Have you considered another mechanism apart from field tags for relating fields to the db (for example a function for marshalling?). It's starting to look like you're creating a dsl stuffed into struct field tags as strings, which gets ugly if it is complex. This is one part of go I'm really not keen on for this reason.
In https://github.com/aodin/sol#sol I separated the table schema, which is a function, from the destination / receiver structs. This approach allows you to build tables programmatically and do condensed selections, such as all IDs into a []int. However, this process still has to match database columns to struct field names for most selections.
There is a table api in qb much like yours. See https://github.com/aacanakin/qb/blob/master/table_test.go. The session api converts tags into table object. I really like the column definitions. Maybe I can convert the definition implementation like you.
I'm also not happy to add tons of tags in struct fields. Can you provide an example on how to do that?
What I understand is to have a function inside the struct namely like "map" and define the constraints there. I think we can do this in the next milestone of qb.
Can you open an issue about this? The github repo is https://github.com/aacanakin/qb
There are also some missing features qb doesn't have and gorm has. For instance, you can define relationships in gorm while in qb, you can only define foreign keys using `qb:"constraints:ref(col, table)"`.
Moreover, I am not entirely sure but I don't think enforcing types is possible in gorm. In qb, consider this struct;
type User struct { id string `qb:"type:uuid; constraints:primary_key"` }
qb understands this as a uuid type although it has string definitions. These are the main reasons why I created qb.
The relationship feature is not clear in my mind. I'd like to have feedback on relationships.