Parser Guide

Vizb supports multiple benchmark frameworks through its parser registry. Use --parser to select the right one for your benchmark output.

Why Multiple Parsers

Each benchmark framework produces output in a different format. Vizb’s parser registry knows how to read them all and convert the results into a common structure — so your charts, grouping, filtering, and merging work identically regardless of where the benchmarks came from.

Choosing a Parser

Go

Default parser. Reads go test -bench text or -json output.

go test -bench . | vizb -o output.html
# or explicitly
go test -bench . | vizb --parser go -o output.html

When JSON auto-detection applies: --parser go is required for JSON input (the go test -bench -json format).

Metrics extracted:

Metric	Description
Execution time	ns/op, with configurable unit
Memory	B/op, with configurable unit
Allocations	allocs/op, with configurable unit
Throughput	MB/s, B/s, GB/s, or custom
Iterations	Number of iterations run

Criterion

Parses cargo bench output from Criterion.

cargo bench | vizb --parser rs:criterion -o output.html

Metrics extracted:

Metric	Description
Latency avg	Mean time per iteration
Latency lower	Lower bound of confidence interval
Latency upper	Upper bound (±)

Note

Criterion output contains ANSI color codes — vizb strips these automatically before parsing.

Divan

Parses cargo bench output from Divan.

cargo bench | vizb --parser rs:divan -o output.html

Metrics extracted:

Metric	Description
Latency fastest	Fastest measured iteration
Latency slowest	Slowest measured iteration (±)
Latency median	Median across all samples
Latency mean	Mean across all samples
Samples	Number of samples collected

Note

The iters column from Divan output is not extracted — it represents per-sample iteration counts rather than an aggregate metric.

Vitest

Parses vitest bench output.

Experimental

Benchmarking in Vitest is experimental and does not follow SemVer. The parser was tested against v4.1.7. Output format changes in future Vitest versions may break parser compatibility.

npx vitest bench | vizb --parser js:vitest -o output.html

Metrics extracted:

Metric	Description
Throughput	Operations per second (ops/s)
Latency min / max	Fastest and slowest run
Latency avg	Mean across all runs
Latency p75 / p99 / p995 / p999	Percentile latency
RME	Relative margin of error (±)
Samples	Number of samples collected

Naming convention: When describe() wraps bench() blocks, names are concatenated with /:

describe('n=100', () => {
  bench('bubbleSort', () => { ... })
  bench('insertionSort', () => { ... })
})

Produces names n=100/bubbleSort and n=100/insertionSort. Use --group-pattern with any valid 2D combination (n/x, n/y, x/n, y/n, x/y, y/x) to split the suite name into its own dimension.

See the Grouping Guide for detailed pattern and regex syntax.

Tinybench

Parses Tinybench console.table() output.

node bench.js | vizb --parser js:tinybench -o output.html

Metrics extracted:

Metric	Description
Latency avg / med	Mean and median latency
Latency RME / MAD	Relative margin of error and median absolute deviation (±)
Throughput avg / med	Mean and median operations per second
Throughput RME / MAD	Relative margin of error (±)
Samples	Number of samples collected

All Parser Keys

Key	Framework	Language
go	Go testing (benchfmt)	Go
rs:criterion	Criterion	Rust
rs:divan	Divan	Rust
js:vitest	Vitest	JavaScript / TypeScript
js:tinybench	Tinybench	JavaScript / TypeScript

Add a New Parser

Want to add support for a benchmark framework or language not listed here? Parser contributions are welcome. Here’s what the process looks like:

1. Choose a key following the <lang>:<framework> convention (e.g., py:pytest, java:jmh)
2. Implement a parse function that extracts []shared.BenchmarkData from the framework’s output
3. Register it via parser.Register("your-key", YourParseFunc) in an init() block
4. Add tests with real output samples and edge cases
5. Open a PR on GitHub — the existing parsers in pkg/parser/ are good references to follow

Tip

The parser registry is designed to be extensible. As long as your parse function returns benchmark data with names, values, and optional metadata, everything else — grouping, filtering, merging, time unit conversion — works automatically.