All null integers, regardless of their type length, were decoded as u8. This causes quaint to read null bigint as Int32(None) instead of Int64(None).
Same goes for floats. Not sure it's causing any issue atm since we're not differentiating f32 vs f64 in the rest of the "Prisma stack", but I did it anyway for good measure.
There were existing tests reading nullable integers, but those tests were only trying to convert the ColumnData to a rust primitive type and assert that primitive type. However, because of those FromSql implementations 👇 which enable converting a ColumnData::U8 and ColumnData::I32 to an Option<i64>, tests were passing.
I left those FromSql implementations since I'm not sure what they're used for atm.
Overview
Related to https://github.com/prisma/prisma/issues/12796 and https://github.com/prisma/prisma/issues/12798
u8. This causesquaintto read nullbigintasInt32(None)instead ofInt64(None).f32vsf64in the rest of the "Prisma stack", but I did it anyway for good measure.There were existing tests reading nullable integers, but those tests were only trying to convert the
ColumnDatato a rust primitive type and assert that primitive type. However, because of thoseFromSqlimplementations 👇 which enable converting aColumnData::U8andColumnData::I32to anOption<i64>, tests were passing.I left those
FromSqlimplementations since I'm not sure what they're used for atm.https://github.com/prisma/tiberius/blob/e88859099fc38e61dd01990567a928872efb08ad/src/from_sql.rs#L64
Tests now also ensure that all integers + floats are decoded as the proper
ColumnDatain the case of null values.