Swift debounce investigation: historical observations
Development session: 2026-10-13

These are selected technical observations from the author's saved development
session. They are not new benchmark runs. They supplement the result files in:
https://github.com/ivan-magda/swift-debounce/tree/990cf04587cf922748c13876d2843071c22873fa/Benchmarks/results

1. Selected symbols observed in macOS sample output

Original long-lived reference:
  DebounceStateMachine.newValue(_:)
  OptimizedDebounce.emit(value:)
  type metadata accessor for ContinuousClock.Instant
  static ContinuousClock.now.getter
  ContinuousClock.Instant.advanced(by:)
  static Duration.+ infix(_:_:)
  __swift_memcpy16_8
  __swift_instantiateGenericMetadata
  swift_getGenericMetadata
  swift_getTupleTypeMetadata3
  swift_getTupleTypeMetadata

Fast prototype after changing the time representation:
  FastDebounce.emit(value:)
  closure #1 in FastDebounce.emit(value:)
  type metadata accessor for Optional
  __swift_instantiateGenericMetadata
  swift_getGenericMetadata
  mach_continuous_time

These lists contain selected observed names, not contiguous call stacks.
They do not encode sample counts, percentages, call order or generated assembly.

2. Historical cross-module optimization A/B experiment

Release LockBench; target 3 seconds per invocation; 10,000 emits between outer
clock checks; debounce interval 3,600 seconds. One listed run per configuration.
The interval keeps callback delivery outside the repeated-emit workload.

Command structure, retaining actual arguments and execution order while
omitting output filtering and cosmetic shell printing:

  for mode in 0 1; do
      CMO=$mode swift build -c release --product LockBench
      .build/release/LockBench profile 3 fast
      .build/release/LockBench profile 3
  done

Saved output:
  CMO=0  fast: emits: 27580000
  CMO=0  article: emits: 15770000
  CMO=1  fast: emits: 257900000
  CMO=1  article: emits: 15680000

CMO toggled the historical Debounce target's -cross-module-optimization flag.
These counts compare throughput over a target time window. They are neither
medians nor directly measured per-call latencies. The clock was checked in
batches, so an invocation could run slightly beyond its target window.

The old profile banner still used the OptimizedDebounce name in fast mode;
the saved harness selection code and FastDebounce sample symbols identify
the actual prototype. The current package is not the original historical
FastDebounce implementation and its current manifest does not expose this
old CMO environment toggle as a reproduction interface.