more efficient implementation of runMarkov #1068

[jwaldmann]

No description provided.

[sss-create]

I am not sure how we feel about adding new dependencies. I requested a review from Alex..

[yaxu]

Yes would like to avoid new dependencies if possible

[jwaldmann]

performance of accessing a sequence by index (inside runMarkov):

• current implementation: linked list O(n),
• this PR: vector (new dependency) O(1)
• another option: containers.Data.Map (existing dependency) O(log n)

It feels wrong to have a contiguous range of Int as keys, and not use the most efficient data structure for that.

What exactly is the problem with adding vector? Did tidal have trouble with too much dependencies in the past? Perhaps my implementation could be rewritten to use array, but that's also a fresh dependency, albeit a trivial one (I guess it comes sort-of pre-installed with ghc but I don't know current status.) Or use containers.Data.IntMap as a work-around (which still feels wrong).

On the other hand, one might ask, what's the problem with inefficient runMarkov? Maybe it's not that important - until we want to use a large state space. Do we? Then it's large-scale programmed music, and no longer live coding?