Jump timing issue with couples/routine mode

[ianklatzco]

I mentioned this in IRC, but wanted to post it somewhere publicly-indexed.

---

Context: We run a yearly couples tournament at RIP.
SM5 rates jumps differently from 3.95. Couples in 3.95 are actually a doubles chart.
The second player's notes are offset by a 1/192nd, and there's a negative warp/stop? value that moves them to the correct timing.

I believe SM5 made a change that breaks this: all jumps receive the same judgment. this breaks couples for our (tournament) use case, because two separate players get different timings.

(this may be for doubles only, i'm unsure about routine in SM5)
(some more info about the problem: https://twitter.com/pluto_chan/status/932608291011809280)

would anyone happen to know if routine mode that this issue?

from tertu in IRC:

in routine mode all notes are judged together
on the same row
it's more that's the way the people who implemented it believed it should be

---

Our current solution is to switch to 3.95 via a wireless button, which allows any DDRIllini members to use the couples game mode. It works for our purposes, but I'd love to get SM5 support.

(Maintaining a 3.95 setup is also nice for legacy mods content that hasn't been converted by @MrThatKid, yet. thank you for all your work!)

[shakesoda]

do both couples and routine modes have this problem? they're not the same thing - and this behavior is correct also annoying for routine.

edit: updated after some testing and further discussion on irc

[ianklatzco]

according to tertu, yes (see quoted block above)

[ianklatzco]

after some research:

• routine mode works by coloring rolls, and rows with rows on them, a different color. this appears to reflect how Only One Couples works. pushing / to switch between p1 and p2 in the editor.
  • the simfile stores two doubles charts, separate by an &
• as tertu said, routine mode scores notes on the same beat as jumps (ie, the later player's timing is the total timing.
• neither lambda nor simply love expose routine mode charts in a player-accessible way
• the 192nd & stops thing appears to be a spooty-created hack to address the "jumps count as one judgment" timing issue. maybe it's like this for 3.95
  • the reason this doesn't cause visual stutters is that at reasonable high bpms (>80), they're sub-16ms (ie sub-frame).

for my purposes, i think i'll continue with the 192nd hack and-couples-are-doubles for RIP11.

todo/roadmap for 192nd hack:

• make a sm5.1 compatible noteskin for couples
  • Negative stops don't work the same way. Might be fixable with warps & stops.
  • Seems to be totally broken in SM5.
• maybe fgchanges (for runaround prompts) need to be fixed.

todo/roadmap for proper fix (hard and i don't know how i'd do it):

• add to the SSC/SM spec for routine charts for a 2nd player exists in routine already
• have jumps judged independently

[ianklatzco]

SM5:

• A stop pauses the chart for that many beats.
• A warp skips that many beats of the chart.

[ianklatzco]

Here's a script that converts SM5 routine to our 192nd hack: http://r21freak.com/SpootyBiscuit/Couples/Couple%20Conversion%20(py).rar

(also available here https://github.com/ddrillini/uiuc-wendy/tree/master/couples/convert-sm5routine-to-3.95double)

I believe modifying that script to insert warps instead of negative stops will solve the problem. Tried & no success. Specifically, P2's notes can't be hit at all. Might be an unrelated issue, but at the very least, solving this problem will not be a simple hack as it was in 3.95.

[quietly-turning]

Maybe helpful, here are my (old) notes on routine mode in SM5:

---

SM5 Routine Mode as of Aug 27, 2014

0. On my ScreenSelectMusic, I made Routine charts accessible if Versus style
   was selected on ScreenSelectStyle with AutoSetStyle set to true by hijacking
   StepsTypesToHide=CustomSetStyle() and reloading metrics before ScreenSelectMusic.

function CustomSetStyle()
	local StepsToHide = "dance-couple,dance-solo,dance-routine"
	
	if SL.Global.CurrentStyle == "None" then
		StepsToHide = ""
	elseif SL.Global.CurrentStyle == "Versus" or SL.Global.CurrentStyle == "Single" then
		StepsToHide = "dance-double,dance-couple,dance-solo"
	elseif SL.Global.CurrentStyle == "Double" then
		StepsToHide = "dance-single,dance-couple,dance-solo"
	end
	
	return StepsToHide
end

ISSUES:

1. Both players need to be joined, but all input goes to P1 (hardcoded? unsure.)
   in gameplay, so P2 rapidly fails and earns no points. This does not happen with AutoplayCPU.

