Question: Naming of `devices` property in the `PoseEstimation` python class

[ehennestad]

I am trying to understand the rule behind naming of properties that originate from links in the schema.

In the PoseEstimation class, there are two unnamed links, for Skeleton and Device types. In the nwbfields list in the PoseEstimation class, skeleton is singular and devices are plural. Are these names based on the quantity key for each of the properties in the schema? I.e Skeleton links can be 0 or 1, while Device links can be 0 or many.

ndx-pose/spec/ndx-pose.extensions.yaml

Lines 126 to 132 in d2ffeba

	links:
	- target_type: Skeleton
	doc: Layout of body part locations and connections.
	quantity: '?'
	- target_type: Device
	doc: Cameras used to record the videos.
	quantity: '*'

ndx-pose/src/pynwb/ndx_pose/pose.py

Lines 111 to 122 in d2ffeba

	__nwbfields__ = (
	"description",
	"original_videos",
	"labeled_videos",
	"dimensions",
	"devices",
	"scorer",
	"source_software",
	"source_software_version",
	"nodes",
	"edges",
	"skeleton", # <-- this is a link to a Skeleton object

[ehennestad]

@rly Are you able to give a quick confirm for the question above?

[rly]

Yes, the name of the Python attribute is based on the quantity field in the spec. HDMF automatically/automagically maps the singular "Skeleton" link to the Python attribute "skeleton" in the PoseEstimation class, as well as to the constructor keyword argument "skeleton" of this class. Similarly, for "devices". This mapping can be customized in a ObjectMapper registered to the PoseEstimation class, as in here: https://github.com/rly/ndx-pose/blob/d2ffebab277253858a1d9259e71e768b2c49387e/src/pynwb/ndx_pose/io/pose.py . For example, by default, HDMF maps the "version" attribute spec on the "source_software" dataset spec to the PoseEstimation class attribute and constructor argument named "source_software__version" with two underscores before "version". This feels inelegant, so I have the PoseEstimationMap ObjectMapper remap this spec to "source_software_version" with one underscore instead.

[ehennestad]

A few more questions/comments on this:

• For file export the property naming is not relevant, as the property name is not used as a name in the file. Correct?
• The "pluralization" rule in HDMF is quite simple and will not always be grammatically correct. I can imagine an author of an extension will use an ObjectMapper to work around this, and ultimately, property names are not guaranteed to be consistent between MatNWB and PyNWB.

What is more valuable? Plural property names or cross-language consistency? The cross-section of such plural properties and the number of people who use both PyNWB and MatNWB is probably very small, just wanted to raise the question.