Daniel Karrasch

Regardless of details, I like your generalization, and obviously, you come with a use case. Does mat-vec-multiply work with your (generalized) BlockMaps and our current `mul!` method?

> it will actually make the BlockMap more type unstable I'm afraid I don't follow. IMHO, it doesn't get any more type stable if you have all the types in...

@jagot I had another look at your PR (finally!). I think this is really a great extension. So, just for myself to clarify, this addresses the case when the LinearMaps...

Thanks, @jagot, for the update. I'm not sure I understand every detail, but I'm sure you're well familiar with what the community has to offer in that direction (like DiffEq...

`DifferentialEquations.jl` has lots of tooling around defining (possibly time- and state-dependent) linear operators, though I'm not sure if they always need to be backed by a matrix. In which way...

> Sorry bothering you with my problems. Don't worry, that's what we're all here for. Since `a` is a function, the LinearMap constructor thinks this should be a `FunctionMap`, which...

> I am not sure, but I have the impression that each time-evaluation of the operator `A(t) `creates an instance of the operator. Could this lead to some performance loss...

There is a new project in development, that does what I was considering to do here at some point: make `LinearMaps` potentially parameter-dependent, i.e., merge `LinearMaps.jl` with the usual DiffEq...

Hi @felixhorger! Thanks for the report. I believe the relevant piece of code for you is https://github.com/JuliaLinearAlgebra/LinearMaps.jl/blob/bd10c6c74ada158d6b820694ccde21235ed01538/src/composition.jl#L176-L182 As you say, when you have a non-mutating `LinearMap`, then `_unsafe_mul!` essentially computes...

Can you try this with the current master branch? If #194 improved things, we should perhaps release it rather sooner than later.