Structured Control-Flow

We obtain from rustc function bodies that have unstructured control-flow (see the chapter explaining the difference). We reconstruct structured control-flow ourselves by analyzing the control-flow graph.

rustc’s MIR records no information about the original loops/ifs/matchs or even blocks, so we only have the control-flow graph to start from. For that reason, the resulting structured control-flow may look different from the original Rust code. It will be semantically equivalent however.

We detect loops by identifying backedges in a postorder traversal of the graph, and based on that we need two more things: loop exits (which are where a break out of the loop will be), and switch exits (where the arms of a if/match join back together). We detect these using heuristics. These heuristics don’t affect correctness, only how pretty the output looks.

By default we sometimes duplicate code to make the output legible. To avoid that, you may pass --no-code-duplication. This changes reconstruction to instead use an algorithm based on the Beyond Relooper paper, which guarantees no duplication by using loop { if foo() { break }; ...; break } to get “forward jumps”.