balazske added a comment. In D152436#4438956 <https://reviews.llvm.org/D152436#4438956>, @NoQ wrote:
> Uh-oh, looks like I'm not paying nearly enough attention to this discussion > (sorry about that!!) > > I'm somewhat skeptical of the decision made in D151225 > <https://reviews.llvm.org/D151225> because the entire reason I originally > implemented `StdCLibraryFunctions` was to deal with false positives I was > seeing. It was really valuable even without the bug-finding part. So I really > wish we could find some way to keep bug-finding and modeling separate. > > I haven't read the entire discussion though, I need to catch up 😓 The problem was that modeling and report generation could not be separated correctly. Both are implemented in one class but are differently named checkers that should run in a specific order because dependency issues, this was not good. Other problem was that if the modeling checker runs first, it will apply state changes for pre and post conditions without generating a bug report even if a bug could be found in the previous state. The old state is then lost and other checkers will not find that bug. For example a case of null pointer argument to a function is always removed by the modeling part of the checker, even if this was a case when a bug report should be generated. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D152436/new/ https://reviews.llvm.org/D152436 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
