Arguments for creates_unique?

[cheerfulstoic]

I just started using ActiveRel.creates_unique and I thought it would always only result in one relationship, but it only does that if all of the properties that I'm giving are equal to another relationship's properties. It would be nice to be able to say which properties it should match on. Then the ActiveRel code can pull all of the rest out into a ON CREATE SET

[subvertallchris]

As a wise man once said... https://www.youtube.com/watch?v=UaU6NQzEGcg

That's really unfortunate.

[subvertallchris]

Is this worth backporting into 5.2.x? It reminds of me that find_or_create issue, where we're changing something that's not technically a bug but it defies expectations.

[cheerfulstoic]

It's a funny one because I don't think it would defy expectations if you know what CREATES UNIQUE is doing. Though if we don't, what are the chances that somebody else will?

Regarding backporting, it's tempting, but I'd also like to have reasons for people to want to update to 6.0 ;)

[subvertallchris]

Yeah, for sure. It makes sense once you know about it...

Since there's so much room for confusion, I think should make the syntax more explicit so you can't accidentally end up with unwanted behavior. In 6.0, maybe we require an argument of nil or :none, :all, on: [keys], or except: [keys]. nil would give the behavior we expected now, the others would do as expected?

I'm going to do a quick spike of this -- already have nil behavior working -- and we can tighten up the options tonight and tomorrow.

[subvertallchris]

That PR is a proposed implementation of the comment above.

[subvertallchris]

Ooh, we can close this.