Changelog

nodejs/node · · 15 commits

Buffer UTF-8 writes get faster on two-byte strings

Node speeds up UTF-8 encoding and decoding paths, plus a SQLite statement cache refactor and several docs/test updates.

codebytere soreavis RafaelGSS Cherry cookesan edemaine leah-1ee chasetonco inoway46 geeksilva97 panva nodejs-github-bot trivikr

Buffer UTF-8 writes now use simdutf for two-byte strings (codebytere55e4ca3)

Node’s Buffer write path now uses simdutf for well-formed two-byte UTF-16 strings instead of routing everything through V8’s slower UTF-8 writer. That should speed up Buffer.from(), buf.write(), fs.write*() with string data, and stream writes when the input is guaranteed to fit.

StringDecoder UTF-8 decoding now shares StringBytes::Encode (codebyteree19c13b)

string_decoder now uses the same simdutf-backed conversion logic as Buffer#toString('utf8'), which should bring ASCII, Latin-1, and UTF-16 inputs onto the faster path. Large invalid inputs still preserve the old ERR_STRING_TOO_LONG behavior and replacement-character handling.

SQLite statements reuse cached column names (geeksilva97cf30b2e)

statement.all() and statement.get() now reuse cached, internalized column names instead of re-interning them on every call. This trims repeated work in the sqlite binding and should help statement-heavy workloads.

Other misc changes

  • Docs clarified fs.Dirent type detection, copyFile symlink behavior, Windows setRawMode write access, per-architecture FFI argument limits, and missing return types in fs.md/buffer.md.
  • Fixed man-page generation on Windows and updated the security release prep command.
  • Bumped undici in /tools/doc and updated vendored googletest.
  • Adjusted tests for WPT backend checks and the event loop delay timer race.