Working on fastplot

Setting up

git clone https://github.com/Rochy111/fastplot
cd fastplot
pip install -e ".[test]"
python -m fastplot warm

Tests

python -m pytest

The tests use matplotlib, cartopy and contourpy as references. They draw the same figures with both libraries and compare limits, ticks, labels, legend and layout geometry, and pixels. GPU tests are skipped when there is no GPU, and tests that compare against matplotlib's own output are skipped on matplotlib older than 3.11, whose text shaping, legends and colormap names differ.

CI runs the suite on Linux (Python 3.10, 3.12 and 3.13, numpy 1 and 2), Windows and macOS on every push.

Guidelines

A change to the drawing code should not change any pixel unless that is its purpose. Compare images before and after the change, and check that the CPU and GPU still produce identical images.

Kernels are numba functions compiled with cache=True and nogil=True. They take writable, contiguous arrays so that each compiles for a single signature. If you add a kernel or change its argument types, regenerate the warm-up manifest with python tools/make_warm_manifest.py.

Measure performance before and after a change. tools/bench.py covers maps, tools/bench_plain.py plain plots and tools/bench_png.py the PNG encoder; each prints medians you can compare.

Building these docs

pip install mkdocs-material
mkdocs serve

The gallery images are made by the scripts in docs/examples/. Run one to redraw its picture into docs/images/.

How it is organised

How it works inside walks through the rendering pipeline, the conventions the code follows and what each module does.