v3.5.2

Patch Changes

• perf: hash a string-backed source without materializing its buffer (by @alexander-akait in #286)

  RawSource.updateHash and OriginalSource.updateHash called buffer(), which
  encodes the string to UTF-8 and memoizes the result — so every source that was
  hashed kept its content twice, as a string and as a buffer, for as long as the
  source lived. Node hashes a string as UTF-8, the same bytes buffer() produces,
  so the string is hashed directly instead and no digest changes.

  Measured on a webpack build of three.js + lodash-es (1033 modules, 2.71 MB of
  module source): retained arrayBuffers after the build drop from 8.74–8.76 MB
  to 5.81 MB with devtool: false, and from 9.35–12.11 MB to 6.42 MB with
  devtool: "source-map" — disjoint over six and five runs. 1029 Buffer.from
  calls per build are removed, one per module. CPU is unchanged: hashing those
  sources costs 9.23 ms via the buffer against 8.18 ms via the string, which no
  whole-build timing can resolve.