feat(matching): use the Hungarian Algorithm for unordered matching

[jpedroh]

Our current approach to unordered node matching relies on a naive assumption: that all nodes possess an identifier. While this holds true for most nodes we've encountered thus far, such as method and property declarations within a Java class, it proves insufficient when attempting to match nodes lacking a label, like static blocks in Java. In such cases, calculations for matchings may yield incorrect results, consequently leading to erroneous merges.

This pull request introduces a solution for matching unordered nodes via the Assignment Problem, utilizing the Hungarian Algorithm to resolve it. This approach mirrors the one used in jDime.

Given the widespread recognition of the Hungarian Algorithm, we rely on the implementation provided by the pathfinding crate. This simplifies our implementation efforts, as we only need to provide the weights matrix and extract the matching information from the solution.

A workaround had to be implemented since pathfinding expects the input matrix weight to have the same number of rows and columns, which might not always be true in our case since nodes can have a different number of children. The solution involves initializing the remaining columns/rows with 0.

For now, our naive label implementation has been bypassed and is not being utilized. In a further pull request, the idea is to resort to the Hungarian algorithm only if the nodes are unlabeled, as it's significantly more complex than merely matching identifiers.

[codesandbox]

Review or Edit in CodeSandbox

Open the branch in Web Editor • VS Code • Insiders

Open Preview

[coveralls]

coverage: 86.126% (-0.3%) from 86.418%
when pulling eb9fe81 on feat-assignment-problem
into 31ea3c8 on main.