PolyModel for ever-more-specific selection information instead of a sparse data model

selection.py

from google.appengine.ext import ndb
from google.appengine.ext.ndb import polymodel


class SelectionBinder(polymodel.PolyModel):
    """
    Base class for selecting content.
    https://developers.google.com/appengine/docs/python/ndb/polymodelclass
    All selections are at the binder level or more specific.
    """
    # the binder in which the selection resides
    binder_id = ndb.StringProperty(indexed=True, required=True)

    # last time the entity was put
    last_modified = ndb.DateTimeProperty(auto_now=True)

    # every selection is for tying some metadata to some content.
    # key.urlsafe() representation of Annotation.key
    # https://developers.google.com/appengine/docs/python/ndb/keyclass#Key_urlsafe
    annotation_key = ndb.StringProperty(indexed=True, required=True)


class SelectionDoc(SelectionBinder):
    """
    Selecting inside of a particular document.
    """
    document_id = ndb.StringProperty(indexed=True, required=True)


class SelectionPage(SelectionDoc)
    page_id = ndb.StringProperty(indexed=True, required=True)


class SelectionRegion(SelectionPage):
    """
    A region can be a paragraph, chart, some object inside of a page.
    """
    region_id = ndb.StringProperty(indexed=True, required=True)


class SelectionText(SelectionRegion):
    """
    When the parent region is a paragraph, the user is selecting some text.
    """
    start_index = ndb.IntegerProperty(indexed=False, required=True)
    end_index = ndb.IntegerProperty(indexed=False, required=True)


class SelectionXY(SelectionRegion):
    """
    Put a point over a region at an X/Y offset from the top-left.
    Only works with static content.
    """
    position_x = ndb.IntegerProperty(indexed=False, required=True)
    position_y = ndb.IntegerProperty(indexed=False, required=True)

sparse.py

from google.appengine.ext import ndb


class SparseModel(ndb.Model):

    def _prepare_for_put(self):
        if self._properties:
            # store delete work for after read is done
            keys_to_remove = []
            for key, value in self._properties.iteritems():
                if value is None:
                    keys_to_remove.append(key)

            # delete properties with None
            for key in keys_to_remove:
                del self._properties[key]

            # finally dig into nested properties
            for prop in self._properties.itervalues():
                prop._prepare_for_put(self)

instead of dropping all None columns from existence on database put via a hack, one can be smart about instantiating entities at the appropriate specificity and reap the rewards of inheritance