Paul Ganssle

Attractive nuisances in software design

Paul Ganssle — Wed, 04 Jan 2023 11:14:28 -0500

When a feature applicable only to a niche use case seems like it would be useful for a common use case, you are as good as inviting your users to do the wrong thing. It is useful to keep this in mind when thinking about expanding your API.

Why naïve times are local times in Python

Paul Ganssle — Tue, 05 Apr 2022 09:22:22 -0400

Starting in Python 3, naïve datetimes stopped being considered "unitless" datetimes and started representing local time with a known mapping to UTC; this post explains why this was the best option for a "local" time zone given the existing constraints.

xfail and code coverage

Paul Ganssle — Mon, 13 Dec 2021 11:50:45 -0500

xfail is great, but because it executes test code as part of failing tests, it can end up artificially inflating coverage numbers. This post discusses the problem and gives one way to ameliorate it.

A pseudo-TDD workflow using expected failures

Paul Ganssle — Mon, 22 Nov 2021 04:21:19 -0500

If, like me, you're not disciplined enough to actually practice test driven development, you can use git history rewriting to get some of the same benefits.

How and why I use pytest's xfail

Paul Ganssle — Tue, 09 Nov 2021 06:47:00 -0500

For tests that currently fail but shouldn't fail, pytest.mark.xfail is a useful way to pre-commit your regression tests.

Why you shouldn't invoke setup.py directly

Paul Ganssle — Tue, 19 Oct 2021 08:55:30 -0400

The setup.py interface is an old convention from distutils, one that has bugs that cannot be fixed to bring it into line with modern Python packaging approaches. As a result, all direct invocations of setup.py are deprecated or discouraged if not formally deprecated yet.

Generating k-Smooth numbers in Python

Paul Ganssle — Fri, 08 Oct 2021 07:56:14 -0400

A post showing how closures and generators can be used to generate Regular numbers (also called Hamming numbers) in Python.

Subtests in Python

Paul Ganssle — Tue, 28 Apr 2020 12:15:24 -0400

An overview of the use of subtests in Python as of April 2020.

Testing an Arch Linux package in Gitlab CI

Paul Ganssle — Mon, 30 Dec 2019 12:14:21 -0500

If your CI runs all jobs as root (like Gitlab's does), you may have trouble testing Arch Linux's makepkg utility, which cannot run as root. This short blog post explains how I solved the problem.

String concatenation in Python

Paul Ganssle — Fri, 15 Nov 2019 16:14:39 -0500

Thanks to some optimizations in Python, building strings by incremental concatenation is cheaper than you might think.

Stop using utcnow and utcfromtimestamp

Paul Ganssle — Tue, 05 Nov 2019 09:56:21 -0500

A public service announcement about the dangers of utcnow and utcfromtimestamp and the benefits of using their replacements.

Testing your python package as installed

Paul Ganssle — Fri, 23 Aug 2019 13:30:00 -0400

An explanation of a few ways to make sure your tests use the version of the package that your users will install.

Testing Weak References

Paul Ganssle — Wed, 13 Jun 2018 11:32:03 -0400

A short post describing one way to write tests that a weak reference is working as expected.

pytz: The Fastest Footgun in the West

Paul Ganssle — Mon, 19 Mar 2018 10:00:00 -0400

An explanation of the differences between dateutil and pytz's time zone models and why people tend to misuse pytz and introduce bugs into their datetime code.

Semantics of timezone-aware datetime arithmetic

Paul Ganssle — Thu, 22 Feb 2018 12:00:00 -0500

A discussion of the competing ambiguous definitions of what datetime arithmetic means when a time zone has a non-constant offset and the definitions that Python uses.

A curious case of non-transitive datetime comparison

Paul Ganssle — Thu, 15 Feb 2018 15:00:00 -0500

An example of how daylight saving time interacting with Python's datetime comparison model can lead to an absurd result.