my databases are all distributed, no FK could work 🙂

my biggest issue with FK-s would be that they only guarantee that whatever they point at exists ... it doesn't mean that they are pointing at the correct thing. so you can still mess up and happily live in an illusion whilst still corrupting data.

I would make a distinction between the concept of a foreign key, an the implementations of foreign keys

just because in my mind the concept of foreign key is pretty synonymous with "pointer" or "reference" or "name", which are useful, and I use all the time

(I think all this discussion has been about the implementation and why folks don't like those constraints in the database itself...?)

At least, that's what I was assuming 🙂

if you put fks in the database, the fk information is(if I recall) available in the result set metadata, which means you can use it for generic database navigation code instead of looking at column names and guessing

Yup, and that's why the (experimental) datafy/`nav` code in java.jdbc lets you pass in a :schema option to override the convention -- which you could auto-generate based on DB metadata. But this is why that's still experimental because there's no one size fits all for how FK concepts are implemented in any given app... ¯\(ツ)/¯