Daniel Wennberg

An option to set the number of channels on the pulseaudio bridge would be greatly appreciated! For default behavior, either enforcing 2 channels or letting module-jack-(sink|source) default on their own...

I use the NI Komplete Audio 6 with 6 outputs: 2 pairs of analog channels, and SPDIF. When pulseaudio is started from cadence, it sees jack presenting a generic 6-port...

> For things that were bigger differences, (e.g. because they do bad scalar ops) I would say those would warrent seperate rules defined in GPUArrays.jl For sure, if a good...

Hitting the same issue with just a basic `jac` specification for `f1!`, i.e., when using `SplitFunction(ODEFunction(f1!; jac=jac1!), f2!)`. Also noting that if I move the `jac` specification to `SplitFunction` only,...

```julia using BenchmarkTools using OrdinaryDiffEq using LinearAlgebra f1!(du, u, p, t) = (du .= u); f2!(du, u, p, t) = (du .= -u); jac1!(jac, u, p, t) = copyto!(jac, I);...

Not ideal to have to add `f1!`'s jacobian to the SplitFunction instance. It's not the jacobian for the whole function, so if I try a non-splitting implicit algorithm on this...

Thought: For everything to work seamlessly with both IMEX and non-IMEX solvers and both manual and autodiff-generated jacobians, maybe the cleanest solution would be to disallow a custom `f.jac` and...

Just to clarify, I wasn't thinking about this feature from an efficiency perspective, as in people deliberately choosing to use SplitFunction because of an efficient way to split the Jacobian...

> If [`f1.jac`]'s defined and someone puts this into Rodas5, what does it do? My mental model here is that the jacobian of `F::ODEFunction` is simply called as `F.jac(u, p,...

Doesn't `AddedOperator` circumvent all such issues? `f1.jac` and `f2.jac` would always be filling their own separate arrays, whether sparse, dense, banded, or otherwise structured, preallocated or not, and then wrapped...