Can you alter a transformation after it has been added?

[BradyAJohnston]

I'm not super familiar with the use of transformations, but once you add them to a universe, can they be altered or removed from that universe?

[orbeckst]

Treat transformations as something that you get to add to a universe a single time.

There is no API at the moment for modifying them in any form.

Part of the reason is that a transformation applied to an in-memory Universe changes the underlying data. There is no safe way to undo the transformation and the apply the new one (or none) except for starting over fresh. In order to avoid the danger of working with corrupted data we don’t support changing transformations.

In order to change this situation we would need to think about safe guards that prevent potential data corruption.

[BradyAJohnston]

ah that's great, I didn't realise you could just grab the filenames from the universe. I think this should mostly cover what I have in mind!

[orbeckst]

yeah, just be aware that we sometimes cheat and filename is not a string, e.g., when we loaded a topology from a buffer/NamedStream (IIRC...)

[BradyAJohnston]

I'm thinking that at least for use in Molecular Nodes, that I might keep around the original untransformed universe, then create copies whenever I want to add / alter a transformation:

u = Universe(TOPOL, TRAJ)

u_smoothed = u.copy()
u_smoothed.trajectory.add_transformation()

I can see this maybe being problematic with particularly large universes though. Will have to do some testing.

[orbeckst]

EDIT: Keeping an unmodified copy sounds like a good idea. I think very few universes are so big that you can't do that.

EDIT: My original suggestion #4477 (reply in thread) to just use filenames may not be such a great ideas and it may better to do the following:

The safer approach is probably to reverse engineer Universe.copy() and the underlying topology.copy() and trajectory.copy() methods — the latter actively patches transformations back in so perhaps writing one's own copy_universe_without_transformations(u) method may be better in the long run (or raise an issue to add a no_transformations kwarg to Universe.copy()...)

[BradyAJohnston]

Ah OK I'll have a look at the copy method as well then. Thanks for all of the details!