Logging is a common and important programming practice, but choosing how to log is challenging, especially in a large, evolving software code base that provides many logging alternatives. Insufficient logging may complicate debugging, while logging incorrectly may result in excessive performance overhead and an overload of trivial logs. The Linux kernel has over 13 million lines of code, over 1100 different logging functions, and the strategies for when and how to log have evolved over time. To help developers log correctly we propose a framework that will learn existing logging practices from the software development history, and that will be capable of identifying new logging strategies, even when the new strategies just start to be adopted.