[clang-tidy] false positive for modernize-type-traits with structured bindings

[matthew-f]

Hi

Using an llvm update from a few days modernize-type-traits is generating what looks like a false positive. An update from today still exhibits the same problem

std::map<int, int> map = {{1, 2}, {3, 4}};

for (const auto& [a, b] : map)
	{
		std::cout << a << b;
	}

generates

main.cpp:36:20: error: use c++14 style type templates [modernize-type-traits,-warnings-as-errors]
   36 |         for (const auto& [a, b] : map)
      |                           ^
      |                           _t
main.cpp:36:23: error: use c++14 style type templates [modernize-type-traits,-warnings-as-errors]
   36 |         for (const auto& [a, b] : map)
      |                              ^
      |                              _t

[llvmbot]

@llvm/issue-subscribers-clang-tidy

Author: None (matthew-f)

Hi

Using an llvm update from a few days modernize-type-traits is generating what looks like a false positve.

std::map<int, int> map = {{1, 2}, {3, 4}};

for (const auto& [a, b] : map)
	{
		std::cout << a << b;
	}

generates

main.cpp:36:20: error: use c++14 style type templates [modernize-type-traits,-warnings-as-errors]
   36 |         for (const auto& [a, b] : map)
      |                           ^
      |                           _t
main.cpp:36:23: error: use c++14 style type templates [modernize-type-traits,-warnings-as-errors]
   36 |         for (const auto& [a, b] : map)
      |                              ^
      |                              _t

[vbvictor]

https://godbolt.org/z/9c7Gab8oc
This is odd TBH, type-traits seems unrelated to structured bindings.

[vbvictor]

This bug gave 63 FPs on clang-tools-extra/clang-tidy which is… very unfortunate.
Need to look into it.

[vbvictor]

This must have broke it #147835.

[mizvekov]

What looks bizarre is that this is using a matcher on TypeLocs, but there shouldn't be a TypeLoc in there, the user didn't write any types at that point. This could be a fake TypeLoc from some misuse / workaround.

[mizvekov]

Another point here is that the matcher didn't previously handle a NestedNameSpecifier of identifier kind.

Now after #147835 these kinds of NestedNameSpecifiers are represented as DependentNameTypes, which this matcher handles.