Taal: AMR callback should "notify" "all" parts of the code when the mesh has changed

[sloede]

I think now that you are already optimizing the dt calculation on have_constant_speed, we should probably skip calculating dt in each time step for those cases and only calculate it once after each mesh adaptation and store it in some cache.

Originally posted by @sloede in #200 (comment)

That's a good suggestion but requires additional logic. For example, the AMR callback and the stepsize callback need to know each other to do that. As you may have seen in other TODO notes, I was thinking about letting the AMR callback notify everything when something has been adapted. If we implement that, we can also add this optimization. Could you please open an issue for that?

Originally posted by @ranocha in #200 (comment)

[ranocha]

What about introducing appropriate methods of resize! to do this? That's similar to the approach of the DiffEq ecosystem and seems to fit well, I think.

[sloede]

Which resize! are we talking about? The one for u_ode?

[ranocha]

No, something like resize!(semi, new_size) and resize!(callback, new_size) which propagate to the appropriate caches.

[sloede]

Hm, so without any arguments for a new size, just as an "something has changed" indicator? In that case, what about using something more descriptive such as mesh_has_changed!? I am not 100% convinced on the name resize!, since technically we do not need to resize anything if we coarsen once and refine once while we still need to update our internal data structures.

[ranocha]

I'm sorry, I forgot the new_size in the first version above. I would like to keep that, of course!