscription to fail due to being unauthorized.
This, however, does not close the connection and the heartbeat messages are still being sent and received.
Thus, after refreshing the token, the connection is still alive but not authorized, which is the same as not having the connection open.
In case the tab gets resumed, it does not get fully reloaded but does reexecute js.
This caused the initial timestamp to get overwritten and lead to the forced reload detection to be incorrect, because it assumed that the app just got loaded.
In case the previous page in the history should be ignored, the back button just went back to the root page (library).
Instead, it should try to go back to the next page in the history until there is no page to go back to anymore.
Most cases were preventing some requests from getting triggered. However, this is not necessary anymore since with 6636cc66b1, this is handled globally now
