Repository navigation
Releases
v0.2.4 — bounded JSON inputs
JSON inputs now have a declared limit of 64 nested arrays or objects, checked before decoding. This fixes Python 3.14 acceptance of overdeep build receipts, ABI receipts and worker results while preserving duplicate-key, finite-number and encoding controls.
Validation: 91 tests per native platform on Linux x64, macOS arm64 and Windows x64, with one POSIX test skipped on Windows. Passive JSON tests pass on Python 3.10–3.14. The wheel was verified after installation. Report, adapter and native-validation schemas remain unchanged; earlier releases and reports are preserved.
v0.2.3
Repair macOS process-group cleanup during worker exit. A denied signal is accepted only after exit is confirmed within the existing five-second limit; an expired wait preserves the signal error.
Native CI passed on Linux x64, macOS arm64 and Windows x64: 79 tests per platform, with one POSIX test skipped on Windows. Packet and native-validation schemas remain unchanged. The pinned OpenIBR controls retain 24 cases per library, 21 unchanged cases and three NaN corrections. The attached wheel was verified after installation.
This release includes the complete worker-evidence checks from v0.2.2. Earlier releases and reports are preserved.
v0.2.2
Successful worker packets now require complete metadata, requested calls and guarded-memory observations. Incomplete or contradictory evidence is rejected before aggregation. Report schema remains inverter-interface-report/2.
Native CI passed on Linux x64, macOS arm64 and Windows x64 with Python 3.12: 75 tests per platform, with one POSIX test skipped on Windows. The pinned OpenIBR controls retain 24 cases per library, 21 unchanged cases and three NaN corrections. The attached wheel was verified after installation.
v0.2.1: JSON input integrity
Reject duplicate object keys, nonfinite numbers and decoder nesting failures in adapters, supplied receipts and worker packets. Build-receipt digests now cover the exact bytes decoded.
Validation: 58 tests on Linux x64, macOS ARM64 and Windows x64 (one POSIX control skipped on Windows). Native header probes and pinned OpenIBR original/fixed parameter controls pass. The installed wheel passes the JSON controls and 642 native software-fixture checks on macOS.
Native validation. Adapter/1 and report/2 are preserved.
0.2.0
Native C/ctypes header-layout checks, guarded Instance storage, return-associated
diagnostic reads and explicit adapter policies. Reports use schema 2; migration
notes are in the README. OpenIBR member/array signal labels remain supported.
Native CI passed Linux x64, macOS ARM64 and Windows x64. Each fixed OpenIBR
library passed 24 cases and 600 checks; each original retained exactly three
NaN failures. The other 21 cases and raw metadata matched. Forty-six software
tests passed on Linux/macOS; Windows passed 45 and skipped one POSIX-only test.
Only Model_GetInfo and Model_CheckParameters are called. These results assess
software interface behavior, without electrical or simulator qualification.
Native CI.
The attached validation summary and installable wheel retain the release identity.
v0.1.0
Initial CLI for native inverter metadata and parameter validation, with declarative adapters for OpenIBR’s public IEEE/CIGRE 1.1 header layout. Calls only Model_GetInfo and Model_CheckParameters in bounded workers. Independent fixtures cover invalid metadata, parameter rejection, caller-buffer writes, crashes, timeouts and output limits.
Native CI passes on Ubuntu x64, macOS ARM64 and Windows x64. The OpenIBR libraries are built with Clang. On each target, the pinned original OpenIBR library reproduces exactly three NaN failures; the corrected library passes 24 cases and 561 checks with identical metadata. Unix runs 15 passing fixture tests; Windows runs 14 with the Unix process-group test skipped.
Scope is metadata and parameter validation. Initialization, electrical stepping, simulator integration and formal compliance are outside this release. MinGW-w64 has not been tested.