Repository navigation
8.0.0
Upstream improvements have improved number-heavy benchmarks by about 4.2%.
When re-using parsers, number-heavy benchmarks improved by over 8%. Direct
buffer exports have improved by 8.4%.
Tail slices on a simdjson.Array proxy are now 1,130x faster.
- Add
parse_many()for in-memory NDJSON andload_many()for bounded-memory
NDJSON reads. (#71) - Keep proxies and iterators valid across parser reuse and destruction (#125).
Proxy.parserattributes no longer keep parsers alive and returnNone
after the originating parser is destroyed. - Preserve embedded NUL characters in object keys and JSON pointers.
- Make array slicing traverse the source once, including reverse slices.
- Require integer array indices and raise
IndexErrorfor out-of-range integers. - Document duplicate-key behavior and fix the unsupported-implementation error message.
- Update bundled simdjson to 5.1.1 using the upstream singleheader release.
- Map simdjson's
OUT_OF_CAPACITYerror toMemoryError. - Correct extension ABI configuration and enable the fallback implementation in all builds (#136).
- Export
simdjson.VERSION, define public exports, and correct typing stubs (#136). - Support general path-like objects in
Parser.load()and map additional parsing errors toValueError(#136). - Restore stdlib shim test collection and update package metadata and development instructions (#136).
- Drop Python 3.9 and 3.10 support; add Python 3.14 and 3.15 wheel builds.
- Update CI build and documentation jobs to Python 3.15.
- Update all GitHub Actions to their latest stable releases.
- Build and test Windows ARM64 wheels for Python 3.11–3.15 on native runners (#141).