Thread by @di_codes

Some researchers downloaded all of @pypi (!) and parsed the source of every package (!!!)

Some insights below 👇

(1/9)

Insight #1: PyPI is growing.

Compound Annual Growth Rate (CAGR) for new packages, active packages, new releases, and new authors are all in the double-digits:

• New packages: 43.28%
• Active packages: 47.31%
• New releases: 51.21%
• New authors: 39.30%

(2/9)

Insight #2: PyPI is becoming more dense.

"...the number of connections or imports per package is increasing", and "annual growth rate in the number of imports [exceeds] the growth rates of the number of packages or releases alone by 10%."

(3/9)

Insight #3: License classification on PyPI is a bit of a mess.

"For example, PyPI does not distinguish between any variations of the BSD License" 🤦‍♂️

We're working on finding a way to improve this, see github.com/pypa/warehouse… for more details.

(4/9)

Insight #4: Software is never done.

More packages are either "Alpha" or "Beta" than "Production/Stable":

• Alpha: 27.70%
• Beta 35.36%
• Production/Stable 24.93%

(5/9)

Insight #5: Everyone wants to live in the future.

Top 10 most imported modules, in-order, are:

• __future__
• os
• sys
• logging
• re
• datetime
• json
• numpy
• unittest
• time

(6/9)

Insight #6: Most packages don't change very much.

"In general, we find that most authors contribute a single package with a small number of releases, and... nearly two in every three packages went without a release during 2018."

(7/9)

Insight #7: That said, the averages of PyPI might surprise you.

On average:

• Each author has 2.15 packages;
• Each package has 6.93 releases;
• A package makes a new release every 65.38 days

(8/9)

PDF of the full paper is here: arxiv.org/pdf/1907.11073…

Thanks to Ethan Bommarito and Michael J Bommarito II for taking the time to do this analysis! 👏

(9/9)