Class ConstructMetrics

java.lang.Object
org.ek9lang.compiler.phase7.generator.ConstructMetrics
All Implemented Interfaces:
Function<ISymbol, FunctionMetrics>
public final class ConstructMetrics extends Object implements Function<ISymbol, FunctionMetrics>
THE resolver of phase-5 quality metrics for one operation, for every IR generator that registers a function entry with CoverageProbePlacer. Hold an instance; do not re-derive the lookup, and do not pass FunctionMetrics.EMPTY as a placeholder.

Sibling to CoverageDisplayName (it carries CoverageIdentity, the join key), and read by FunctionProbes.enter(CoverageProbePlacer, ISymbol, DebugInfo), THE registration every generator uses. When it was written there were four registering generators (OperationDfnGenerator, AggregateDfnGenerator, ServiceDfnGenerator, TextDfnGenerator) and three passed EMPTY, so every method and operator of every class, record, trait, component, service and text construct reported zero for all metrics — complexity, cognitive, nesting, statements, expressions, injections and coupling alike. Only free functions, which go through OperationDfnGenerator, ever carried real numbers. Any per-construct quality signal built on that data would have read every aggregate member as trivial and perfectly safe.

Two lookup rules, and they differ deliberately:

  • Body metrics (complexity, cognitive, nesting, statements, expressions, injections) are the operation's own, falling back to the enclosing scope only when the operation carries none at all — the synthetic-method case, where phase 5 measured the declaration rather than the generated _call.
  • Coupling is always inherited from the enclosing type, because efferent coupling is a property of the construct, not of an operation within it. Phase 5 chains its coupling check onto construct declarations only — never onto a method or operator — so an operation's own symbol never carries it. Every method of a twelve-dependency class is risky to change because the class is entangled; that is the sense in which coupling is change risk.
  • Defaulted is provenance rather than a metric, read straight off the phase-1 DEFAULTED marker. It travels with the metrics because the consumer that needs it (change-risk ranking) has only this record to go on, and the alternative — inferring "the compiler wrote this" from zero body metrics — is wrong for every trivial hand-written method.