balazs-benics-sonarsource wrote: > What is the relationship of this new callback with the `BranchCondition` > callback which is evaluated in `ProcessBranch` where the checker splits the > execution path into multiple branches? It would be nice if you could document > the difference between these two callbacks.
Great question. They are for different purposes, but one should prefer `BranchCondition` over the strictly more generic `BlockEntrance` callback. State splits work as in any other checkers. If you split the state in both callback, you will end up with 4 splits for the cases that are `if`, `for`, `while`, statements with conditions and 2 paths for every other control-flow edge, like entering the first basic block from the Entry CFG node, or following a `break`, `continue` or `goto` (unconditional) CFG edge. Documented this now in c2bd657065e0edcac742941c1a535444a90b1638, and added a unit test for the case when both callbacks are defined in 910abb384af9c8d51f32d2109a73ab900682957e. > > What is the relationship of this new callback with the BranchCondition > > callback > > +1, I am also interested to learn what is the main motivation behind the new > callback. Do you have some example use cases in mind? I think users might get > confused which one to use. My motivation is to drive enabling or disabling an internal checker by entering or leaving a certain CFG block. Concretely, if a short-circuiting operator leads to the guarded CFG block (aka. the true branch for the `&&` operator and the false branch for the `||` operator), enable a checker to detect certain constructs in a guarded block. And once we are leaving the guarded block disable the checker. I hope it clarifies the motivation and demonstrates the usefulness of such a callback. I believe that there could be other similar use cases of this. https://github.com/llvm/llvm-project/pull/140924 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
