Reuse hash

[liamsi]

After reviewing celestiaorg/celestia-core#351 I've realized that #38 introduced an unnecessary alloc per leaf and inner node.

This PR reverts the changes of #38, caches the root s.t. it only gets computed when necessary but keeps that test that aims to reproduce the issue in #37. (draft as I want to confirm this really does not re-introduce the issue in test withing ll-core, too celestiaorg/celestia-core@1522021)

[codecov]

Codecov Report

Merging #40 (945e12f) into master (a25c390) will increase coverage by 1.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #40      +/-   ##
==========================================
+ Coverage   93.77%   94.78%   +1.01%     
==========================================
  Files           7        7              
  Lines         305      307       +2     
==========================================
+ Hits          286      291       +5     
+ Misses         14       11       -3     
  Partials        5        5              

Impacted Files	Coverage Δ
proof.go	97.05% <ø> (ø)
hasher.go	100.00% <100.00%> (+5.66%)	⬆️
nmt.go	93.10% <100.00%> (+0.14%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a25c390...945e12f. Read the comment docs.

[Wondertan]

It is so cool that you immediately realized how to apply here the change @jsign proposed. Love this!

[liamsi]

Thanks :-) We were already doing the suggested change (of celestiaorg/celestia-core#351) here. Only very recently (#38) we changed this for (#37). But yeah reviewing that code made me realize that our "bugfix" (quotes as the bug only happened during unit tests), wasn't the smartest one and we should have kept the existing code in favour of the fix here.
In general there is plenty of room for optimizations in this library (nmt).

Two low hanging fruits are:

• have one function call that combines Push and Root (takes all leaves and immediately returns the hash)
• get rid of the nebulous labs code for proofs (remove the dependency entirely)

[liamsi]

IMO, the test in #38 is probably enough confirmation that this does not re-introduce the problem pre #38/#37.

[adlerjohn]

Is this branch up-to-date against master? It's hard to tell from the force push.

[liamsi]

Yes, the force push was to rebase on master.