Скачать или смотреть Experiments for your Android builds driven by Gradle Profiler - Ubiratan Soares

Recorded at Android Summit 2020 https://androidsummit.org

For large Android projects, Gradle builds might be a real pain. While tools like Gradle Build Scans and/or Gradle Enterprise are available out there, eventually they are not affordable or ideal for every team or they don't event represent the right tool for a given problem. For instance, one would like to evaluate experiments for Gradle builds, measuring and experimenting with locally generated data in order to figure out if the local build can actually improve given some particular change...

In this talk, we will learn more about Gradle Profiler, an open-source tool provided by the Gradle team which allows profiling and bench-marking of builds driven by a high level description of scenarios to be exercised. Using this tool, we'll learn how define, execute and check results for real experiments - using as targets some well know open-source Android projects - defining the scenarios and the hypothesis we want to validate.

We'll behold "when Data Science meets Build Engineering" and learn how the data generated by Gradle profiler can be used as an input of a classical Statistics tests in order to verify if we found something meaningful (or not) with our experiments. By leveraging such approach, you can evaluate if Gradle builds can effectively improve given some meaningful change you did, for instance something at Dagger usage, some Gradle configuration, some change on dependencies and/or modules setup, and etc. All of that using the scientific way.