Remove functionally duplicate artist/genre values that were read from
a file.
This caused a indexer crash in 3.1.2 due to the switch to music sets,
which no longer made duplicate values group the song twice. This then
cascaded to a failure in song finalization, as it expects there to be
the same amount of artists/genres as raw artists/genres.
Fix a regression where the loading process will never stop on a no-op
refresh operation.
This was an oversight made in the redocumentation of the loading
process.
Further document the music indexing process.
It's so aggressively parallelized as to require some more extensive
comments to actually make it clear what's going on.
Recognize artists sort, albumartists sort, and album artists tags.
These are written by mediafile, so they are probably also written by
other software.
Do not hang when an error halts the discovery process.
This was an oversight with the previous band-aid fix regarding handling
errors in music loading. If something failed, the channels would not
close, resulting in the main loop consuming the channel hanging.
There's probably a deeper issue causing this in 3.1.2, but with this
fix I can actually start digging for it.
Use sets for all child music information.
Unlike parent information, which usually has an ordering derived from
file information, child music information more or less doesn't, and
will be consistently re-interpreted by the app to apply user-configured
sorts.
Use a Set when storing all song, album, artist, and genre information.
These all have no consistent ordering due to parallelization, and
should be communicated as such.
Sort song informtion in all MusicParent instances.
This is a temporary hack to band-aid music consistency between reloads,
as with the aggressive parallelization song order is no longer
consistent.
Fix a regression where partial ReplayGain adjustments missing an album
or track component would not be indexed.
This was an oversight made when moving adjustments to the cache, as the
nullability model of ReplayGain adjustments changed.
Properly group albums by raw artist keys, instead of by raw artist
instances.
This was an accidental regression in 3.1.1 that resulted in duplicate
albums in some circumstances.
Resolves#475.
Only use fallbackToDestructiveMigration for all databases.
I thought you had to selectively enable downgrading as well, but
apparently that disables other destructive migrations. This was not
obvious at all and caused countless issues. Absolutely flooring.
Do not extract ReplayGain adjustments on the fly, instead doing them as
we go along.
This prevents an issue where the ReplayGain information would only be
applied a short period after playback start, which is heavily jarring.
Make UserLibrary return some kind of error indicator if something
fails.
I don't have the framework for how the app will display these errors
just yet.
Make it so that the DeviceLibrary constructor streams all song
information instead of building the library on completion.
This has no measurable effect on loading times, but does appear visibly
faster to the user since the loading process is no longer stuck on the
"Loading your music library" step.
Actually bother to make the way music is grouped consistent, based on:
- The first track for albums
- The earliest album for artists
- The first song for genres
Fix redundant separator parsing obliterating loading performance.
If there are no separators configured, the parsing function would not
short-circuit and instead do a useless O(n) iteraton (including
escaping!), massively reducing performance.
Song build performance still isn't the best (blame intelligent
sorting), but it's definitely better now.
Fix a few problems with the current comparison algorithm:
1. It wasn't actually comparing the raw music information, only the
UIDs, which was redundant.
2. The comparison in the main music loading process occurred on the
main thread, which causes massive freeze-up issues.
Resolves#457.
Log whenever a listener is added/removed.
This might make it easier to track concurrency issues stemming from
listener dispatch and addition occuring at the same time.
Group albums implicitly linked to an artist via the "artist" tag into
their own section called "Appears on".
This makes Auxio's artist model a bit more apparent to users.
Resolves#411.
Reduce the accepted datatype of extractors down to a list of songs,
moving the other datatypes to the UI layer.
This massively reduces the amount of components that must be managed,
and enables functionality related to playlist editing.
