Open ADBond opened 3 weeks ago
Took a quick look and haven't got anywhere but this is probably useful when we get more time to have a proper look
Iirc the issue is basically this:
I think the actual issue should be relatively easy to sort out, but I wonder if a wider change might be warranted. Maybe changing how (or if) caching works when we're in debug mode
One thing (probably separate from that specific issue) is that I think we should be outputting the intermediate tables with a unique name (e.g. with asci_uuid), and then passing them into the next step. Doesn't feel very robust to physically instantitiate tables with very generic names to the db.
If we turn on
debug_mode
, u-training breaks, but only if we have already runestimate_probability_two_random_records_match
. We getBinder Error: Table "l" does not have a column named "__splink_salt"
This is to do with us materialising a table when salting wasn't required, but then trying to access it when salting is required.
I have a weird sense of déjà vu with this - maybe came across it another time in the last few weeks? But couldn't find any other issue, so opening one here.