The goal of course isn't to chase some coverage number, instead adding tests has already helped us uncover multiple issues, simplify many implementation details, push forward the C++17 migration, and gives us more confidence during refactoring and when adding new features.

Here's a graph from the time we started using Codecov. It seems that something wasn't configured correctly in the beginning, thus the big chunk of 25% code coverage that later dropped again.

https://app.codecov.io/gh/SFML/SFML?search=&trend=all%20time