I just had a mega win idea.

When you try to save a scripted replay, e.g., if you the use rocket boots script, the replay will be all glitchy and it won't save right. The only way to fix that is to shift-space every frame and move a joint every frame. So, I was thinking, what if somebody just made a script that alternated between contracting and extending tori's neck every frame. Then you could just add that to the other scripts.

I don't know a whole lot about scripts but that would fix the replay glitching, right?

It's already been tried, and doesn't work. Also, the replays would become enormous really fast. A single frame with inputs adds around 5.3 KB to the replay file size. The game also bugs when trying to read files with over a certain number of frames with inputs.

Best bet is to write a proprietary replay format that stores forces or force equations, and then have a proprietary Lua replay parser that handles playback.

Maybe I'll work on one of those in about a week.

...
A script changing things causes the problem, and you think a script moving the neck will fix it?

It's a good idea. It just doesn't work. Normally, you can force a replay frame write by changing joint states, grip states, or breaking or fracturing something, which causes a frame to be written with joint states, breaks and fractures, and position and rotation information. But that only works if the joint state changes during or right after a freeze for inputs.

Unfortunately, using the Lua command to freeze the game, doing your joint state changes and then unfreezing via Lua doesn't trigger a frame write to the replay.

But since lua makes the limbs move, won't it not show it?

So how could you fix a replay that is glitchy, if you did not use any scripts or mods?

Really kick ass observation and predicting skills, good knowledge of the syntax of .rpl files, and a hell of alot of time.

Do you have these things Nuya?

I have a hell of alot of time and that's it :P

I actually did this when I was working on punchout... I basically wrote a function that collected joint changes each frame and dumped them out to a rpl file in the right format with the right header, etc...

It works up until about 150 frames, then Toribash goes kablooey when it tries to load said replay. 150 frames is not enough to be useful, so I scrapped it. Like Dafe said, size is an issue, even without storing the quats and other stuff.