Jon Pretty

Oh yeah... caught in the lie! ;) But I meant that the `T` is inferred from the instance `field`, and that `T`, LUBed with other `T`s, helps to infer a phantom type, while the instance `field` is quickly raised to a `BaseKey`... and everything is fine because of erasure...

Clever or too clever? Also, could it be made more obviously correct?

Not sure. I've been playing around with inferring types through variance for a while. I don't like the use of implicits, especially. And maybe use-site variance (i.e. bounded existentials) might work better...

Alternatively, given(a, b, c).propagate(d) { ... } could be replaced with (a & b & c).propagate(d) { ... } without the @ uv or implicits. WDYT?

That's symbolic, so also not good. Maybe I'll get a chance to play with it today.

Fair warning: you'll need to compile Mercator first, which Pyroclastic depends on. I also discovered that dotc does not infer implicit function types the same as if they're explicit lambdas at the callsite, which is problematic. Best to use scalac until I work that out.

Is https://search.maven.org/artifact/com.propensive/mercator_2.12/0.1.1/jar … not good enough?

Probably not. I don't think that version defines sequence for monads...