[PATCH] D116512: [clang-tidy] Limit non-Strict mode to public functions

[Mehdi AMINI via Phabricator via cfe-commits]

mehdi_amini abandoned this revision.
mehdi_amini added inline comments.


================
Comment at: clang-tools-extra/docs/clang-tidy/checks/misc-unused-parameters.rst:43
+   a human reader, and there's basically no place for a bug to hide. On the other
+   hand for non-public functions, all the call-sites are visible and the parameter
+   can be eliminated entirely.
----------------
aaron.ballman wrote:
> mehdi_amini wrote:
> > aaron.ballman wrote:
> > > Call sites are not always visible for protected functions, so this seems a bit suspicious. The changes are missing test coverage for that situation.
> > You're using `public` for "access control" while I was using the linkage aspect: my reasoning is that if a method isn't "externally visible" from the current translation unit, you see all the call-sites. This is orthogonal to public/private/protected as far as I know.
> > 
> > I am likely missing a check for "isDefinedInMainFile" (or whatever the api is called) to not flag included headers.
> > You're using public for "access control" while I was using the linkage aspect: my reasoning is that if a method isn't "externally visible" from the current translation unit, you see all the call-sites.
> 
> Oh, thanks for pointing out that I was confused! I'm not used to "public" when describing linkage, usually that's "external" or "externally visible". Any appetite for rewording in those terms? Something like "On the other hand, for functions with internal linkage, all the call sites are visible and parameters can be safely removed."
Sure: happy to reword.

(We use public/private for symbol visibility at a module level in MLIR unfortunately, I've been "contaminated" ;) ).


================
Comment at: clang-tools-extra/test/clang-tidy/checkers/misc-unused-parameters.cpp:157-159
+// CHECK-FIXES: C() {}
+  C(int i) {}
+// CHECK-MESSAGES: :[[@LINE-1]]:9: warning
----------------
aaron.ballman wrote:
> mehdi_amini wrote:
> > aaron.ballman wrote:
> > > I think this demonstrates a bad fix -- this changes code meaning from being a converting constructor to being a default constructor, which are very different things.
> > Oh you're right: so we can't do it for a Ctor with a single parameter...
> > 
> > But we also can't do it for a Ctor with two parameters as it'll turn it into a converting ctor. Unless you can eliminate both parameters, in which case it is become a default ctor (which can conflict with an existing one, in which case it can be just deleted?)
> Yeah, I think we need to not transform ctors at all currently.
In this case I can fix the existing bug by disabling the fixit  (as discussed in D116513) and abandon this revision I think.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D116512/new/

https://reviews.llvm.org/D116512