[clang-format] regression from clang-format v13

[paulhoad]

The following source

namespace A {

template <int N> struct Foo<char[N]> {
  void foo() { std::cout << "Bar"; }
}; // namespace A

is formatted as:

namespace A {

template <int N> struct Foo<char[N]>{void foo(){std::cout << "Bar";
}
}
; // namespace A

using clang-format version 15.0.0 (https://github.com/llvm/llvm-project.git ee4ba9f
82397c60e17f14823eb22b2e5831c7)

Using the following .clang-format file

BasedOnStyle: LLVM

[llvmbot]

@llvm/issue-subscribers-clang-format

[llvmbot]

@llvm/issue-subscribers-bug

[mydeveloperday]

possibly caused by the presence of

[N]

given

namespace A {

template <int N> struct Foo<char[N]>{void foo(){std::cout << "Bar";
}
}
; // namespace A

namespace B {

template <int N> struct Foo<char> {
  void foo() { std::cout << "Bar"; }
}; // namespace A

[mkurdej]

With clang-format from 2022-01-05, commit 5109737:

clang-format-main-20220105-T1352-5109737c.exe --style=file test.cpp
namespace A {

template <int N> struct Foo<char[N]> {
  void foo() { std::cout << "Bar"; }
};

} // namespace A

And with clang-format from 2022-01-10, commit 5c2e7c9:

clang-format-main-20220110-5c2e7c9ca043.exe --style=file test.cpp
namespace A {

template <int N> struct Foo<char[N]>{void foo(){std::cout << "Bar";
}
}
;

} // namespace A

I think it's the commit 5c2e7c9 that provoked that.

[mkurdej]

Commit 72e29ca also fixed a regression on it.
I think that the solution would be to call tryToParseLambda only if the previous token is < (template opener) or , (template argument comma) when we are in the template context.

[mydeveloperday]

How did you get to the commit so quickly! (are you keeping a copy of the .exe at every commit?)

[mydeveloperday]

https://reviews.llvm.org/D118911