PProf

Load PProf and FlameGraphs together to activate the profile export extension. Replay a completed bundle without rerunning its workload.

From examples/kitchen-sink in the source checkout, run julia setup.jl extras. Then start Julia with julia --project=.controller/extras. Replace ACTUAL-PROFILE-RUN and ACTUAL-BUNDLE with paths from your completed profile run.

using PerfChecker, PProf, FlameGraphs
bundle = read_run_bundle("results/ACTUAL-PROFILE-RUN/bundles/ACTUAL-BUNDLE")
write_folded_profile(bundle, "exports/cpu.folded"; metric="julia.cpu.samples")
write_speedscope_profile(bundle, "exports/cpu.speedscope.json";
    metric="julia.cpu.samples")
write_pprof_profile(bundle, "exports/allocations.pb.gz";
    metric="julia.alloc.bytes", max_samples=1000)

profiles.jl applies this to CPU, wall and allocation metrics, reporting a missing stack selection instead of producing an empty profile. The flame graphs in the CPU, wall and allocation pages are observed stack data. A separate replay of the real Bibliography allocation bundle produced this PProf export, with its source manifest hash in the export receipt. The download uses relative runtime filenames; the receipt distinguishes the original and published export hashes. Stack names, line numbers, samples and weights are preserved. The adapter quantizes recorded allocation weights into sample repetitions; max_samples=1000 controls that approximation. It collects no new measurements.