Improve necessary_attributes and recommended_attributes format

[toddrjen]

Currently, the attribute list used to check attribute types uses the following format:

(('time', pq.Quantity, 1),
('signal', numpy, 3),
('datetime', datetime))

Samuel proposed the following format

[{'name' : 'time', 'type' : 'quantities', 'ndim' : 1 },
{'name' : 'signal', 'type' : 'numpy', 'ndim' : 3 },
{'name' : 'rec_datetime', 'type' : 'datetime', },]

I would propose the following alternative:

{'time': {'type': pq.Quantity, 'ndim': 1},
'signal': {'type': pq.Quantity, 'ndim': 3},
'rec_datetime': {'type': datetime}}

[samuelgarcia]

What about evrything in str ?

{'time': {'type': 'pq.Quantity', 'ndim': 1},
'signal': {'type': 'pq.Quantity', 'ndim': 3},
'rec_datetime': {'type': 'datetime'}}

[toddrjen]

I don't as much like the string approach. This is because it is very easy to get a string from a class (just .name), but essentially impossible to get a class from a string. It would require IO classes to hard-code a dict of name/class relationships, which would make it much harder to add additional data types down the road.

[samuelgarcia]

OK not problem.

Your formalism is fine to me.

[rproepp]

+1 for dict of dicts and storing the actual types

[toddrjen]

There is one disadvantage to this approach: dicts are mutable. This increases the risk of accidentally changing the values of the dict. Doing so would change the value for all instances of the class.

As an alternative, it would be easy to create a property that would return a dict from the existing immutable tuple structure.

[samuelgarcia]

Yes but I like the idea that a end user can add at class level some attributes in that dict for a particular use case (some kind of neo extension). We a struct with such code would look ugly. We are between adults...

[toddrjen]

So are we agreed on this approach?

[samuelgarcia]

And what about adding 'require' key in the dict:

[{'name' : 'time', 'type' : 'quantities', 'ndim' : 1, 'require' : False },
{'name' : 'signal', 'type' : 'numpy', 'ndim' : 3 , 'require' : True },
{'name' : 'rec_datetime', 'type' : 'datetime', 'require' : False },]

[toddrjen]

That would be possible. However, I think it would require more complicated and slower code to analyze and wouldn't add any clarity or simplicity compared to the existing pattern. Right now it is obvious at a glance what is required and what isn't, while in that version it would take careful reading. It is okay in that example since all the "require" keys are roughly in a line, but in practice they will be all over the place, which I think will make it much harder to read.

[samuelgarcia]

outdated