Contributing to Pyrodigal

For bug fixes or new features, please file an issue before submitting a pull request. If the change isn’t trivial, it may be best to wait for feedback.

Coding guidelines


This project targets Python 3.5 or later.

Python objects should be typed; since it is not supported by Cython, you must manually declare types in type stubs (.pyi files). In Python files, you can add type annotations to function signatures (supported in Python 3.5) but not in variable assignments (supported only from Python 3.6 onward). However, Cython allows you to use f-strings even when compiling the code for Python 3.5.

Interfacing with C

When interfacing with C, and in particular with pointers, use assertions everywhere you assume the pointer to be non-NULL. Also consider using assertions when accessing raw C arrays, if applicable.

Setting up a local repository

Make sure you clone the repository in recursive mode, so you also get the wrapped code of Prodigal which is exposed as a git submodule:

$ git clone --recursive

Running tests

Tests are written as usual Python unit tests with the unittest module of the standard library. Running them requires the extension to be built locally:

$ python build_ext --debug --inplace
$ python -m unittest discover -vv

Running benchmarks

Connection scoring

The benches folder contains benchmarks for evaluating the performance of the node connection scoring step, essentially to make sure that the SIMD code makes it faster.

Start by building pyrodigal locally:

$ python build_ext --debug --inplace

Then make sure you have the required packages and data:

$ pip install --user -r benches/connection_scoring/requirements.txt
$ python benches/connection_scoring/data/

Finally, run the benchmarks and plot the results:

$ python benches/connection_scoring/ -d benches/connection_scoring/data/ -o times.json
$ python benches/connection_scoring/ -i times.json --show

Setting up the Prodigal comparison

Julian Hahnfeld has developed a pipeline for comparing Pyrodigal and Prodigal to make sure they produce exactly the same results on a set of genomes:

To use it, first install the local version of Pyrodigal that you want to compare against Prodigal:

$ pip install --user . --force-reinstall

Then compile the patched Prodigal binary from the code that is ditributed with Pyrodigal:

$ make -C vendor/Prodigal/

Finally, clone the comparison repository, download example genomes, and run the comparison:

$ git clone
$ cd prodigal-pyrodigal-comparison
$ wget -i test_genomes.txt -P genomes
$ python --prodigal ../vendor/Prodigal/prodigal --genome genomes/*