Want json-sort for sorting json objects

[nfitch]

As far as I've searched, I haven't found an app doing external merge sorts of json objects. For sorting json objects in Manta, it seems that someone would need to extract a field in the first pass, sort using the standard unix sort and then remove the field in a final pass. Of course, all this could be done as part of the user's "standard" data processing, but that requires more complexity than most people want/need.

manta-compute-bin may not be the right place for a tool like this, but it's a good place to record the idea. Also, it'd be nice to have these options:

1. Control over what happens to object that don't have the field
2. Filter rows that don't need to be in the final sort

[davepacheco]

I've needed something like this in order to use "diff" with large json objects. It's poorly documented, but it's here if you want to use it: https://github.com/davepacheco/kartlytics/blob/items/tools/json_normalize

It seems like you should be able to do this with "json -e" as well, but I haven't been able to make that work.

[trentm]

Would something first-class in json be useful/desirable here? I've occasionally wanted to be able to sort arrays-of-objects with json -s key or something

[davepacheco]

Potentially, yes. The "json_normalize" tool I linked to sorts all object
keys lexicographically and lets the user specify a field on which to sort
arrays of objects. It lets you specify N of these keys that it uses for
the first N arrays it finds, depth-wise. This approach assumes the objects
basically look alike, though. I wonder if there's a better way to express
it.

[nfitch]

I wouldn't mind if it ends up in the json-tool as long as trentm/json#49 is done before-hand. That said, though, the json tool can't assume that all the json objects will fit in memory- it's going to have to implement some sort of external merge sort (like the unix sort command does today). I wasn't sure that sorting complexity should be put in the json-tool, but it's really up to you, trentm.