Despite being true positives, these results just confuse users. So
filter them out.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
clang-tools-extra/include-cleaner/lib/Analysis.cpp | ||
---|---|---|
37 | shouldIgnoreMacroReference would better distinguish vs macro definition | |
43 | Comment nit: I'm having trouble imagining cases that are actually*unrelated* to the stdlib. If stderr is an impl-defined macro, then the only way to use the name to refer to something else is if it's not defined (inside ifndef stderr, or if you can be sure your TU doesn't include stdio first). Seems implausible... That said I feel like we've had this conversation before and I've just forgotten the details. | |
44 | Nit: "false positives" is a little unclear: positives for this function are negatives for walkAST and could be either for diagnostics. Also, I think you mean *references* rather than declarations? |
clang-tools-extra/include-cleaner/lib/Analysis.cpp | ||
---|---|---|
43 |
That's absolutely right. The issue here is macro expansion triggers independent of the context, e.g. #include <stdio.h> namespace ns { void stderr(); } void foo() { ns::stderr(); } here we have a (well two) reference to stderr macro from stdandard library, which is not user's intent, but rather a quirk of the language. | |
44 | this should've been "false negatives" from the "usedness perspective", we basically drop the reference to ns::stderr in above example, because the name is not spelled in the main file (but rather spelled inside a macro body). |
still LG, comments are still confusing me a little
clang-tools-extra/include-cleaner/lib/Analysis.cpp | ||
---|---|---|
43 | Sure, but this code is not valid C++ (since stderr is not guaranteed to expand to a single identifier). Is this actually a common/motivating case? | |
44 | And now it's backwards if your perspective is this function itself! just avoid "false positives"/"false negatives" terminology and describe the actual effect? |
clang-tools-extra/include-cleaner/lib/Analysis.cpp | ||
---|---|---|
43 |
That's true. Most of the standard library implementations I checked does that, but I also don't like that reliance, so making it a little bit more generic (which meant some more plumbing, sorry for the big diff :( ), by making sure that we apply this to all the macros that expand to themselves. |
clang-tools-extra/include-cleaner/lib/Analysis.cpp | ||
---|---|---|
42 | still call out stdout/stderr? |
shouldIgnoreMacroReference would better distinguish vs macro definition