Rewrite Parsers

[richardjgowers]

All parsers need to have their output changed to a Topology object. These should only now return values that are defined in the format (no guessing & no default values).

For details on the new structures, see:
https://github.com/MDAnalysis/mdanalysis/wiki/Guide-to-new-Topology-format

It is also possible to add in Attributes for values that were previously ignored

• CRDParser
• DLPolyParser
• DMSParser
• ExtendedPDBParser
• GMSParser
• GROParser
• HoomdXMLParser
• LAMMPSParser
• MOL2Parser
• PDBParser
• PDBQTParser
• PQRParser
• PrimitivePDBParser
• PSFParser
• TOPParser
• TPRParser
• XYZParser

[orbeckst]

Is this really easy for a newcomer or "just" easy for someone in @MDAnalysis/coredevs ?

[richardjgowers]

I think I called it easy because it's (mostly) just filling numpy arrays with stuff from the file, rather than filling Atom objects. So it's just converting the existing parsers really.

[jbarnoud]

What is the status of the tests? Are they still relevant? I am mostly concerned about MDAnalysisTests.topology.test_topology._TestTopology (that is MDAnalysisTests.test_topology._TestTopology in the develop branch). Indeed, I plan to tackle the TPR parser, most tests of which inherit from _TestTopology.

[richardjgowers]

Long story short, the current tests aren't very good because they need to create a Universe. The new Topology object should work as a standalone object, so the new tests for topology won't have to involve a Universe.

I'll try and get a new test template done soon, but it will likely just have to inspect each of the numpy arrays inside each Attribute in the Topology object.

If you want to look at doing TPR, on the issue-363 branch, GROParser is a fairly good example of what needs to be done.

[richardjgowers]

@jbarnoud
Base tests now exist, with an example for GRO

https://github.com/MDAnalysis/mdanalysis/blob/issue-363/testsuite/MDAnalysisTests/topology/base.py

https://github.com/MDAnalysis/mdanalysis/blob/issue-363/testsuite/MDAnalysisTests/topology/test_gro.py

[jbarnoud]

@richardjgowers
Great! I'll see what I can do for the TPR parser then.

Quick remark: it may be good to list the attributes a child of ParserBase must define in the docstring of the class. Of course, one can read the code or copy TestGROParser but it is better when a class has its own doc.

[dotsdl]

@richardjgowers clearly you had a relaxing vacation. :D

[richardjgowers]

Wasn't too bad in the end :D We need to get the benchmarks up really, show everyone why we're even bothering to get this done

[dotsdl]

Do you want help on that front? Did you check if the performance of any of these parsers improved from these changes alone?

[richardjgowers]

Yeah I was going to add the select_atoms benchmarks in too, so if you could
throw the notebook up and maybe add some quick selection benchmarks that'd
be good.

On Sat, 19 Dec 2015 17:57 David Dotson notifications@github.com wrote:

Do you want help on that front? Did you check if the performance of any of
these parsers improved from these changes alone?

—
Reply to this email directly or view it on GitHub
#572 (comment)
.

[richardjgowers]

Oh, and I've not checked individual parsers, but I expect none will be
slower, many will be faster. I did a little refactoring for a few.

On Sun, 20 Dec 2015 08:40 Richard Gowers richardjgowers@gmail.com wrote:

Yeah I was going to add the select_atoms benchmarks in too, so if you
could throw the notebook up and maybe add some quick selection benchmarks
that'd be good.

On Sat, 19 Dec 2015 17:57 David Dotson notifications@github.com wrote:

Do you want help on that front? Did you check if the performance of any
of these parsers improved from these changes alone?

—
Reply to this email directly or view it on GitHub
#572 (comment)
.

[dotsdl]

@richardjgowers I'll throw it up as a gist today. Any selections in particular you think showcase the best the new behavior? Also, on which system?

[dotsdl]

Here's the link: https://gist.github.com/dotsdl/0e0fbd409e3e102d0458

Haven't added selections yet, but can update the notebook later on.