fs.readFile now completes small files in one thread-pool round trip (
542e2b2)
Node’s readFile path now uses a new ReadFileJob fast path for regular files that fit in one chunk, collapsing open/fstat/read/close into a single libuv trip. The promises API gets the same treatment, which should cut latency and reduce thread-pool contention for common small-file reads.
buffer writes now validate offsets before overflow-prone math (
59c976f)
Buffer string write helpers now coerce and bounds-check offset before using it to derive the remaining length, preventing offset overflow from slipping through. The native slow path was fixed too, so out-of-range offsets now fail cleanly instead of being miscomputed.
sqlite sessions no longer allow reentrant close during changeset generation (
d6bbf57)
session.close() now throws if a changeset or patchset is being generated, preventing a use-after-free when the session is closed from inside SQLite callbacks. The change is documented and covered by a regression test.
FFI float setters now reject non-numbers instead of coercing them (
f3a8f1a)
setFloat32() and setFloat64() now require actual numbers and throw ERR_INVALID_ARG_VALUE for strings, booleans, objects, and other non-number inputs. This brings the setters in line with the stricter validation already used by the integer and argument conversion paths.
FFI fast-pointer handling now rejects direct SharedArrayBuffer inputs (4bd9821)
The optimized pointer conversion path was tightened so direct SharedArrayBuffer values are rejected consistently, matching the generic wrapper’s validation. That avoids optimized/unoptimized behavior drift for pointer arguments.
Streams async pull now surfaces aborts that happen during final flush (c79868f)
The async pipeline iterator now checks for aborts after transform iteration completes, even when a flush resolves to null and yields no final batch. That fixes a case where an abort could previously be swallowed and the pipeline incorrectly complete.
JUnit output no longer breaks on empty diagnostics (
03fb384)
The test runner’s JUnit reporter now handles empty diagnostic output correctly, fixing a regression in generated reports.
module package.json lookup now caches per directory (9c641c5)
Nearest-parent package.json discovery is now memoized by directory instead of by file, eliminating redundant native lookups and repeated deserialization work. This should reduce overhead for module loading in package-heavy trees.
Other misc changes
