ENH: IO interfaces to JSON files

[oesteban]

A grabber/sink pair to read/write parameters from json files.

[coveralls]

Coverage increased (+0.01%) when pulling cb17ee6 on oesteban:enh/JSONFileInterface into ccc4166 on nipy:master.

[satra]

this looks good. could you please update the changes file?

[oesteban]

@satra sorry, I forgot about that. Fixed.

[coveralls]

Coverage decreased (-0.02%) when pulling fe6a054 on oesteban:enh/JSONFileInterface into ccc4166 on nipy:master.

[satra]

how about adding an input that accepts a dictionary?

[oesteban]

Sorry @satra, I don't understand what you mean :S.

[satra]

something like in_dict = Dict(desc='input JSON dictionary')

so that you can pass a dictionary output of another node directly into the jsonsink node? and it will populate the relevant fields.

[oesteban]

Ok I see 👍

[satra]

should we consider:
a. adding the elements using on_trait_change for in_dict, i.e., whenever this is changed, instead of only a single input?
or
b. making it behave more like datasink with the a[.[@]b[.[@]c] formulation to create arbitrary dictionaries?

this would allow connecting multiple nodes to the same JSONsink.

[oesteban]

Hi @satra, I've been thinking this around, but I can't see what would be a good end to this PR.

From your last comment, I don't see why it is not possible to connect multiple nodes to the same JSONSink, provided with you specify the input in input_names.

Regarding your considerations, I'd prefer not to go for b), as I'm not fond of the datasink formulation. I find it particularly cumbersome (of course, this is just one opinion).

Regarding a), you mean that every time in_dict changed, a new json file would be written/updated?.

[satra]

@oesteban

Regarding your considerations, I'd prefer not to go for b), as I'm not fond of the datasink formulation. I find it particularly cumbersome (of course, this is just one opinion).

happy to entertain other thoughts on that. but we will leave it for a separate issue.

Regarding a), you mean that every time in_dict changed, a new json file would be written/updated?.

i meant that this input acts as an accumulator, but i can now see how that wouldn't be great. let's merge this and then consider the datasink formulation (also what an alternative to becoming less cumbersome would be).