Improve ujson test coverage or clean up dead code

[WillAyd]

I've been spending some time cleaning up the integration with ujson in our source code. After enabling some debugging directives and running the test suite I noticed that there were quite a few places in objToJSON.c which were not being hit. This can point to either dead code or a lack of test coverage, so these can be whittled down.

In my own repo I have a branch of this where each of the items are commented with // not hit in tests which you can see here:

https://github.com/WillAyd/pandas/blob/dead-json/pandas/_libs/src/ujson/python/objToJSON.c

Planning to go through these individually and see what we can do. Extra set of eyes would certainly be welcome

[WillAyd]

@jbrockmendel you might be interested

[TomAugspurger]

I’m hopeful that we can eventually remove all of this and use arrow and it’s JSON support. Given that, it may not be worthwhile spending too much time here.

[jbrockmendel]

@TomAugspurger any idea what "eventually" means in this context? i.e. are we talking months or years?

[TomAugspurger]

https://issues.apache.org/jira/browse/ARROW-694 was the initial blocker, and that just landed in arrow 0.13.0. So, with a good amount of effort, months seems reasonable.

[jbrockmendel]

Extra set of eyes would certainly be welcome

Yah, there's definitely some remove-able stuff in there.

• L65-67 is a check for py<2.5
• L163-167 is a PY3 check
• is the NPY_API_VERSION < 0.7 check on L729 removable?
• i have no idea what the semantics are for L737-753
• L648 when would obj.size be negative?
• is the PyLong_Check on 1785 redundant with the check on L1771?

[WillAyd]

Yep thanks @jbrockmendel . FYI some of the compat stuff I already ripped out in #26211

[WillAyd]

This has been / is being tackled elsewhere