I am yet again bumping into the limits of extract_table_column_map
as it does not use a real parser and does not recognize generated
columns, which causes issues when I use that column list to enforce
It probably would be better to move even farther away from extracting
columns and just asking the database instead.
Anyway we get slightly improved ergonomics of insert_filler at the cost
of not being able to enforce all columns within this function.
For a while I was attracted to string IDs because of the notion
that "if you're not going to perform arithmetic, it shouldn't be
an int". But ints make way better use of the allotted bits and
I'm loving getrandbits for urandom ID generation.
This change has really made it easier to reason about database
transactions in my projects. When you use 'with db.transaction'
you know for sure that either the db will commit or rollback at
the end and you won't leave in a dirty state. And it will lock
out all other writers so nothing gets messed up.
Previously I was conflating atomicity of each function with
committing of the entire transaction, and that was causing me
grief. I think this is closer to correct.