Add missing MkProducer argument, force compiler errors when they're missing

[bplommer]

I'm not 100% happy with the added boilerplate, or with it being needed, but I don't have a better idea. Any thoughts?

[bplommer]

Exciting update: it doesn't compile with Scala 2.12 anyway. I guess I'll just leave the ambiguous implicits stuff out, but I'm still not happy with that.

[bplommer]

Exciting update no. 2: It now compiles in Scala 2.12, courtesy of added ugliness.

[LMnet]

Could you please explain why do we need this? As far as I understand, without mkAmbig1 and mkAmbig2 default instance of the Mk??? instance will be used (which is derived from the Sync). And what is the problem with that?

[bplommer]

Could you please explain why do we need this?

It's not strictly needed, but it's an extra safeguard against us writing factory methods that don't require an Mk argument.

As far as I understand, without mkAmbig1 and mkAmbig2 default instance of the Mk??? instance will be used (which is derived from the Sync). And what is the problem with that?

There's no problem with the default instance derived from Sync being used, but that derivation needs to happen at the call site in user code so that the user can override the default behaviour if needed.

[LMnet]

Oh, so that's for library maintainers mostly, not for users. Ok, I got it.

[bplommer]

This is intended to catch bugs like the one this PR fixes on lines 642-63 of KafkaProducerConnection.scala, where the stream method delegates to the resource method. Rather than stream taking a MkProducer argument it was autoderiving an instance based on Sync, which would prevent use of a custom instance by user code.

[bplommer]

Oh, so that's for library maintainers mostly, not for users. Ok, I got it.

👍 didn't see your reply before my second comment