tyrelr

@leshow > Again, I would like to PR this, just point me in the right direction. Unfortunately I can't dig into the root cause right now. But if you want...

Thanks for the update, @leshow. I gave this a look this afternoon and was able to construct an update query with incorrect nullibility. I believe I found the root-cause and...

Yep, I just double checked that after todays rebase to get a fix for an unrelated CI failure. Thanks for the good reproduction steps. Those make life so much easier.

Thanks for the schema & query, I've added a [partial] fix into PR #1960 fixing a smaller reproducer. That fix doesn't seem to work for your query though, so there...

I know next to nothing about the other sqlx database backends. But sqlite does have some tricky calculations/trade-offs to do when it attempts to do type checking. Long story short......

That's interesting, @yzernik. Nothing on the macro side inspects actual data rows within the tables, as far as I'm aware. Sqlx retrieves the declared table column types, and the query...

Overall, I think this is a good approach and it worked when I tested it. It is a bit unfortunate to need a 3rd way of storing column-states. But there...

In practice, I expect that a different result would never actually happen. It requires a very atypical query plan for the issue above to occur, I expect sqlite would never...

This should fix #1923, but I haven't yet checked against the original reproduction steps. Left as draft for now, as there are probably additional 'branching operations' which could select branch...

Yep, it fixes #1923 (which impacts the calculated nullability).