Steven G. Johnson

Just needs someone to add a test and create a pull request.

Here is a more concrete (but still very basic) implementation: ```jl import Richardson using DifferentialEquations function solveextrap(prob, alg, args...; kws...) DiffEqBase.isadaptive(alg) && error("non-adaptive algorithms only") opt = Dict(kws) @show dt0...

> However, given all of that though... I am not too bullish on extrapolation in ODEs because in any case you extrapolate, you could in theory derive a more efficient...

> I think the difficulty is that, if you use extrapolation, do you keep the SSP property and what's the SSP coefficient in that case? My suggestion was to use...

(But if you're using this for PDEs, you'd probably want to increase the spatial resolution at the same time you are increasing the time resolution, in which case you'd want...

Alternatively, since you only need a low-level interface to matplotlib, you could call `matplotlib.pyplot` directly via PythonCall.jl and only borrow small pieces of PythonPlot.jl as needed.

> Do you think there would be any TTFP improvement? Not currently. Right now, PyCall and PyPlot are significantly faster to load than PythonCall and PythonPlot. The following timings are...

It just draws an arrow at an arbitrary position in the plot. Here's an example with PyPlot: 

I think it's totally fine for AbstractFFTs to depend on ChainRulesCore…

Looks good, modulo a couple of minor tweaks noted above.