Saving (pickling) Topology between sessions

[richardjgowers]

It would be nice if we could save the current state of the Universe topology so it could be saved and loaded (possible with another trajectory) again.
Eg something like

u.save_topology('out.pkl')

# Then load a new Universe, which instead of reading a Topology file,
# just unpickles an old topology
u2 = mda.Universe('out.pkl', 'other.traj')

The current way for doing this would be to make a .pdb or something and load the new Universe with that, but that puts you at the mercy of the limits of the given format.

It will also have to work with selections of a Universe too:

ag = u.atoms[:100]

ag.save_topology('out.pkl')
with mda.Writer('out.trr', numatoms=len(ag)) as w:
    for ts in u.trajectory:
        w.write(ag)

unew = mda.Universe('out.pkl', 'out.trr')

Thinking about it, this is kind of a step towards Issue #173 which is the pickling of everything. This should be possible because it sidesteps the problems in Issue #173 which is reinitialising the Reader object.

[orbeckst]

1. Why pickle instead of an existing format such as (extended) PSF?
2. Is this meant to be a "stable" way to save all the topology information or more like a cache (similar to the XTC index) so that we don't have to recompute e.g. guessed bonds?

Somewhat related when reading through your use case: Would it make sense to have a Universe.topology object similar to Universe.trajectory and bundle a lot of the topology-related stuff there?

[richardjgowers]

It's just to avoid having to fit in to an existing format and losing information. Eg maybe PSF doesn't do something I want. There should be a 100% way of moving topology info.

Could definitely do a .topology object and try and tidy the distinction between trajectory and topology.

[richardjgowers]

But having a PSF (and all other formats) writer isn't a terrible idea either.

[richardjgowers]

Closed, go see #643 for more coherent ideas on this :D