How do I share a Criterion.rs benchmark report?

Run `cargo bench`, then publish the whole `target/criterion/report/` folder to Comma. `index.html` becomes the report body and the linked per-benchmark pages and SVG plots upload alongside it with their references rewritten, so the cross-links work at the URL.

Why does the Criterion report look broken when I open it locally?

It is a folder, not a file — `index.html` links to a page per benchmark and to generated SVG plots in sibling directories. Send just the index and every link is dead; zip the folder and the reader has to unzip it first and find the right entry point.

Does Criterion's regression detection still show up?

Yes. Criterion compares against the previous saved baseline and renders the change with its confidence interval — the 'Performance has improved/regressed' verdict and the PDF/violin plots are part of the generated HTML, so they publish with everything else.

No. Criterion falls back to the plotters backend if gnuplot is not installed, which produces the same report with slightly different plots. Either way the output is the same folder structure.

Can I keep a history across commits?

PATCH the same report id from each benchmark run. One URL per suite, one revision per run — which pairs well with Criterion's own baseline comparison, since the revision list shows what the baseline was at the time.

Share a Criterion.rs benchmark report

Criterion does the statistics most benchmarking skips. It estimates the mean and the distribution, compares against a stored baseline, and tells you whether the change it measured is outside the noise — with plots that make an outlier obvious rather than something you have to infer from a single number.

cargo bench                       # → target/criterion/report/index.html

And then the report is a directory:

target/criterion/
├─ report/index.html          ← the entry point
├─ parse_json/
│  ├─ report/index.html       ← per-benchmark page
│  ├─ pdf.svg  violin.svg  regression.svg
│  └─ change/                 ← comparison against the previous baseline
└─ serialize/…

Sending index.html on its own gives the reader a page of dead links. Zipping it gives them a chore. Screenshotting the violin plot throws away the reason Criterion drew a violin plot.

Publish the folder

Drag target/criterion/report/ — or the whole target/criterion/ if you want the per-benchmark drilldowns — into the app:

index.html becomes the report body.
The per-benchmark pages and every generated SVG upload alongside it, and their relative references are rewritten to the uploaded copies, so clicking through to a specific benchmark works.
Scripts and inline SVG render inside a sandboxed iframe (allow-scripts, no allow-same-origin).

Or PATCH the entry file from a script:

cargo bench

curl -fsS -X PATCH "https://commareports.com/api/v1/reports/$REPORT_ID" \
  -H "Authorization: Bearer $COMMA_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d "$(jq -n --rawfile html target/criterion/report/index.html \
        --arg t "Criterion — parser, $(git rev-parse --short HEAD)" \
        '{title: $t, html: $html}')"

Publish the comparison, not just the run

The most valuable page Criterion generates is change/ — the comparison against the saved baseline, with the verdict and its confidence interval. Save a named baseline before a change and compare against it after:

git checkout main
cargo bench -- --save-baseline main

git checkout my-optimization
cargo bench -- --baseline main

The generated report now contains "Performance has improved" (or has not) with the interval attached. That is the artifact worth a URL in the pull request, because it is the one a reviewer can disagree with on the evidence.

From CI

Runner noise makes absolute numbers unreliable, but Criterion's own noise modelling means the verdict holds up better than a raw comparison would. PATCH one report id per suite and set a wide regression threshold:

cargo bench -- --baseline main --noise-threshold 0.05

One URL, a revision per run, and the revision list makes the runner's variance visible instead of leaving you to guess at it.

What review adds

Anchored threads on the benchmark under discussion — see commenting on HTML.
Revisions, so a claimed speedup is checkable later.
Access per report — private, team, domain-gated or named reviewers. See the sharing model.

Limits

Entry HTML: 5 MB. Assets: 25 MB per file, 250 MB and 500 files total. Criterion generates several SVGs per benchmark, so a suite with hundreds of benchmarks is the case that approaches the 500-file cap — publish report/ alone rather than all of target/criterion/ when it does.
60 requests/minute per token.

Try it

Comma is free — unlimited reports, unlimited commenters, unlimited revision history.

Publish a benchmark →