NYTProf vs Benchmark: Best Tool for Initial Bottleneck Identification in Perl

Goal

Determine which documented profiling approach—Devel::NYTProf or Benchmark.pm—offers the most reliable insight into where a Perl program spends its time before any code changes.

Constraints & Uncertainty

• NYTProf provides per‑line and per‑subroutine inclusive/exclusive timings but incurs measurable overhead that skews absolute numbers.
• Benchmark measures only code supplied to it, requiring prior suspicion of hotspots and suffering from loop‑count sensitivity.
• Deciding whether to trust inclusive or exclusive times in NYTProf can lead to different optimization targets.

Questions

1. For a large, complex script, should I start with a full NYTProf run to locate hotspots, or use Benchmark on a subset of suspected functions?
2. When NYTProf’s inclusive and exclusive time rankings diverge, which metric should guide the first rewrite?
3. What loop‑count strategy in Benchmark best mitigates overhead bias for micro‑benchmarks of small subroutines?