Link Search Menu Expand Document
spicypath

Supported formats

Inputs are detected by extension and a content sniff, gunzipped transparently (.gz via DecompressionStream), and normalized into the canonical model described in architecture.md. Nothing is uploaded — parsing happens in the browser (or in Node for the test harness).

Input formats

Format Extension(s) Plane Multi-value Notes
folded / collapsed .folded, .txt aggregated — a;b;c 42 lines; the lowest common denominator
pprof .pprof, .pb.gz aggregated ✅ protobuf; keeps all sample value types
V8 .cpuprofile .cpuprofile time-ordered — Chrome / Node CPU profile; per-sample timing reconstructed from the node tree + timeDeltas
speedscope .speedscope.json time-ordered — both sampled and evented variants; evented O/C events are reconstructed into per-interval timed samples
perf script .perf, .txt time-ordered ⚠️ perf script text output; blank-line-separated sample blocks
OTLP Profiles .otlp aggregated ✅ OpenTelemetry profiling signal (profiles/v1development, protobuf); shares the pprof value-type mapping. Lossless pprof ↔ OTLP in our model
Gecko .gecko.json time-ordered — Firefox Profiler / samply processed-profile JSON (meta.version 5); detected by content sniff. Per-sample timing preserved
JFR .jfr time-ordered — JDK Flight Recorder (binary, JVM-only); a native browser-pure decoder (chunk → metadata → constant pool → jdk.ExecutionSample). Detected by the FLR\0 magic. Per-sample timing preserved

“Plane” is whether per-sample timing survives. Aggregated formats carry no timestamps, so the time-ordered flame chart is unavailable for them (the chart tab hides itself when hasTiming is false — see architecture.md).

JFR is the one format with no synthetic golden round-trip (it has no emitter — you can’t fabricate a faithful .jfr). Instead test/parse-jfr-test.ts generates a real recording with the local JDK at test time and validates the parser against the jfr tool’s own output; it skips cleanly when no JDK is installed. Nothing binary is committed.

Export formats

The loaded profile can be saved as:

  • .speedscope.json — re-importable by this viewer and by speedscope.
  • .folded — the universal lowest-common-denominator text.
  • .svg — a static flame-graph snapshot of the current view.

Adding a parser

  1. Write a pure core src/parse-<format>.js that takes bytes or text and returns the canonical model via ProfileBuilder (see src/model.js). Keep it free of Node APIs so it runs in the browser unchanged.
  2. Register detection in src/ingest.js — an extension match plus a content sniff.
  3. Add a fixture and an emitter under test/ and wire it into the golden round-trip suite so parse → model → re-emit stays stable.

Keeping the parser core pure (bytes/text in, model out) is what lets the same code run in the browser and under the Node test harness.

spicypath — MIT. Nothing is uploaded; parsing happens in your browser.