This directory contains Python utilities used during development of the la-stack crate.

The Python tooling in this repo is managed with uv.

just python-sync # or: uv sync --group dev

The plotter reads Criterion output under:

• target/criterion/d{D}/{benchmark}/{new|base}/estimates.json

And writes:

• docs/assets/bench/vs_linalg_{metric}_{stat}.csv
• docs/assets/bench/vs_linalg_{metric}_{stat}.svg

To generate the single “time vs dimension” chart:

By default, the benchmark suite runs for dimensions 2–5, 8, 16, 32, and 64.

1. Run the benchmarks you want to plot (this produces target/criterion/...):

# full run (takes longer, better for README plots) just bench-vs-linalg lu_solve # or quick run (fast sanity check; still produces estimates.json) just bench-vs-linalg-quick lu_solve

2. Generate the chart (median or mean):

# median (recommended) just plot-vs-linalg lu_solve median new true # median + update README's benchmark table (between BENCH_TABLE markers) just plot-vs-linalg-readme lu_solve median new true # or mean just plot-vs-linalg lu_solve mean new true

This writes:

• docs/assets/bench/vs_linalg_lu_solve_median.csv
• docs/assets/bench/vs_linalg_lu_solve_median.svg (requires gnuplot)

(For stat=mean, the filenames end in _mean instead of _median.)

Plot a different metric:

uv run criterion-dim-plot --metric dot --stat median --sample new uv run criterion-dim-plot --metric inf_norm --stat median --sample new

Plot a different statistic:

uv run criterion-dim-plot --metric lu_solve --stat mean --sample new

Plot the previous (baseline) sample instead of the newest run:

uv run criterion-dim-plot --metric lu_solve --stat median --sample base

Use a log-scale y-axis:

uv run criterion-dim-plot --metric lu_solve --stat median --sample new --log-y

Write to custom output paths:

uv run criterion-dim-plot \ --metric lu_solve --stat median --sample new \ --csv docs/assets/bench/custom.csv \ --out docs/assets/bench/custom.svg

CSV only (skip SVG/gnuplot):

uv run criterion-dim-plot --no-plot --metric lu_solve --stat median --sample new

SVG rendering requires gnuplot to be installed and available on PATH.

Install (macOS/Homebrew):

brew install gnuplot

Verify the installed version:

gnuplot --version

This repo has been tested with gnuplot 6.0 patchlevel 3 (Homebrew gnuplot 6.0.3).

# Full regeneration from all history just changelog # Prepend only unreleased changes for a new version just changelog-unreleased v0.3.0

just changelog runs git-cliff -o CHANGELOG.md followed by postprocess-changelog (strips trailing blank lines). Configuration lives in cliff.toml at the repo root.

just tag v0.3.0 # create annotated tag from CHANGELOG.md section just tag-force v0.3.0 # recreate tag if it already exists

The tag-release CLI (in tag_release.py) extracts the matching version section from CHANGELOG.md, validates semver, and handles GitHub's 125KB tag-annotation size limit.

See docs/RELEASING.md for the full release workflow.