perf(track): Persist GoslingTrackModels

[etowahadams]

Fix #1043
Toward #

Change List

• Persist GoslingTrackModels between draw calls
• Switch from structuredClone() to JSON.parse(JSON.stringify())

Screen.Recording.2024-02-22.at.9.47.14.AM.mov

Checklist

• Ensure the PR works with all demos on the online editor
• Unit tests added or updated
• Examples added or updated
• Documentation updated (e.g., added API functions)
• Screenshots for visual changes (e.g., new encoding support or UI change on Editor)

[etowahadams]

Persisting GoslingTrackModels is something I originally tested in #1039. This offers substantial performance improvements since now the GoslingTrackModels don't have to be regenerated with every draw() call.

It does seem that JSON.parse(JSON.stringify()) is about 40% faster than structuredClone() in Firefox so I switched it back to that https://www.measurethat.net/Benchmarks/ShowResult/506888

[sehilyi]

This is nice! I assume this also increases the performance to some extent.

[sehilyi]

Looks great! Thanks a lot for such a quick fix!

[sehilyi]

JSON.parse(JSON.stringify()) being faster than structuredClone() is surprising. I thought it was the opposite.

[etowahadams]

Its worth noting that this reduces, but doesn't completely get rid of the freezing that sometimes occurs when zooming and panning. As far as I can tell, the reason for this is because of the HiGlass renderer, where an infinite loop can happen. The renderer detects that the scales have changed before the drawing has finished so it tries redrawing everything according to the new scales. But after it redraws, it sees that the scales have been updated again so it draws everything again.