fletchr-measurand

Arrow-native bit-fragment extraction and engineering-unit decoding for binary protocol data. Parses parameter and measurand spec strings into evaluable objects; runs them against pa.Tables whose data columns are fletchr.uintn extension arrays.

Why?

Bit-packed protocol data routinely needs two things:

• Bit-fragment extraction — "this value is word 3 concatenated with word 4 as a 16-bit value." Done by hand, it's masking and shifting; done at scale, it's bugs.
• Engineering-unit decoding — "raw × 0.1 + (-273.15)" for a scaled measurement.

fletchr-measurand makes both declarative. Parameter("[1+2]") extracts word 1 concatenated with word 2; Measurand(parameter=..., encoding="2c", euc=EUC(...)) adds decoding (two's complement, float, etc.) and scaling.

Features

• Parameter DSL — [1] (single word), [1+2] (multi-word, MSB first), [1-3] (range), [1+3+5] (non-contiguous concatenation); plus generator / iteration patterns for repeated bit-fragments at successive column offsets.
• Level 1 encodings via byteforge — unsigned, sign-magnitude, one's-complement, two's-complement, offset-binary, IEEE 754 (32/64), MIL-STD-1750A (32/48), packed BCD, Gray code, IBM/DEC/TI floats.
• Level 2 EUC — linear scaling (scale_factor, data_bias, scaled_bias) and arbitrary callable transforms.
• Arrow-native build chain — Parameter.build and Measurand.build consume pa.Table directly and return pa.ExtensionArray / MeasurandResult. Validity bitmaps propagate through both decode levels.
• Field-metadata convention — the fletchr.encoding / fletchr.scale_factor / fletchr.data_bias / fletchr.scaled_bias keys defined in SPEC-decode.md give a language-neutral wire format for decode rules attached to Arrow column metadata.

Install

uv add fletchr-measurand        # or: pip install fletchr-measurand

Requires Python 3.9+. Pulls in fletchr-uintn, pyarrow >= 16, numpy >= 2.0, byteforge, lark, and antlr4-python3-runtime.

Quickstart

import pyarrow as pa
from fletchr_uintn import uintn_array
from fletchr_measurand import EUC, Measurand, parameter_parser

# A FrameArray-shaped pa.Table: one column per word position (c1..cN,
# 1-based), each a fletchr.uintn column. Real users get this from
# fletchr-core's FrameArray; constructed manually here for illustration.
table = pa.table({
    "c1": uintn_array([0x12, 0x34, 0x56], bits=8),
    "c2": uintn_array([0xFF, 0x80, 0x00], bits=8),
    "c3": uintn_array([0x10, 0x20, 0x30], bits=8),
})

# Parameter spec is 1-based: [2] selects word 2 (column c2). Decode it as
# signed two's-complement, then linear-scale to Celsius from a centikelvin
# raw representation.
m = Measurand(
    parameter=parameter_parser.parse("[2]"),
    encoding="2c",
    euc=EUC(scale_factor=0.1, scaled_bias=-273.15),
)

result = m.build(table)
result.dns.to_pylist()        # [255, 128, 0]       — raw 8-bit DNs
result.decoded.to_pylist()    # [-1, -128, 0]       — Level 1: two's-complement
result.values.to_pylist()     # [-273.25, -285.95, -273.15]  — Level 2: scale + bias

Multi-word values use + to concatenate, MSB first:

# Words 2 and 3 as a single 16-bit unsigned value (c1 high, c2 low).
m16 = Measurand(parameter=parameter_parser.parse("[2+3]"))
m16.build(table).values.to_pylist()    # [65296, 32800, 48]

See SPEC-decode.md for the full encoding registry and the field-metadata convention.

Public API

from fletchr_measurand import (
    # Parameter / Measurand DSL
    Parameter, ParameterSpec, parameter_parser,
    Measurand, MeasurandSpec, measurand_parser,
    resolve_parameter, resolve_measurand,
    # Result + Level 2 transform
    MeasurandResult,
    EUC, SamplingStrategy,
    # Calculator parser (for the SPEC's expression-mode hook)
    calculator_parser,
)

Status

The runtime data model (Parameter, Measurand, MeasurandResult, EUC) is stable. The wire-format convention in SPEC-decode.md is draft v0.5 — the encoding registry is locked, but one open question remains (the expression-mode allow-list) before incrementing to v1.0. Field-metadata read/write helpers that round-trip the SPEC keys onto pa.Field.metadata are tracked as future work.

Links

• Source: https://github.com/fletchr-labs/fletchr
• Issues: https://github.com/fletchr-labs/fletchr/issues
• Decode metadata spec: SPEC-decode.md

License

MIT.