Various performance issues

[karl2883]

Hi! I have been profiling modded Celeste maps recently. The GetSensitivity function of SensitiveFlags.cs has a measurable performance impact (around 1.3% of Update time in an average map) even when Chronia Helper is not used at all. This seems to be because of Regex use, which is not cheap performance wise. The function gets called multiple times per frame through MapProcessor.OnLevelUpdate -> ChroniaFlagUtils.SetFlag -> SensitiveFlags.GetSensitivity.
Perhaps a decent solution would be caching whether flags are sensitive? A dictionary lookup seems like it would be significantly cheaper than a Regex check. Maybe you can think of other solutions as well.
Lmk if you have any questions! :)

[karl2883]

Additional important performance issue that I discovered:
ITeraBlock.cs has two hooks that have a significant performance impact on Actors: OnActorUpdate and OnSetLiftSpeed (which contributes to noticeable lag when there are many actors in a room). The reason for the performance impact is the use of DynamicData.For (expensive!) in those two hooks. To fix this, I would suggest not using DynamicData and instead having a component for entities that stores tera block related data (ideally the component would only exist once an entity actually gets affected but even if it always exists for actors it would already be a huge improvement!).

[karl2883]

Ok I'm just gonna dump all my performance issues here :)

The next one is the PolygonCollider. It hooks Collider.Collide, and checks whether one of the entities is a Polygon. The hook is relatively expensive (around 1.8% of Update time in a fairly laggy map) because the underlying method gets called so frequently, the isinstance calls are not completely free. I would strongly recommend changing the PolygonCollider to inherit ColliderList instead, which means that it would not require any hooks.
Also, the implementation seems to basically be a 1 to 1 copy to the one from VivHelper, where I mentioned the same thing. I'm not sure why this one exists, currently this inefficient hook basically runs twice lol.

Also, I should mention, if you don't have time to address these issues right now, I can also resolve them myself and make a pull request if that is preferred.