Always write UGRID (results) file

[evetion]

Non ugrid follow up of #969.

Does it have to be non-UGRID? And why? Color and width rendering for flow works out of the box.

QGIS in combination with MDAL is the only data model that seemed to work align well; temporal vector data was very slow because it didn't index into a time axis; it was filtering over the entire table continuously (possibly on each pan/zoom action).

If it's a case of matching the geometry, this can still be done easily with UGRID, just repeat the flow values for all the "sub-edges" of the true link. It'll need some modification of the to_xugrid method (possibly a bool argument) -- it should probably default to using the detailed geometry if available.

Originally posted by @Huite in #2171

We can always write a UGRID netcdf from Python, and Julia could write the results to it? This would also make visualization easier.

EDIT: we now have #2497, which is NetCDF-CF, not UGRID.

[visr]

I'd prefer to just write the whole UGRID from Julia. Also for to_fews we are working towards Delft-FEWS being able to read our UGRID files. If the UGRID look good and load fast in QGIS out of the box that is also a good argument to do this.

There is the currently private https://github.com/Deltares-research/UGrid.jl that we made at the start of Ribasim. It would be a bit more work, but we could also consider writing it on top of e.g. https://github.com/gbaraldi/StaticHDF5.jl.

[evetion]

Ok, so temporarily solution is QGIS (this week) and then work on UGrid in one of the next sprints?

[Huite]

There is the currently private https://github.com/Deltares-research/UGrid.jl that we made at the start of Ribasim. It would be a bit more work, but we could also consider writing it on top of e.g. https://github.com/gbaraldi/StaticHDF5.jl.

Wouldn't you need the Julia equivalent of h5netcdf as well? It's probably mostly a one-to-one translation, but it's still about 2000 or 3000 lines.

(It's a nice and useful project: Python h5netcdf is sometimes faster, sometimes slower. In particular, h5netcdf is slower when there is a large amount of variables since it's running a slow Python loop -- a Julia h5netcdf is probably faster all the time!)

[visr]

Yeah something like h5netcdf would be great, though probably it would be less work to just call HDF5 APIs directly to make valid NetCDF for our use case only, as a start.

Note there is also the pure Julia https://github.com/JuliaIO/JLD2.jl which could potentially be used as well, and actually is already an indirect dependency.