Embed vs Final

[greydot]

Hello!

With the recent removal of withLowerToIO and resourceToIO, I've been left wondering. Is it correct to assume that polysemy is moving in the direction of removing Embed effect completely? As of right now, many things become impossible in polysemy 1.9 as opposed to version 1.7.

[KingoftheHomeless]

Embed will stay -- it's less powerful than Final, but that makes it more flexible to interpret. Compare runEmbedded to finalToFinal.

withLowerToIO, resourceToIO, and asyncToIO were based on a horrific, horrific hack that abused thread forking and communication via MVars in order to work, and often didn't work. With respect to Sandy, withLowerToIO was an abomination that should never have existed.

As of right now, many things become impossible in polysemy 1.9 as opposed to version 1.7.

What things? Final's the safe, sane replacement of the old usages of withLowerToIO. Why can't you use resourceToIOFinal instead of resourceToIO, or withWeavingToFinal/withStrategicToFinal instead of withLowerToIO? Sure, on the surface they're more difficult to use and less flexible than withLowerToIO, but the flexibility and simplicity of withLowerToIO was a lie -- withLowerToIO is horribly unsafe and unsound, and the library ever offering it was a mistake.

Admittedly, withStrategicToFinal's interface is pretty bad. In the WIP core rewrite we're gonna replace its current interface with an interface that's closer to MonadBaseControl in use.

[greydot]

Final's the safe, sane replacement of the old usages of withLowerToIO. Why can't you use resourceToIOFinal instead of resourceToIO, or withWeavingToFinal/withStrategicToFinal instead of withLowerToIO? Sure, on the surface they're more difficult to use and less flexible than withLowerToIO, but the flexibility and simplicity of withLowerToIO was a lie -- withLowerToIO is horribly unsafe and unsound, and the library ever offering it was a mistake.

Mostly, the lack of (easy to find) examples. Another problem is the need to combine Embed and Final in the same effects stack in the presence of third-party libraries, although this can be mitigated with embedToFinal, I'd still to only have to deal with one of them.

Admittedly, withStrategicToFinal's interface is pretty bad. In the WIP core rewrite we're gonna replace its current interface with an interface that's closer to MonadBaseControl in use.

I would love to see that.