2. P1's gameplay modifiers are used; P2's are ignored. Not necessarily a problem,
   per se, but confusing initially.

3. Noteskins are (more or less) forced by _fallback metrics:

RoutineNoteSkinP1=RoutineSkinP1()
RoutineNoteSkinP2=RoutineSkinP2()

Those functions are defined in 03 Gameplay.lua, and simply return strings
dependent on game:

local RoutineSkins = {
	dance	= { P1 = "midi-routine-p1", P2 = "midi-routine-p2" },
	pump	= { P1 = "cmd-routine-p1", P2 = "cmd-routine-p2" },
	kb7		= { P1 = "default", P2 = "retrobar" },
	-------------------------------------------------------------
	default	= { P1 = "default", P2 = "default" }
}

Again, not a problem, per se, but worth... thinking about.

4. P1 must be the MasterPlayer via my #0 hack. If P2 was the MasterPlayer,
   Routine steps were not listed in the stock MusicWheel.

5. Arguably unrelated, but with my #0 hack, if P2 selected double style, the
   music wheel displayed ONLY routine steps. If P1 selected double style, double
   steps were shown properly (and routine hidden, correctly) but that player was
   forced to use P1's input. Pressing P2's start could latejoin P2 which would
   crash when any song was selected.

6. Since single steps and routine steps could coexist in my MusicWheel, I tried
   entering PlayerOptions while P1 was on a routine steps and P2 on a single steps.
   P2 was ignored; my ScreenPlayerOptions only listed the routine steps as selectable.
   The inverse scenario held true; with P1 on a single steps and P2 on a routine,
   ScreenPlayerOptions only listed the single steps as selectable. This was actually
   better than I'd anticipated (I was expecting a crash), but I'm noting it here.

7. _fallback's IsRoutine() method could stand to be improved for the sake of themers.
   For example, it works properly ScreenGameplay (used it in my ScreenFilter code) but not
   ScreenPlayerOptions (when you are currently already locked into playing a routine chart).

[ianklatzco]

I've tested. It looks like SM5 converts negative stops to warps (if you re-save the simfile in the editor, the SSC will have warps, tho the sm will still have negative stops). There is no difference in behavior between the two while playing.

(Specifically, P1, the un-warped player, cannot hit notes that appear both on P1 and P2. P2, the warped player, has no issue.)

It seems SM5 broke Spooty's hack to judge two players in couples differently.

I imagine I'll have to modify SM5 source to fix this problem.

[ianklatzco]

Possible locations:

• ScoreKeeper*.cpp
• TimingData.cpp
• ScreenGameplay.cpp

Pausing for now.

[ianklatzco]

I'm interested in discussing whether couples and routine should be separate.

Basically, there's two potential games here:

• Versus doubles
• Cooperative doubles

"Routine" is, I believe what it's named on the pump side of the house. "Couples" is what it's named on the ITG customs side of the house. I think it should be internally called couples, and exposed as such as well.

One idea: engine exposes both scores: versus and coop. Lua can choose which of these to display.

In any case, I think the routine/couples distinction should be removed. very few charts have been written for these modes, if any.

Here's where routine/couples are defined (GameManager.cpp):

stepmania/src/GameManager.cpp

Line 432 in 27fdfb3

static const Style g_Style_Dance_Routine =

[ianklatzco]

I've modified Simply Love to expose routine on the couples branch of the DDRIllini fork.

There are a few engine bugs, as per Dan's notes.

I would recommend naming them Versus and Coop Couples. I suggest that the choice between the two be at this screen of Simply Love:

Handling high scores between the two different modes is a decision that needs to be made. If they're separate Styles, this is easy. Otherwise, I am less sure.

Personally, I don't really care about versus doubles, but I see why it's useful/interesting. I think there could be some very interesting charts, but I think they would not be terribly commonly used, because a skill comparison for doubles isn't really something people care about, and it's very hard to have fair charts (you'd need the same patterning twice, which doesn't really work musically in most cases). I don't see why you'd want a versus doubles mode to track separate scoring.

So, I think that the best option is to have one couples Style in the engine, and have the theme-side distinction between individual and collaborative scores. For my use case and modification to Simply Love, I will most likely print all three scores (coop and 2 indiv.).

I think I will:

• Rename routine style to couples.
• Scrap couples style.

@shakesoda, I am interested in your opinion on my conclusions here, and if you think they're the right way to proceed or not.

[ianklatzco]

Well, I:

• spent 3 hours spelunking through cpp source
• spent 3 hours spelunking through fallback and simply love lua

SL: given a chart w 32 notes, i'll have 32 misses on P1 if I don't do anything. P2 will have nothing (0 misses).

(Keep in mind stepmania will lie to you and only show you notes from P2 in the total note count on MusicSelect. Another bug, I guess.)

Fallback handles couples in a more reasonable way; namely, that all input is forced to P1. P2 still gets 32 misses (32), but P1 gets 32 inputs.

It seems that fallback has both ?players? actually enabled? And SL doesn't?

From diving through C++, I learned (again) that stepmania is terrible spaghetti.