Discussion: unique names for elements, pro and cons

[LucaMarconato]

I propose to enforce unique names among the elements of one SpatailData object. Here I explore the implications.

Cons:

• The user has less freedom since different types can't have the same name. Still, when merging/concatenating different SpatialData objects (say 1 sdata = 1 sample), one could have exactly the same name for the same element types (like all the Labels are called cells), so the user is already limited in this regard.

Neutral:

• The file storage can be kept as it is (because we can infer types when we read), or we can choose to make it completely flat.

Pro:

• In transformations, see discussion in the comment below this one, we can have function calls like this

points = sd.transform(sdata, element_name='cells', tgt="tgt_space")

instead of this

points = sd.transform(sdata, element_type='labels', element_name='cells', tgt="tgt_space")

• (big pro): the tables, the column "instance_key" can be a simple string like "cell_segmentation", now it needs to be "/labels/cell_segmentation" and this feels super weird.

[LucaMarconato]

If we go for unique names we could implement the set_transfomation() and remove_transformation() functions with a signature like sdata.set_transformation(self, Union[str, SpatialElement], ...etc), where the string is the name of the element.

EDIT: these functions have different signature now.

[LucaMarconato]

Another argument in favor for unique names is that napari layers must have unique names.

[LucaMarconato]

Unique names have been implemented. The function sd.transform hasn't been modified, but if implementing this signature or not:

points = sd.transform(sdata, element_name='cells', tgt="tgt_space")

should be done eventually in another issue.