Module bench_corpus 

Deterministic synthetic corpora for benchmarking.

Every benchmark that needed pages used to generate them inline, so the numbers were only comparable within a single bench file: bench_scalability and incremental_1000_pages wrote different front matter and different body lengths, then reported timings as though they measured the same work. A published figure has to be reproducible by whoever reads it, which means one generator, one shape, and no hidden inputs.

Determinism

Content is derived from a seed and the page index through a small xorshift generator rather than the rand crate: the corpus must be byte-identical across machines, architectures and toolchain versions, and rand’s output is explicitly not stable across releases. The same seed therefore yields the same corpus in a year’s time, which is the property that lets a benchmark number be checked rather than believed.

Body length and tag selection vary per page — a corpus of identical pages measures the cache, not the compiler — but vary predictably.

Examples

use ssg::bench_corpus::{generate_corpus, CorpusSpec};
let dir = tempfile::tempdir().unwrap();
let spec = CorpusSpec::new(64);
let written = generate_corpus(dir.path(), &spec).unwrap();
assert_eq!(written, 64);

Structs

CorpusSpec

Shape of a synthetic corpus.

Functions

generate_corpus

Writes spec.pages Markdown files into dir, returning the count written.