Need to use function "getAsCXXRecordDecl" of ASTMatchFinder.cpp in a clang-tidy check

[Richard Smith via cfe-commits]

On Tue, Feb 9, 2016 at 4:15 PM, Cong Liu <congliu at google.com> wrote:
> Hi Richard,
>
> Thank you for your reply. Yes, the case I need to deal with is like what you
> said:
>>
>> If you want to make the assumption that the primary template will be
>> used for an unknown specialization, you'll need something like that
>> function in ASTMatchFinder.
>
>
> For example,
>
>   template <typename T>
>   struct Base {};
>   template <typename T>
>   struct Derived : Base<T>{};
>
>   Derived<int> T1;
>
> In this case, I need to firstly get the CXXBaseSpecifier from line 4, then
> get the QualType of primary template (Base<T>), then get its declaration.
> For this case, that function in ASTMatchFinder works but
> Type::getAsCXXRecordDecl does not.
>
> So, what do you suggest?

Using that may not be correct when analysing virtual function
overrides. Consider this:

template<typename T> struct Base {
  virtual void f(T);
};
template<> struct Base<void> {
  virtual void f();
};
template<typename T, bool = is_same<T, void>::value> struct Derived : Base<T> {
  virtual void f();
};
template<typename T> struct Derived<T, false> : Base<T> {
  virtual void f(T);
};

Here, it would be wrong to report that the Derived primary template
fails to override a virtual function from the Base primary template,
as the Derived primary template is actually only ever used when T ==
void, and there's a specialization of the Base template for that case.
I have no idea whether that's acceptable for your check.

In principle, I'm fine with us moving the functionality in
ASTMatchFinder (that I recently renamed to
getAsCXXRecordDeclOrPrimaryTemplate to better express its purpose) to
somewhere more central, but my concern is that most uses of it will in
fact be subtle bugs.