fix: remove node buffers

[achingbrain]

Replaces node buffers with browser-friendly Uint8Arrays.

Previously we didn't want to do this because there's no equivalent to node's Buffer.allocUnsafe which doesn't set all entries of allocated buffers to 0. The thing is we only actually use that in one place so to address this I've detected the existence of of Buffer.allocUnsafe before using it, otherwise falling back to new Uint8Array which is what the node Buffer polyfill does anyway.

Running the benchmark suite in this module shows the performance is comparable to or even slightly better than master (I think due to not having to convert Uint8Arrays to Buffers any more though please don't quote me on that 😆 ):

Before:

$ node benchmark.js
Initializing handshake benchmark
Init complete, running benchmark
handshake x 59.95 ops/sec ±11.20% (75 runs sampled)
handshake x 54.68 ops/sec ±10.81% (68 runs sampled)
handshake x 50.42 ops/sec ±11.55% (65 runs sampled)
handshake x 53.41 ops/sec ±11.84% (68 runs sampled)
handshake x 50.25 ops/sec ±11.80% (66 runs sampled)

After:

$ node ./benchmark.js
Initializing handshake benchmark
Init complete, running benchmark
handshake x 61.48 ops/sec ±11.71% (76 runs sampled)
handshake x 59.43 ops/sec ±11.13% (73 runs sampled)
handshake x 56.09 ops/sec ±12.02% (71 runs sampled)
handshake x 60.05 ops/sec ±11.69% (74 runs sampled)
handshake x 59.66 ops/sec ±10.59% (74 runs sampled)

[codecov-commenter]

Codecov Report

Merging #125 (ee01e42) into master (61251ed) will decrease coverage by 0.52%.
The diff coverage is 85.71%.

@@            Coverage Diff             @@
##           master     #125      +/-   ##
==========================================
- Coverage   89.13%   88.60%   -0.53%     
==========================================
  Files          16       16              
  Lines        1831     1826       -5     
  Branches      243      246       +3     
==========================================
- Hits         1632     1618      -14     
- Misses        199      208       +9     

Impacted Files	Coverage Δ
src/noise.ts	76.51% <50.00%> (-0.83%)	⬇️
src/handshakes/abstract-handshake.ts	92.51% <72.72%> (-1.64%)	⬇️
src/handshake-xx-fallback.ts	92.94% <75.00%> (-0.09%)	⬇️
src/encoder.ts	88.75% <76.00%> (-6.78%)	⬇️
src/handshake-ik.ts	92.30% <92.30%> (-0.05%)	⬇️
src/crypto.ts	97.67% <100.00%> (-0.29%)	⬇️
src/errors.ts	100.00% <100.00%> (ø)
src/handshake-xx.ts	97.12% <100.00%> (-0.02%)	⬇️
src/handshakes/ik.ts	89.03% <100.00%> (-0.08%)	⬇️
src/handshakes/xx.ts	90.21% <100.00%> (-0.06%)	⬇️
... and 2 more

---

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 61251ed...ee01e42. Read the comment docs.

[dapplion]

@achingbrain I've been benchmarking the memory usage of Uint8Array vs Buffer and Uint8Arrays take double memory to represent the same data. See benchmark results here https://github.com/ChainSafe/ssz/pull/219/files#diff-afd2490ed095916808da2e4d8c22e4c7ebc2062c020c722574214fdb908ef640

noise shouldn't persist too much data at once but it's an important consideration to be aware of

[achingbrain]

That's interesting. The percentage memory difference between new Uint8Array(...) and Buffer.from(....) seems to decrease as the size of the bytes increases so it may be some sort of fixed overhead, which is all weird as Buffer extends Uint8Array. Maybe worth opening an issue against node itself to get some clarification?

As long as these things are getting garbage collected it shouldn't be a showstopper though?

[wemeetagain]

LGTM

[wemeetagain]

noise shouldn't persist too much data at once but it's an important consideration to be aware of

As long as these things are getting garbage collected it shouldn't be a showstopper though?

👍 Thats my thought on it.

It would be reassuring to eventually find the root cause of the Buffer/Uint8Array discrepancy tho 😅

[dapplion]

@achingbrain Sorry forgot to go back to this

As long as these things are getting garbage collected it shouldn't be a showstopper though?

I did a few heap snapshots and I can't see noise retaining any substantial amount of binary data. So good on this front 👍