Skip to content

PikaPython v2.0.0-rc5 Release Announcement

Pre-release
Pre-release

Choose a tag to compare

@pikasTech pikasTech released this 02 Sep 04:57
· 6421 commits to master since this release

PikaPython v2.0.0-rc5 improves JSON container serialization memory behavior
while preserving the CPython-aligned module contract and the V2 runtime's
performance and resource-efficiency goals.

New Features and Improvements

  • JSON memory optimization: Reduced temporary allocation peaks during
    container serialization by avoiding an intermediate growing-string copy.
  • Capability accuracy: Native callback binding coverage is now explicitly
    reported as unsupported until its ABI is implemented, preventing unsupported
    tests from being selected as if the feature were available.
  • Release configuration: The JSON project template continues to use the
    Program Image module format for lower startup and resident memory overhead.

Validation

  • The Linux JSON workload with 32 objects remains semantically correct: 384
    output bytes, 32 decoded items, and the expected abcdefgh tail value.
  • V2 Program Image measurements were stable across three runs at a malloc peak
    of 17,437 B, with 15 live blocks and a predicted target heap peak of 28,521 B.
  • Valgrind Memcheck reported zero errors and zero leaks across the repeated
    JSON memory runs.
  • CPython-aligned JSON checks cover dumps, nested containers, loads from text
    and binary inputs, trailing-data rejection, invalid separators, and error
    recovery.

Known Issues

  • PikaPython v2 remains a release candidate; APIs and project configuration may
    still change before 2.0.0.
  • Thread-backed concurrency and lock APIs are not included. eventloop
    provides cooperative scheduling without thread support.
  • CPython compatibility covers the documented embedded subset and bundled
    module contracts, not the complete CPython standard library.
  • The cross-version V1/V2 memory comparison uses different default JSON output
    formatting and allocator sampling boundaries; its results are comparative,
    not a strict byte-identical A/B benchmark.