Periodic/Nonperiodic keywords to selections

[richardjgowers]

So currently we have some geometric selections which allow you to select atoms based on absolute and relative positions. These can work with and without periodic boundary conditions, both are useful at different times.

The use of PBC is currently controlled by flags, which I think we're moving towards removing (#315) so instead a keyword in the selection could be used.

u.select_atoms('A around B distance C selection D')

u.select_atoms('A point B x y z C distance D')

So with the above selections, there are positions A -> D where we could put the keyword.

Position A would be tricky as the keyword precedes the token, but it could be done. ('periodic around 7.0 protein')

D is messy for the around case as generally anything after C gets made into another selection.

I think C reads best.. ('around 7.0 periodic protein')

So tl;dr

• add periodic keyword?
• what should the default be? (probably same as currently?)
• where to put the periodic keyword?

[kain88-de]

@richardjgowers I'm sorry I think something when wrong with your formating. Are A .. D supposed to be options to choose from or atoms?

[kain88-de]

I might replace periodic with pbc

[richardjgowers]

They're places in the selection string we could insert the keyword, so options

[jbarnoud]

I do not have a strong opinion on the question, but I'll go with D. It makes the keyword easier to omit, and existing code will not look odd when compared to new one.

@kain88-de Why would pbc be better than periodic?

What should be the opposite keyword? noperiodic (or nopbc)?

[richardjgowers]

@jbarnoud the problem with D is that the periodic keyword can end up a long way away (as selection can be arbitrarily long).. so

'around 7.0 protein periodic'  # this reads well!

'around 7.0 not resid 7 or name OH periodic'  # this is getting worse?

And then the implementation problem is usually those selections read the distance (7.0 here) then pass all following tokens to be parsed again as the selection. So we'd need some sort of way of passing back of the periodic keyword, not impossible though.

nonperiodic / nopbc is the opposite

[jbarnoud]

@richardjgowers I see your point. I do not like A because I much prefer the selection keyword to be first (much easier to parse, even for a human). I think the periodic keyword applies to the distance (or the coordinate) so C would make sense.

[mnmelo]

I for one think D is ok. It might also be better to have it as a separate kwarg instead:

u.select_atoms('A around B distance C selection', pbc=True)

One other aspect is that one might want different PBC treatment for different subselections in the expression. C would be best for that, as we can have one per subselection.

[richardjgowers]

Ahhh, I didn't even think of a kwarg, maybe that's even better. I'm assuming that it doesn't mess with the current usage unless you have a group called pbc?

[mnmelo]

Exactly. I think it's ok to have a couple of reserved words for this and other purposes (see @jbarnoud's comment in #175).

[jbarnoud]

@mnmelo Are we sure there is no usecase where we would like to mix periodic and non periodic selection keywords?

What about around 7.0 periodic of protein and not around 10.0 nonperiodic of resname POPC? Why would we want to prevent that?

[mnmelo]

Yea, if we want to keep that flexibility, then C is definitely the way to go.

[orbeckst]

I like the kwarg solution best for clarity and I think it covers the overwhelming majority of use cases; in fact, you could argue that you are more likely to get "bad" selections by inadvertently mixing periodic and non-periodic selections. It does occur a flexibility penalty but I can't think of any use cases where you'd need to mix. I am pretty sure you could get a mixed selection by using two calls to select_atoms() and groups.

I also favor the periodic keyword (more explicit than pbc).

edit: Clarified "two groups = two different calls to select_atoms().