Should these TimeSeries/ElectricalSeries be :py:class: intersphinx references?

Ditto for below on TimeIntervals / DynamicTable

I would suggest intersphinx to the nwb-schema docs since this is for extensions

Just reading this, it's hard to grasp 'how big' something is to be less than 64 KB

Since most metadata are strings, maybe it would be useful to mention a relative scale to that?

A standard Python string under 64 KB would correspond to ~1300 characters, from

import sys

sys.getsizeof("a")
> 50

64_000 / 50
> 1280

Google also says the average number of characters per word in English is 4.7, so that's ~272 words per string

Mostly relevant to fields like description, or sometimes reference_frame

I like the idea of relating this to characters. Do you know how big characters are within HDF5? Is it the same as Python strings in memory?

Hmm I'll have to investigate that a bit deeper, but most likely not the same as Python... I seem to remember us having some issues related to their byte string types here on the Inspector, but maybe that was for fields that are technically set as datasets, not attributes

Even in Python, if you apply a specific encoding it can change the size quite a bit

sys.getsizeof("a".encode("utf-8"))
> 34

which would put the max closer to ~400 words

Well, from https://docs.h5py.org/en/stable/strings.html#storing-strings and https://docs.h5py.org/en/stable/special.html#h5py.string_dtype and manually inspecting some of these attributes with HDFView

It seems HDF5 stores with UTF-8 encoding, so about ~400 words on average from above

~400 words

words or characters?

words or characters?

words using Google's estimate of average characters per word in English

Or just shy of ~1900 characters if that's easier to communicate