Christopher Rackauckas

Maybe it could be due to not using a residual control?

I wonder if we can save out values of the history function and plot that just to make sure those are following the true trajectory. I think there may be...

Now he's something that's interesting: ```julia function f_dde_rober4(du, u, h, p, t) # compute repeating terms x = 4e-2 * u[1] y = 1e4 * h(p, t-1e-16)[2] * u[3] z...

I think it's highlighting that we're not handling step rejections properly? It seems we `advance_ode_integrator!(integrator)` on every `perform_step!`, but don't pull back if the step failed or if `integrator.EEst>1`.

That test case was bad because it requires using the isout handling. However, I am narrowing down a bug which is requiring isout when dt is constrained to be smaller...

On the `bug` branch, it errors if history extrapolation is ever used. I run the problem: ```julia using DelayDiffEq function f_dde_rober4(du, u, h, p, t) # compute repeating terms x...

Nevermind. That "bug" was just because finite differencing went too far.

These docs mysteriously started building correctly again. SciMLSensitivity.jl and SciMLDocs both have the issue too though...

I just looked in there: that's just a standard Legrangian interpolation, right? Is there really no Julia package which already does this that you can just call? How does Polynomials.jl...

Ahh, I looked through the code again. I misread it the first time as you were constructing the polynomial, instead of directly doing the interpolation at the point. It think...