Implicit or in selections (Issue #345)

[richardjgowers]

Allows selections such as:

u.select_atoms('name C* N* O*')
# == u.select_atoms('name C or name N')

u.select_atoms('name N and resname GLY LEU')
# == 'name N and (resname GLY or resname LEU)'

u.select_atoms('resid 1:10 14 15')

u.select_atoms('resid 1:100 200-300 and not resname MET GLY')

Defines how to detect keywords in selection strings with is_keyword (Issue #347).

I'm not sure we agreed we wanted these, but it turned out to be pretty easy in the end.

[kain88-de]

Would that screw with us trying to formalize the selection syntax?

[richardjgowers]

Formalise as in move to pyparsing?

[kain88-de]

Yes. Pyparsing would require us to write a formal selection grammar if I remember correctly.

[richardjgowers]

We've technically already got a language, but yeah it would be another feature we'd have to implement if we moved

[dotsdl]

I do wonder if pyparsing is necessary at this point? @richardjgowers has done a lot of work fixing up the selection machinery, and I wonder what we would specifically gain from moving to pyparsing? Also, what would we have to twist pyparsing's arm to do?

[richardjgowers]

A lot of the work I did wouldn't be replaced by pyparsing, it would just replace this class:

mdanalysis/package/MDAnalysis/core/Selection.py

Line 857 in 554d276

class SelectionParser(object):

But all the Selection objects above would probably be 90% the same.

That said, I had a play with it yesterday and didn't really get it immediately? So my current state of mind is, if it ain't broke don't fix it.

[dotsdl]

Unless there's a compelling reason to switch, I say we stick with what we have. As long as new selection features don't have weird consequences/inconsistencies, I'm good with them.

@kain88-de do you know anything we're missing out on without pyparsing?

[kain88-de]

Nope I don't know much about pyparsing. It was just in the back of my mind that we wanted to do that I think @jbarnoud or @mnmelo showed interest in that.

If you are happy with the current parsing approach I'm fine with it. I'd just thought I ask since I don't know anything about pyparsing.

[kain88-de]

Otherwise the idea of @richardjgowers is nice.

[richardjgowers]

Ok, so range selections now work with implicit OR, (see resid example in top comment)

Also some overdue docstring love

[dotsdl]

Very nice work here! I'll go through it in detail.

[orbeckst]

Technically, this can't go into 0.13.1 --- it's a new feature so we have to increase the minor version and hence it must be 0.14.0. Ultimately, only the release manager @richardjgowers has to remember to make the "0.13 bugfixes" milestone actually a 0.14.0 release (and the #363 topology-rewrite will then become 0.15.0).

Nice work here. VMD has these range selections and I find them very nice.

[dotsdl]

I just realized how the new topology system might make documenting the selection language right here difficult. Completely fine for this PR, but we'll have to think about this. I imagine it might be as simple as saying "the token is the singular attribute name, followed by values to filter on."

[richardjgowers]

Unless we hack __doc__ on the fly too?

[dotsdl]

Yeah, we'll have to figure something out. Not sure yet.

[dotsdl]

Looking really good. A few questions to resolve and then we'll merge.

[richardjgowers]

Cool, should be good to go