* Rename removeFromQueueByPredicate to removeFromQueueIf
Follow-up to PR comment in #9511
* Make Download hashCode stable
Mutating pages would previously change the Download hashCode, which
breaks HashMap lookups.
* Convert Donwloader subscription to coroutine
Replace downloadsRelay with activeDownloadsFlow. Instead of managing
a PublishRelay independent from the queue, derive a Flow of active
downloads directly from the queue StateFlow. (This will allow
updating the queue without pausing the downloader, to be done in a
follow-up PR.)
When a download completes successfully, the downloads is removed from
queueState. This updates activeDownloadsFlow and causes the
downloaderJob start the download job for the next active download.
When a download fails, the download is left in the queue, so
queueState is not modified. To make activeDownloadsFlow update
without a change to queueState, use transformLatest and use the
Download statusFlows to suspend until a download reaches the ERROR
state.
To avoid stopping and starting downloads every time
activeDownloadsFlow emits a new value, maintain a map of current
download Jobs and only start/stop jobs in the difference between
downloadJobs and activeDownloads. To make sure all child download
jobs are cancelled when the top-level downloader job is cancelled,
use supervisorScope.
* Remove obsolete main thread references in Downloader
Thread safety of the queue state used to be guaranteed by running all
queue mutation on the main thread, but this has not been true for
some time. Since the queue state is now backed by a StateFlow,
queueState can be safely updated by any thread.
Issues:
- Apache implementation relies on methods unavailable on lower Android API levels
- Using input stream implementation doesn't seem to read some files properly, but using
ZipFile implementation still requires reading the entire thing into memory
Fetch each source image URL immediately before downloading each image
instead of fetching all URLs and then downloading all images.
Source image URLs may change, so the downloader may fail if there is
too long a delay between fetching the image URL and downloading the
image.
Closes#9255, sort of. The example is a bad edge case though, where chapter numbers are repeated across versions,
so realistically only the first 113 will appear but the later 113(s) won't despite being "different". Those realistically
should be in different manga entries, not all mixed together, so this is just a crappy source.
