FunctionDecl::getBody() returning nullptr

[Aaron Ballman via cfe-commits]

On Wed, Aug 19, 2015 at 11:33 AM, Aaron Ballman <aaron at aaronballman.com> wrote:
> When I run the following test code through clang-tidy -checks=*, I get
> a crash from some relatively strange behavior with FunctionDecl.
>
> template <class T> struct remove_reference      {typedef T type;};
> template <class T> struct remove_reference<T&>  {typedef T type;};
> template <class T> struct remove_reference<T&&> {typedef T type;};
>
> template <typename T>
> typename remove_reference<T>::type&& move(T&& arg) {
>   return static_cast<typename remove_reference<T>::type&&>(arg);
> }
>
> AnalysisConsumer::getModeForDecl() is called, and it has code that
> does: D->hasBody() ? D->getBody()->stuff : stuff;
>
> What's strange is that hasBody() returns true, but getBody() returns
> nullptr. I don't think that this should be possible. I'm wondering
> whether it's purposeful that hasBody() does not check
> Definition->Body?

Looking into this a bit further, the issue is that hasBody() checks
for Definition->Body *or* Definition->IsLateTemplateParsed when
deciding to return true. In my case, Body is null, but
IsLateTemplateParsed is true, so hasBody() returns true. However,
getBody() doesn't have any special logic for late template parsed
function bodies.

> Also, the FunctionDecl in question is for move(), which does have a
> body, so I do not understand why the definition would claim there is
> no body.
>
> Ideas, or should I file a bug report?

>From my further look, I wonder if the correct solution to this is to
simply call getBody() and handle a nullptr return instead of calling
hasBody() first. It basically makes the check to ignore late parsed
template bodies an implicit one.

~Aaron