NagyDonat wrote: > I wonder if we could have something in between. I'm thinking of having the > concrete offset as a separate note, instead of having it part of the primary > message. That way after BR selection, we would still deterministically pick > the shortest parh, and also have the offset in the path notes.
Actually I had a better solution for achieving this: during my earlier experimentation I noticed that the `Profile()` methods of the `BugReporter` use the (long) `Description` field (which is "Access of `<region>` at negative byte offset -124", the final note on the bug path), but ignore the `ShortDescription` field (which is "Out of bound access to memory preceding `<region>`", the stand-alone warning message). In the commit that I just pushed, I tweaked `Profile()` to ensure that it uses the short description and ignores the long one -- which yields the right deduplication behavior. I'd guess that `ShortDescription` was missing from the `Profile()` method because it was introduced at a later point in the development, and there are very few checkers that actually set different short and long descriptions. (For more details see the commit message of the commit that I just pushed onto this PR.) https://github.com/llvm/llvm-project/pull/98621 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
