Two-Stage Patch Synthesis for API Migration from Single API Usage Example

Abstract

Third-party libraries are widely used and constantly evolving. When migrating client code to a new API, a major challenge is editing client code to adapt to incompatible changes of the API. Some tools provide automated migration that synthesizes a generic patch from migration samples in a pair of beforeand after-migration snippets. However, they still have limited applicability due to the difficulty in retrieving an adequate pair in which before- and after-migration snippets invoke the sourceand destination-API of the migration respectively. In this paper, we present AUTOMIG, which addresses the problem by a patch synthesis from a single API usage example that invokes the destination-API. Due to the absence of a beforemigration version of the API usage example, it is nontrivial to find out the invariant context of API invocation through the migration. To synthesize a patch that preserves the context in the client code, AUTOMIG synthesizes patches in two-stage patch synthesis procedure. The first stage synthesizes a generic patch to describe a replacement of API. The second stage modmes the generic patch to preserve the context of the API invocation in the client code. Our experiment using a dataset that provides real-world Java codes shows that AUTOMIG synthesized a correct patch for 92.2% of the code on average. Our manual investigation on synthesized patches shows that AUTOMIG can synthesize highquality patches that preserve the context of API invocation of each client code.