improve performance and memory usage in HashUtils.djb2

[skimbrel-figma]

Runtime profiling indicates this method generates several many memory allocations.

Comparing to the JS implementation, we saw the intent of the hash &= hash line was to force the JS runtime to keep the number as a 32-bit integer. This is indeed the correct way to do it in JS, but not in Ruby; as a result, the hash local will grow ever larger, requiring more and more memory since Ruby supports unbounded integers.

Fix: truncate the hash value on each iteration with the same 32-bit 0xFFFFFFF constant used at the end instead.

[skimbrel-figma]

Benchmarking before/after with a 256-char string on my laptop:

irb(main):199:0> puts Benchmark.measure { n.times { djb2_fixed(short_str) }}
  0.003365   0.000043   0.003408 (  0.003415)
=> nil                                                                                       
irb(main):200:0> puts Benchmark.measure { n.times { djb2(short_str) }}
  0.024926   0.001164   0.026090 (  0.026234)
=> nil    

A quick, imprecise comparison by inspecting GC.stat[:total_allocated_objects] before/after also shows 60% reduction in memory allocations.

I spot-checked a handful of input values to ensure the output hash value did not change.

[tore-statsig]

Nice find! Pulling this in to run tests on it now

[skimbrel-figma]

@tore-statsig actually i just realized we can do even better — the only thing inside the each loop is .ord, which is available as each_codepoint! i'll update.

[tore-statsig]

The original version of this is released

https://github.com/statsig-io/ruby-sdk/releases/tag/2.4.2

[skimbrel-figma]

great! we've been running the second commit (switching to each_codepoint) as a monkeypatch for most of a week now and it seems totally fine + reduces memory still further 👍