Detecting Higher-Order Merge Conflicts in Large Software Projects

Abstract

Merge conflicts can occur when multiple developers work concurrently on the same source code corpus. Diverging textual changes (in the same lines of code) are typically easy to resolve with help of common tools like Git. More challenging are higher-order merge conflicts. They arise as the result of unintended interactions between changes in different parts of the source code. Higher-order merge conflicts can be caused by a combination of changes, and so even thorough testing of the individual development branches might not be able to identify them. We suggest an approach based on static analysis and a prototypical tool to detect potential higher-order merge conflicts. Our method identifies potentially dangerous dependencies between changed code fragments in a call graph. An evaluation on SAP HANA, a very large industrial product in C++, shows that the approach is able to identify 62% of higher-order merge conflicts causing build failures over 22 months project development time. The same prototype finds no instance of higher-order merge conflicts causing test failures in SAP HANA during a two month development period. In summary, our method scales well and can identify higher-order merge conflicts which escape traditional testing.