clang-format 19-rc1 regression: member function pointers as template arguments

[vogelsgesang]

In our internal code base, we have a function templated with a member function pointer.
Upgrading from clang-format-17 to 19.1.0-rc1, clang-format proposed the following change, which is incorrect / unexpected

-template <llvm::Value* (llvm::IRBuilderBase::*func)(llvm::Value*, llvm::Value*, const llvm::Twine&, bool, bool)>
+template <llvm::Value* (llvm::IRBuilderBase::* func)(llvm::Value*, llvm::Value*, const llvm::Twine&, bool, bool)>
 llvm::Value* LLVMTranslator::generateArithmetic(const IRBinaryOp& op) {

[llvmbot]

@llvm/issue-subscribers-clang-format

Author: Adrian Vogelsgesang (vogelsgesang)

In our internal code base, we have a function templated with a member function pointer. Upgrading from clang-format-17 to 19.1.0-rc1, clang-format proposed the following change, which is incorrect / unexpected

-template <llvm::Value* (llvm::IRBuilderBase::*func)(llvm::Value*, llvm::Value*, const llvm::Twine&, bool, bool)>
+template <llvm::Value* (llvm::IRBuilderBase::* func)(llvm::Value*, llvm::Value*, const llvm::Twine&, bool, bool)>
 llvm::Value* LLVMTranslator::generateArithmetic(const IRBinaryOp& op) {

[rymiel]

Bisected to 579dc7f. This looks intentional, though

[owenca]

Bisected to 579dc7f. This looks intentional, though

Yeah, IMO it's a bug fix rather than a regression.

[tru]

Can this be closed or removed from the LLVM 19.x milestone?

[vogelsgesang]

Yes, given this was an intentional change, let's close it