veccat([]) -> Sparsity instead of MX

[jgillis]

No description provided.

[jgillis]

But what if we want SX?
Is DM a useful common ground?

[jgillis]

The following passes:

    f = Function('f',[DM(0,1)],[])
    print(f)
    f = Function('f',[DM(0,1),SX.sym("x")],[])
    print(f)
    f = Function('f',[DM(0,1),MX.sym("x")],[])
    print(f)

So DM makes sense

[jgillis]

Check if C++ variants match up

[jgillis]

veccat, vertcat, horzcat, diagcat should all fall back to DM when receiving empty arguments.

[jgillis]

Do we give shape (0,1) also to vertcat, (1,0) to horzcat and (0,0) to diagcat? Or just in the case of veccat?

[jgillis]

Matlab?

[jaeandersson]

Yeah, why would we want to cast [] to Sparsity?

[jaeandersson]

Nevermind, I get it now. I think you can solve it by modifying the casting priorities in SWIG. They decide in which order checks are made. Sparsity should have lower priority than DM. MATLAB uses the same typemaps

[jgillis]

Changing casting priorities is opening a can of worms. vertcat(spA,spB) would then return type DM because of (possibly dubious) upconversion in our SWIG.

[jaeandersson]

I don't think we should cast Sparsity to DM. Isn't that constructor explicit?