Class InterproceduralLockAnalyzer

java.lang.Object
org.ek9lang.compiler.phase5.lockanalysis.InterproceduralLockAnalyzer
public final class InterproceduralLockAnalyzer extends Object
Phase C.2 of slice-2 deadlock detection: compute per-function mayEnter summaries by composing direct enters with reachable enters via the call graph and CrossThreadEdge list.

Algorithm: monotone worklist iteration to fixpoint. For each function with an outgoing edge to g, add to mayEnter(f) every TransitiveEnterRecord in mayEnter(g), with the edge kind composed against the path kind (any cross-thread edge poisons the whole path). Sets only grow, so the iteration terminates even with cyclic call graphs.

Edge kind composition (per EK9_CONCURRENCY_MODEL.md §5.2): SAME ∘ SAME = SAME; any other combination is CROSS_THREAD. See EdgeKind.composeWith(EdgeKind).

First-cut simplifications:

  • No explicit SCC condensation. Worklist iteration handles cycles by construction (monotone sets). Tarjan-based topological processing would reduce iteration count for deep linear chains; can be added later if profiling justifies it.
  • No parameter substitution. Lock identities expressed as parameter-rooted tokens in the callee propagate as-is, even though semantically they alias caller-supplied values. The dominant EK9 pattern is field-rooted locks (identities are stable across call boundaries), so this affects a minority of cases. Phase C.3 may over-flag parameter-rooted reentrancy until a follow-up adds proper substitution.

Outputs: every FunctionLockSummary in LockAnalysis.summariesByFqn() has its mayEnter field populated with the transitive closure. Summaries are auto-created for any function reached in the call graph that doesn't already have one.

  • Constructor Details

    • InterproceduralLockAnalyzer

      public InterproceduralLockAnalyzer(CallGraph callGraph, LockAnalysis lockAnalysis)

  • Method Details

    • analyse

      public void analyse()