A few facts of life:
- Guest difficulties are at this point a staple of mapping.
- People are very much used to flinging `.osu`s around (because there's
no better alternative).
- Currently there are two ways to get an `.osu` out of lazer. You can:
- Export the beatmap as "compatibility" to an `.osz`, then
transmogrify the `.osz` to a `.zip`, then extract the `.zip`, then
pluck out the `.osu`. This is the "correct" way to make sure stable
works, but is also stupidly arcane.
- Use "edit externally" to mount the beatmap files to disk, then
copy-paste out the `.osu`. This is the *wrong* way to make sure
stable works, because the mounting process exposes the raw "for
editing" format with features stable doesn't support, but it the
actual easy one.
- Reports about guest difficulties exported from lazer "working wrong on
stable" are prevalent. Probably mostly because of the preceding point.
What this PR does is introduce a *third* method to export an `.osu`,
which is designed to be both the easiest one yet *and* correct. I am
hoping this will curb the complaints until support for direct submission
of guest difficulties is added - which I still hope to see, but it will
be a significant effort *client-side* (the server side has been ready
for years now).
And yes, you will notice that much of the code added in
`LegacyBeatmapExporter` related to manipulation of the path is
copy-pasted from `LegacyExporter`. I don't care enough to invent
protected / abstract / whatever else OOP faff for something that may not
survive review and is mostly a weird semi-temporary wart.
Done for two reasons:
- During review it was requested for the logic to be moved out of
`BezierConverter` as `BezierConverter` was intended to produce
"lazer style" sliders with per-control-point curve types,
as a future usability / code layering concern.
- It is also relevant for encode-decode stability. With how the logic
was structured between the Bezier converter and the legacy beatmap
encoder, the encoder would leave behind per-control-point Bezier curve
specs that stable ignored, but subsequent encodes and decodes in lazer
would end up multiplying the doubled-up control points ad nauseam.
Instead, it is sufficient to only specify the curve type for the
head control point as Bezier, not specify any further curve types
later on, and instead just keep the double-up-control-point for new
implicit segment logic which is enough to make stable cooperate
(and also as close to outputting the slider exactly as stable would
have produced it as we've ever been)