donald.anderson, the non-JSON changes for dump are in, can you check on the JSON changes?

Copied from WT-2381:

keith.bostic:

Based on inspection, I think the JSON support for dump/load needs review as well.

First, it looks to me like the JSON dump isn't handling source= configurations in the colgroup/index lookup for tables (at least I don't see anything in dump_json_table_cg, which is where I think it would be).

Second, the test suite has separate JSON dump tests, and LSM isn't included. I suspect the changes Sue LoVerso has been making for the non-JSON dump tests need to be copied into the JSON dump tests.

Third, would it make sense in dump to change the JSON support to be a translation of the non-JSON dump information, that is, always do a non-JSON dump and then translate the "final" schema information into JSON as we're printing it out? I freely admit I don't know much about how JSON fits into dump/load, but currently we have separate JSON and non-JSON paths in dump and load, and if we're doing surgery, it might make sense to refactor so that JSON is a translation layer instead of doing its own interactions with the WT library. I think that's Donald Anderson's code, so he should probably weigh in on if we're missing source= support in there and/or if refactoring makes sense.

michael.cahill:

Lastly, w.r.t. JSON dump: let's get the non-JSON tests passing on a single branch, then add JSON into the test and see what needs to be fixed. We could potentially merge the branch into develop with the JSON tests disabled for now so there is a clean handover to Don in a separate ticket. I like the idea of having JSON share most of its logic with non-JSON dump: my preferred way to do that long-term is to move the logic under the cursor API but it may be expedient to fix the JSON versions of the utility first and make the structural change later.