Christian Schilling

@lbenet, the current release version still prints these warnings. Maybe you could make a new patch release.

I cannot explain why, but if you replace https://github.com/JuliaDiff/TaylorSeries.jl/blob/16fe4790a3eeec0d3e6afca5753d0d06696e4eb6/src/evaluate.jl#L217-L218 by ```julia evaluate(a::TaylorN, vals::NTuple{N,

I finally debugged this properly and the problem is in the solver of the elimination algorithm. If `presolve` is disabled (which is activated by default), the problem does not occur...

The conversion to vertices and computing the convex hull is efficient enough in 2D. Setting up an LP has higher overhead. You would probably need a lot of constraints to...

Alternatively, we could have only one `OnDemand` type that dispatches on its argument.

Creating such a wrapper seems too complicated in practice. My new preference is to let the existing `Intersection` store the concrete intersection (there is already a cache, so this would...

For me this already crashes with the simpler set (see the stacktrace below). Not sure why this works for you. In any case, as written in the warning box [here](https://jump.dev/JuMP.jl/stable/tutorials/algorithms/parallelism/#Using-parallelism-the-right-way),...

The call to `minkowski_sum` does quite a lot in the background. In particular it converts the zonotope to a `VPolytope` and in this process it also removes redundant vertices. This...

I think we should at least use a clearer name then, like `backend_polyhedra` (not sure I like that name), because right now the difference is not transparent.

Maybe after #1493 these two options can actually be unified (not sure).