Skip to content

formulas.md

Latest commit

 

History

History
102 lines (77 loc) · 3.76 KB

formulas.md

File metadata and controls

102 lines (77 loc) · 3.76 KB

Transformation Formulas

A practical guide to transforming values during decode/encode.

Overview

Formulas let you convert between raw on‑wire values and engineering values. You can express them as inline lambdas or as classes implementing java.util.function.Function.

  • Decode formula: raw -> target field type
  • Encode formula: target field type -> raw

Note: Lambda formulas require annotation processing for compile-time code generation. The fastproto bundle includes this by default. For Android projects, see Android guide for setup details.

Quick Steps

  1. Pick the field type and its data annotation (e.g., @UInt32Type).
  2. Decide if a lambda is sufficient, or a reusable class is better.
  3. Provide the correct input/output types for each direction.
  4. Test round‑trip: value -> encode -> decode -> value.
  5. Keep formulas pure (no side effects, deterministic).

Core Concepts

Direction & Types

  • Decode: Function<RawType, FieldType>
    • Example: @UInt32Type raw is Long; field may be Double (engineering).
  • Encode: Function<FieldType, RawType>
    • Example: reverse the above conversion.

Lambda vs Class

  • Lambda: @DecodingFormula(lambda = "...") / @EncodingFormula(lambda = "...") — included in fastproto bundle
  • Class: @DecodingFormula(MyFunc.class) / @EncodingFormula(MyFunc.class) where class implements Function<In, Out>
  • Precedence: Class‑based formulas win if both lambda and class are present.

Lambda Support

Lambda expressions are supported out of the box when using the fastproto bundle:

<dependency>
    <groupId>org.indunet</groupId>
    <artifactId>fastproto</artifactId>
    <version>4.1.0</version>
</dependency>

For Android projects, see android.md for separate configuration.

Lifecycle & Errors

  • Formulas are compiled/bound during graph resolve and invoked per field during decode/encode.
  • Throwing exceptions inside formulas will surface as DecodingException / EncodingException at call sites.

Examples

Lambda‑based unit conversion

import org.indunet.fastproto.annotation.*;

public class Weather {
  @UInt32Type(offset = 14)
  @DecodingFormula(lambda = "x -> x * 0.1")            // raw uint32 -> Pa * 0.1 -> engineering
  @EncodingFormula(lambda = "x -> (long) (x * 10)")    // engineering -> raw
  double pressure;
}

Class‑based (reusable) conversion

import java.util.function.Function;

public class PressureDecode implements Function<Long, Double> {
  @Override public Double apply(Long raw) { return raw * 0.1; }
}

public class PressureEncode implements Function<Double, Long> {
  @Override public Long apply(Double eng) { return (long) (eng * 10); }
}
public class Weather {
  @UInt32Type(offset = 14)
  @DecodingFormula(PressureDecode.class)
  @EncodingFormula(PressureEncode.class)
  double pressure;
}

One‑way formulas

  • You may specify only one side if the other is identity, e.g., only @DecodingFormula.

Advanced

  • Constants & context: Prefer constructor parameters or class constants inside class‑based formulas; avoid hidden globals.
  • Composition: Split complex logic into smaller Function classes and delegate.
  • Performance: Lambdas are compiled once and reused; keep formulas simple and avoid heavy allocations in hot paths.
  • Validation: Add unit tests with known vectors; ensure symmetry if both directions are specified.

Best Practices & Troubleshooting

  • Ensure types match the codec direction (e.g., UInt32 raw is Long).
  • Document non‑obvious conversions (units, offsets) near the field.
  • If results are off by a factor, verify integer vs floating arithmetic and cast points.
  • If exceptions bubble up, catch and wrap with clear messages in class‑based formulas to ease debugging.