btw /benchmark is outdated and i was not able to run it using PS@0.11

This may already be in the works or may have already been decided; however, I am wondering if it would make sense to add a performance comparison to test the note in #83 that the implementation in this PR might be more efficient.

@ethul I would need to update /benchmark as it's outdated. will try

so:

• /benchmark should be update to ps@0.11
• current Trampoline implementation should be copied there (it should uses current Free)
• update Trampoline benchmarks to use all three implementations (old, current and next)

Is the Lazy/thunk even necessary? Why isn't the spine of Free Identity enough to act as a trampoline?

Example: http://try.purescript.org/?backend=core&gist=f89133003a49095821f592f0dc2b8c16

FWIW, I'm not sure that Free Lazy memoizes in a very interesting way. If you defer under a monadic bind, it won't share anything because it creates a new Lazy every time you apply the bind.

It's useful to form cyclic structures, and I think some could be generated purely using Lazy which would then share some results. Things like finding comonadic fixed points jump to mind. That might be useful, but whether it's useful enough to add the performance overhead for the simple case, I don't know.

What if we just remove it as there are already 3 ways to get trampoline? (Lazy, (->) Unit, Identity)

Uh, sorry, I was thinking of Cofree, not Trampoline.

What if we just remove it as there are already 3 ways to get trampoline? (Lazy, (->) Unit, Identity)

?