As Android platform evolves in a fast pace, API-related compatibility issues become a significant challenge for developers. Existing works mainly focus on detecting whether an app performs sufficient checks to avoid invoking incompatible APIs on platforms that do not support them. Besides checking the presence of compatibility issues, another important problem is how to fix/work-around these issues. To shed light on the current practice of fixing evolution-induced API compatibility issues, this paper performs the first large-scale study of compatibility issues for about 300,000 Android market apps, and more importantly, their solutions (if exist). However, it is in general very challenging to determine if developers have put in counter-measure for a compatibility issue, as different APIs have diverse behaviors, rendering various repair. We propose a learning-based approach, RAPID, that leverages static analysis and machine learning to build a classifier to determine if a compatibility issue has been addressed. Results show that the trained classifier is quite effective by achieving a F1-score of 95.21% and 91.96% in the training stage and the validation stage respectively. With the help of RAPID, our study yields a number of interesting findings, e.g. only 38.4% of the incompatible API uses are actually provided alternative implementation as a fix; the fix ratio for incompatible APIs with Google recommendations is significantly higher than those without recommendations, but Google only provides recommendations for very few APIs; developers find more ways to repair such issues than Google’s recommendations and there is a large room to improve the current status of compatibility issue handling.