docs v7 rework

[ArnoStrouwen]

Is this template what you had in mind @ChrisRackauckas?

[ChrisRackauckas]

Yes exactly.

[ArnoStrouwen]

I came up with a way to automate the same example over all the libs, but it's a pretty ugly hack.

[ChrisRackauckas]

Looks like it works, I'll take it.

[ArnoStrouwen]

The current dev version of docs removed documentation about threading, while this is still there on stable. Is this a regression?

[ChrisRackauckas]

That's a regression

[ArnoStrouwen]

Are the Anas* methods really OrdinaryDiffEqAlgorithm and not OrdinaryDiffEqAdaptiveAlgorithm?
Same question for:

RKO65
RKM
MSRK*
PSRK*
Alshina6

[ChrisRackauckas]

Are the Anas* methods really OrdinaryDiffEqAlgorithm and not OrdinaryDiffEqAdaptiveAlgorithm?

IIRc they are non-adaptive. You can check the perform_step! on whether they calculate an EEst.

[ArnoStrouwen]

How does HighOrderRK fit into the story? If I understand correctly, the recommended solver page skips over them in favor of Vern for lower tolerances.

[ChrisRackauckas]

Yes they aren't recommended for most things. DP8 is legacy fallback to dop853 but Verner is just generally better unless you really need a non-lazy interpolation.

[ArnoStrouwen]

AB5: Adams-Bashforth Explicit Method
The 3-step third order multistep method. Ralston's Second Order Method is used to calculate starting values.
"""
struct AB5 <: OrdinaryDiffEqAlgorithm end

Seems weird to me the 5 from AB5 does not occur anywhere in the description.

[ChrisRackauckas]

Seems weird to me the 5 from AB5 does not occur anywhere in the description.

5th Order Adams-Bashforth method. History is prestarted using Ralston's 3rd order Runge-Kutta method.

[ArnoStrouwen]

I'm not sure what of this JVODE stuff needs a docstring:

OrdinaryDiffEq.jl/lib/OrdinaryDiffEqNordsieck/src/algorithms.jl

Lines 7 to 20 in dc0e1e7

	struct JVODE{bType, aType} <: OrdinaryDiffEqAdamsVarOrderVarStepAlgorithm
	algorithm::Symbol
	bias1::bType
	bias2::bType
	bias3::bType
	addon::aType
	end
	function JVODE(algorithm = :Adams; bias1 = 6, bias2 = 6, bias3 = 10,
	addon = 1 // 10^6)
	JVODE(algorithm, bias1, bias2, bias3, addon)
	end
	JVODE_Adams(; kwargs...) = JVODE(:Adams; kwargs...)
	JVODE_BDF(; kwargs...) = JVODE(:BDF; kwargs...)

And even less sure what then should be in that string.

[ChrisRackauckas]

It should have a warning as experimental. JVODE is an adaptive BDF implementation in Nordsieck form, similar to VODE. However, FBDF is recommended in almost all scenarios, and so this is more of a development artifact.

[ChrisRackauckas]

We should also audit to ensure that every algorithm has its source reference in the docstring. I think almost all do, but a few may be missing some.

[ArnoStrouwen]

@mortenpi we have some large pages in these docs. On Windows I need to do size_threshold_ignore = ["implicit\\Rosenbrock.md"]), instead of size_threshold_ignore = ["implicit/Rosenbrock.md"]), is this expected behavior?

[mortenpi]

On Windows I need to do size_threshold_ignore = ["implicit\\Rosenbrock.md"]), instead of size_threshold_ignore = ["implicit/Rosenbrock.md"]), is this expected behavior?

No. We should be normalizing the paths so they work cross-platform. Do you mind opening an issue?

Just to clarify: if you use the Unix-y path ("implicit/Rosenbrock.md"), it then doesn't work on Windows, right? A workaround you can do to get the PR passing would be to set is as joinpath("implicit", "Rosenbrock.md").

[ChrisRackauckas]

Looks like this got deleted.

[ChrisRackauckas]

This should have been moved to low storage RK.