Overwritten get_contravariant_vectors leading to error

[benegee]

Currently some (all?) examples fail to run threaded.

E.g. julia -t 2 --project=. -e 'include("../examples/elixir_moist_euler_moist_bubble.jl")' leads to:

ERROR: LoadError: MethodError: no method matching static2val(::Int64)
The function `static2val` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  static2val(::Static.StaticInt{N}) where N
   @ TrixiAtmo ~/trixi/TrixiAtmo-vanilla.jl/src/solvers/dgsem_p4est/containers_2d_manifold_in_3d_cartesian.jl:48

Stacktrace:
  [1] get_contravariant_vector
    @ ~/trixi/TrixiAtmo-vanilla.jl/src/solvers/dgsem_p4est/containers_2d_manifold_in_3d_cartesian.jl:51 [inlined]
  [2] macro expansion
    @ ~/.julia/packages/Trixi/6nxkA/src/callbacks_step/stepsize_dg2d.jl:97 [inlined]
...
  [8] max_dt ...
    @ Trixi ~/trixi/TrixiAtmo-vanilla.jl/run/dev/Trixi/src/callbacks_step/stepsize_dg2d.jl:93
  [9] calculate_dt
    @ ~/trixi/TrixiAtmo-vanilla.jl/run/dev/Trixi/src/callbacks_step/stepsize.jl:91 [inlined]
...

The reason is that @batch is used in max_dt (https://github.com/trixi-framework/Trixi.jl/blob/e93de2a5f5c9cab36e7593dc2fc7cdb8d216af48/src/callbacks_step/stepsize_dg2d.jl#L90), which turns Arrays into PtrArrays (https://github.com/JuliaSIMD/Polyester.jl?tab=readme-ov-file#important-notes). Thus, TrixiAtmo's get_contravariant_vectors

TrixiAtmo.jl/src/solvers/dgsem_p4est/containers_2d_manifold_in_3d_cartesian.jl

Line 49 in e7c1de7

@inline function Trixi.get_contravariant_vector(index, contravariant_vectors::PtrArray,

is picked up.

Can we somehow refine the dispatching s.t. "usual" setups keep using Trixi's version?

[tristanmontoya]

Thanks for pointing this out. I get the same issue as you for the new equation sets that use the "standard" Trixi.jl solvers, but the Cartesian and covariant solvers for PDEs on surfaces still seem to work with threading. I'll have to look what the type parameters of PtrArray are, but we could potentially dispatch on one of them, or otherwise make the Cartesian solver for surfaces not use the get_contravariant_vector function at all (the latter approach would likely require repeating more of Trixi.jl's code in TrixiAtmo.jl, which would of course be undesirable).

Edit: I refined the dispatch in #62 and it seems to work, but perhaps someone more familiar with PtrArray can comment on whether what I did makes sense and/or suggest a better solution.