On Wed, Aug 29, 2012 at 10:43 AM, Casey W. Stark
Hey Matt.
I think this would be a big improvement, but I was wondering how it interacts with other yt pieces. Does each output have geometry and coordinate_handler objects as attributes?
Yup, that is the plan. The idea is that we move all of the IO and particle/fluid selection into the geometry handler, and the handling of spatial layout to the coordinate handler. This would mean, for instance, that we could push periodicity as well as path length into the coordinate handler; this would remove some of the need to constantly do wraparound checks and the like. There is still somewhat the issue that *selection* of points to understand coordinate systems, which will require a bit more thought in the future but I think is still a tractable problem.
Is it possible to replace axis_name, axis_id, x_axis, and y_axis with only axis_names = ['x', ...]?
It is, but not with abc.abstractproperty. (Initially I figured the cost of creating the dicts was low enough that we could do this to avoid worrying about mutable, class-level properties, but I think perhaps I like yours better.) I'll remove some of the fancier ABC-stuff and slim it down.
- Casey
On Wed, Aug 29, 2012 at 7:28 AM, Matthew Turk
wrote: Hi all,
I've issued a pull request to the 3.0 repository, as I think it warrants discussion. It's here:
https://bitbucket.org/yt_analysis/yt-3.0/pull-request/5/initial-import-of-co...
This includes a first pass at a coordinate handling system. This is distinct from a geometry handling system; the coordinates here refer to how we handle coordinates and spatial locations internally, whereas geometry refers to how data is distributed throughout a domain and throughout places on disk. For instance, coordinate handling would be cartesian, polar, spherical.
The reason I'm bringing it up for discussion is that I believe we want to move as much coordinate handling and transformation into a separate, well-defined class as possible. Periodicity, distances and so on are all currently scattered throughout the code, and I'd like to try to consolidate them. Additionally, as new coordinate systems (polar, spherical) are added, we'll need clear ways to delegate responsibility for things like "How do I calculate path length as I integrate?" or "What's the way to turn this into an image?" I believe the best way to do that is to attach a coordinate system to the dataset object itself. (We now have a polar pixelizer http://i.imgur.com/a4UGg.png !)
The interface is currently set such that you need to define these methods and properties in order to implement a coordinate system:
coordinate_fields (this may go away, but it's for the analogs of 'x', 'y', 'z', as well as volume) pixelize convert_from_cartesian convert_to_cartesian axis_name axis_id x_axis y_axis period
Some of these currently live in dictionaries in yt/utilities/definitions.py, which is pretty sub-optimal. I'd like to ask for feedback:
1) Do these methods sufficiently cover everything we need to know in yt about a coordinate system? Should any be added? 2) Do we need to directly address dimensionality as a separate subclass? 3) Should any of these be removed?
This will also help address these issues:
https://bitbucket.org/yt_analysis/yt/issue/418/use-a-right-handed-coordinate...
https://bitbucket.org/yt_analysis/yt/issue/422/ray-casting-in-cylindrical-co...
https://bitbucket.org/yt_analysis/yt/issue/421/refactor-non-cartesian-geomet... https://bitbucket.org/yt_analysis/yt/issue/345/non-cartesian-geometry https://bitbucket.org/yt_analysis/yt/issue/205/periodicity
-Matt _______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
_______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org