Add benchmarks to Universum

[chshersh]

I suppose some functions should be benchmarked in order to investigate their performance and to see that we don't have some drawbacks.

First canditate: it may be good to add hashNub function near ordNub. Because it may be more efficient.

[chshersh]

This benchmark shows that sometimes it make more sense to use hashNub function instead or implement ordNub in a different way to make more profit: https:/ChShersh/effective-haskell/tree/master/nub

[chshersh]

We must consider reading carefully all pragmas and fixing (adding) them appropriately according to this tutorial: http://mpickering.github.io/posts/2017-03-20-inlining-and-specialisation.html

[vrom911]

@chshersh I would like to try myself in this one. But before that I would like to clarify what benchmarks need to be implemented additionally to hashNub and ordNub?

[chshersh]

@vrom911 Thanks for help with this issue! Currently I see only next set of benchmarks:

Leaving only unique elements in the list:

1. Stable nubbing: nub ordNub hashNub
2. Unstable nubbing: Set.toList . Set.fromList map head . group . sort

If I come up with more suggestions or anyone wants to benchmark something else, this list will be extended.

[Martoon-00]

How about:
3. Check whether (...) usage spoils performance when applied to many-arguments function, comparing to plain ways of function application
4. concatMapM vs concatMap when used monad is Identity - (?)

[vrom911]

@Martoon-00 , thank for your suggestions!
I have a question about (...) operator. I don't see a good way to benchmark it, how do you think it could be done? Will we observe difference on some small examples, and how to come to bigger examples for that?

[Martoon-00]

Well, nice question )
I suppose, that if no difference is observed on very small examples, you have two ways of extension: apply resulting function to more arguments and apply (...) more times. If, say, 10 usages of (...) x application to 10 arguments doesn't increase time much (time stays within measurement error, my hope in that it will be enough), then we don't care in practice about possible further performance penult.

Does it sound reasonable?

[vrom911]

Fair enough. Thank you! :) Will try to implement that.