The interaction of runEval, rpar, and rseq is underdocumented

[jberryman]

(IMHO). It has never really been clear to me when I should use rseq, even after reading this chapter: http://chimera.labs.oreilly.com/books/1230000000929/ch02.html#sec_par-rpar-rseq

What I'm lacking I think is an intuitive sense of how or when we "block" waiting for a parallel computation to return. Is it the case that once a spark is "converted" to a thread then evaluation of the sparked thunk will block? At what point exactly do sparks fizzle, and how does this relate to the whims of lazy evaluation? Does -feager-blackholing come into play here? Do data dependencies matter at all in this?

I was motivated to open this after reading https://ghc.haskell.org/trac/ghc/ticket/14330 and realizing I don't have a strong sense of what's going on when doing pure parallelism.