This seems like a really difficult issue to solve in the general case, but I found that solving it for the specific case I had was a tractable problem. I have

This seems like a really difficult issue to solve in the general case, but I found that solving it for the specific case I had was a tractable problem. I have a need to diff the output of a query and compare it to the last time it ran, to do regression testing. Just diffing the resulting CSVs wasn't very useful, because I needed the ability to do things like ignore new columns, and report the exact column that had differences from the previous version. I was able to do that by defining a primary key on which I could outer join the two tables. Missing or new rows would be the ones that didn't join, and then I could do a per-column comparison for each row that did join.