add ismassaction and get rid of some temp arrays

[isaacsas]

@TorkelE Mind taking a look that this looks ok to you? ismassaction is so useful I've found that I wanted a separate routine for it; we can drop it from DEBio now. This also re-arranges a bit so that the ratelaw calculation is only called for non-mass action jumps. On a big system that should help system construction time.

[isaacsas]

@ChrisRackauckas I think with this we are done with ReactionSystems and their conversions over here, modulo we should add a RegularJump system and conversion with the new leaping code coming online.

[isaacsas]

but let me know if you see anything we've missed.

[ChrisRackauckas]

Looks good. Could a follow up PR flesh out the docs for JumpSystem and ReactionSystem?

We need to think about how we want to do the conversions there to regular jump problems. I think full_regular_jump or something like that would be to get to standard tau-leaping, and we want to leave open things like hybrid_regular that have double ups with extra parameters etc.

[isaacsas]

Yup, doc updates are on the TODO list next over here. I didn't want to advertise all this stuff till we had it working / somewhat finalized, but we are there now.

Hybrid systems will need a bit of thought, but I can go ahead with the basic RegularJumpSystem to start at least. If we later allow RegularJumpSystems to be hybrids instead of making a separate HybridJumpSystem we can update over here as necessary.

[isaacsas]

Should hybrid be working at the DiffEqJump / StochasticDiffEq level already?

[ChrisRackauckas]

Should hybrid be working at the DiffEqJump / StochasticDiffEq level already?

Yes, you just can't mix that hybrid with actual SDEs/ODEs yet, which is my next goal.

Hybrid systems will need a bit of thought, but I can go ahead with the basic RegularJumpSystem to start at least. If we later allow RegularJumpSystems to be hybrids instead of making a separate HybridJumpSystem we can update over here as necessary.

It would be good to keep hybrid separate for SimpleTauLeaping.

One thing that would be interesting too would be a static splitting, but it might be difficult to choose which jumps go where.

Then dynamic would be a bigger goal down the line, which is possible by making every jump part of both and adding parameters, but we'd need solvers to "know about it" in a sensible way.

[ChrisRackauckas]

https://travis-ci.org/github/SciML/ModelingToolkit.jl/builds/688124601#L570

That tolerance should be made higher since it seems reliant on compiler optimization levels.

[isaacsas]

I'm away from my computer the next few hours, but will bump that up (and double check the generated code) later tonight.

[TorkelE]

Good stuff.