System constructor cannot be found? about molly.jl

[axsk]

I ended up using exactly that approach and it's fine. A little more verbose then a System(oldsys, coordinates=newcoords) but certainly no priority.

I was just surprised that the 'common' approach did not work (and the type parameter list was too long to spot the problem :D) and Accessors failed too. IMHO in the long run compatibility with Accessors or a 're-constructor' would be nice to have, but I'm fine off knowing about it now.

Thanks for the info!

[jgreener64]

Perhaps there is use for a System constructor that takes in an existing System and keyword arguments for any properties that should be changed, returning a copy.

This is now on master, e.g. System(sys; coords=(sys.coords .* 0.5)).

[jgreener64]

The type parameters of System require some calculation and don't map directly to the fields. This is to allow certain properties to be put into the type domain for dispatch and performance. If you are constructing directly with fields... you would need to specify the type parameters, see:

Molly.jl/src/types.jl

Lines 391 to 394 in b172d0d

	return System{D, G, T, CU, A, AD, PI, SI, GI, CN, C, V, B, NF, L, F, E, K}(
	atoms, atoms_data, pairwise_inters, specific_inter_lists,
	general_inters, constraints, coords, vels, boundary, neighbor_finder,
	loggers, force_units, energy_units, k_converted)

You can copy a System with deepcopy too. In general, though, I would use the keyword argument constructor for System since it won't be the bottleneck in most code.

Perhaps there is use for a System constructor that takes in an existing System and keyword arguments for any properties that should be changed, returning a copy. At the moment I just do something like:

sys2 = System(
    atoms=sys.atoms,
    coords=sys.coords,
    ...
)