Closes https://github.com/ppy/osu/issues/29832.
The underlying reason for the incorrect sample playback was an equality
comparer failure.
Samples are contained in several pools which are managed by the
playfield. In particular, the pools are keyed by `ISampleInfo`
instances. This means that for correct operation, `ISampleInfo` has to
implement `IEquatable<ISampleInfo>` and also provide an appropriately
correct `GetHashCode()` implementation. Different audible samples must
not compare equal to each other when represented by `ISampleInfo`.
As it turns out, `VolumeAwareHitSampleInfo` failed on this, due to not
overriding equality members. Therefore, a `new
HitSampleInfo(HitSampleInfo.HIT_NORMAL, HitSampleInfo.BANK_NORMAL,
volume: 70)` was allowed to compare equal to a
`VolumeAwareHitSampleInfo` wrapping it, *even though they correspond to
completely different sounds and go through entirely different lookup
path sequences*.
Therefore, to fix, provide more proper equality implementations for
`VolumeAwareHitSampleInfo`.
When testing note that this issue *only occurs immediately after
placing an object*. Saving and re-entering editor makes this issue go
away. I haven't looked too long into why, but the general gist of it is
ordering; it appears that a `normal-hitnormal` pool exists at point
of query of a new object placement, but does not seem to exist when
entering editor afresh. That said I'm not sure that ordering aspect of
this bug matters much if at all, since the two `IHitSampleInfo`s should
never be allowed to alias with each other at all wrt equality.
This fixes an issue identified with the WASAPI implementation in
https://github.com/ppy/osu-framework/pull/6088. It has no real effect
on current `master`, but fixes a deadlock that occurs with the
aforementioned framework branch when one lets a preview track play out
to the end - at this point all audio will stop and an attempt to perform
any synchronous BASS operation (playing another track, seeking) will
result in a deadlock.
It isn't terribly clear as to why this is happening precisely, but
there does not appear to be any need to stop and seek at that point,
so this feels like a decent workaround even if the actual issue is
upstream (and will unblock pushing out WASAPI support to users).
This paves the way for the framework code quality change
(https://github.com/ppy/osu-framework/pull/4873) which stops exposing
the constructor.
Most of the restructuring here is required to give `PreviewTrackManager`
an adjustable target to apply the global mute.
The previous implementation of `SampleInfo`'s equality members was not
completely correct in its treatment of the `sampleNames` array. While
`Equals()` compared the values of `sampleNames` using `SequenceEqual()`,
therefore performing a structural check that inspects the contents of
both arrays, `GetHashCode()` used `HashCode.Combine()` directly on the
arrays, therefore operating on reference equality. This could cause the
pooling mechanism of samples to fail, as pointed out in #11079.
To resolve, change the `GetHashCode()` implementation such that it also
considers the contents of the array rather than just the reference to
the array itself. This is achieved by leveraging
`StructuralEqualityComparer`.
Additionally, as a bonus, an array sort was added to the constructor of
`SampleInfo`. This is intended to be a "canonicalisation" processing
step for the array of sample names. Thanks to that sort, two instances
of `SampleInfo` that have the same sample names but permutated will also
turn out to be equal and have the same hash codes, given the
implementation of both equality members. This gives `SampleInfo`
set-like semantics.
