Use pprint to represent DataContainers, not forcing HDFStubs

[niklassiemer]

to_builtin got a new argument load_stubs=True such that loading HDFStubs can be avoided.

Might close #1031

[niklassiemer]

With these changes, the DataContainer representation is better to read on the command line and in notebooks - at least in my opinion. However, it also slightly changes for subgroups, which are not shown as DataContainer any more:
For the DataContainer

dc = DataContainer({"a": 1, "b": 2, 'c':{'A':1}})

we get with the current main

DataContainer({'a': 1, 'b': 2, 'c': DataContainer({'A': 1})})

but with these changes we would get

DataContainer({"a": 1, "b": 2, 'c':{'A':1}})

Furthermore, it (currently) omits everything nested deeper than the first two levels by '...'.

[pmrv]

load_stubs should have a short documentation even though it's fairly obvious. If I see correctly _no_str_keys is not used.

[niklassiemer]

Yes, documentation is missing and I forgot to pass the _no_str_keys to the sub groups. However, it is used and switches off the key conversion such that the keys stay e.g. int if they are.

[pmrv]

With these changes, the DataContainer representation is better to read on the command line and in notebooks - at least in my opinion. However, it also slightly changes for subgroups, which are not shown as DataContainer any more: For the DataContainer

dc = DataContainer({"a": 1, "b": 2, 'c':{'A':1}})

we get with the current main

DataContainer({'a': 1, 'b': 2, 'c': DataContainer({'A': 1})})

but with these changes we would get

DataContainer({"a": 1, "b": 2, 'c':{'A':1}})

Furthermore, it (currently) omits everything nested deeper than the first two levels by '...'.

I have a slight preference to keep the class name in the output to differentiate true dict members from DataContainer sub groups or their subclasses (e.g. Group from Sphinx). Omitting after a certain depth is ok with me.

[niklassiemer]

I agree, however, the pprint does not naively work on the DataContainer with depth truncation. Therefore, I use the to_builtin under the hood, which of course converts to plain dict... I'll take another shot at it. Maybe I find a way to tell pprint to use the DataContainer directly.

[niklassiemer]

Also a bit the question: Do we want to not load the stubs in this case? Also the nice jupyter lab interface would not show the full container from the beginning but the stubs. To have the full view one would need to call to_builtin beforehand.

[pmrv]

I found this and this, which suggests to subclass the PrettyPrinter. Your call if that's worth it. :')

Not forcing stubs is prefered, I think, since it might incur significant I/O.

[pmrv]

See also the second answer from the SO link and this POTW. Those manage without subclassing, by doing __repr__ in certain way, but haven't looked at the details.

[pmrv]

Last link I promise..

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.