improve performance for single-effect uses

[suhailshergill]

MTLs State is very fast and difficult to beat when it comes to single-effect uses. Can this be beaten? Similarly for non-determinism, ContT is difficult to beat.

[samuelpilz]

moved: benchmark report

[suhailshergill]

potential solution:

use metaprogramming. specifically, make Union and Eff data families with special cases for single-effect and pure computations respectively. this would allow us to then use newtypes for these special cases as well as UNPACK for the rest.

would require some effort to test out.

[suhailshergill]

so, we tried using data families for Union, but didn't see a speedup.

• http://okmij.org/ftp/Haskell/extensible/Eff3.hs
• http://okmij.org/ftp/Haskell/extensible/OpenUnion53.hs

perhaps if we also were able to convert Eff into a data family we may see results? unclear at present.

[greydot]

There is now fused-effects package, which seems to be quite performant too. I will probably benchmark it against extensible-effects some time later.