cf-basic-signal 0.1.7

Cogniflow basic signal+IO step package (InlineSource, HandleSink, FIFO/average) with optional C++ acceleration.

cf.basic.signal StepPackage

Basic IO + signal-processing steps packaged for Cogniflow (InlineSource, HandleSink, FIFO buffer, averages, optional OPC UA source).

Steps

• cf:InlineSource - emit inline payload (JSON if prefixed with inline://)
• cf:HandleSinkStep - stub sink that forwards to a handle (duckdb://, parquet://, inline://, etc.)
• cf:FifoWindowBufferStep - windowed buffer
• cf:AverageStep - arithmetic / geometric / harmonic mean
• cf:OpcuaPhSource - virtual OPC UA pH source

Files

• cf-step-package.yaml - manifest
• steps.ttl - ontology fragment for the steps
• cf_basic_signal/steps.py - implementations (with optional C++ backend for signal ops)
• examples/inline_to_sink.json - minimal IO pipeline (InlineSource -> HandleSinkStep)
• examples/golden_fifo_avg.json - minimal signal pipeline (InlineSource -> FIFO -> Average)

Pipeline backend plugin

The package now registers itself as a pipeline_backend plugin:

• C++ ops (impl="cpp"): fifo_window_buffer, average (registered via capsule register_cpp_op)
• Python ops (impl="python"): inline_source, handle_sink, opcua_ph_source
• Metadata: pipeline_plugin.json (api_version=1, Data ABI Data-v1)
• Importing cf_basic_signal will auto-register C++ ops when available; otherwise the Python fallbacks are registered.

Quick check

1. Install package:

from core.step_package_manager import StepPackageManager
m = StepPackageManager("path/to/gui_sandbox.duckdb")
m.install_from_manifest("sand_toys/cf_basic_steps/cf_basic_signal/cf_basic_signal/cf-step-package.yaml")

2. Load server and verify:

from core.algorithm_server import AlgorithmServer
s = AlgorithmServer("path/to/gui_sandbox.duckdb")
assert s.step_registry.has_step("cf:InlineSource")
assert s.step_registry.has_step("cf:HandleSinkStep")
assert s.step_registry.has_step("cf:FifoWindowBufferStep")
assert s.step_registry.has_step("cf:AverageStep")

3. Run the example pipelines via your runner. Expected outputs:
   • inline_to_sink: status "handled via inline://dev-null"
   • golden_fifo_avg: window [4,5,6], mean 5.0, small SE.

Wheel builds & publishing

• Wheels are built with cibuildwheel for CPython 3.9-3.12 across Windows, Linux, and macOS (config in pyproject.toml).
• The workflow builds the single cf-basic-signal wheel (now including IO steps) and uploads it as an artifact.
• PyPI/TestPyPI publishing happens on tagged pushes via the trusted publisher action (pypa/gh-action-pypi-publish).