[clang] [Sema] Don't call isNonConstantStorage on incomplete variable types (PR #161590)

[Erich Keane via cfe-commits]

================
@@ -14919,7 +14919,7 @@ void Sema::CheckCompleteVariableDeclaration(VarDecl *var) {
 
   // Apply section attributes and pragmas to global variables.
   if (GlobalStorage && var->isThisDeclarationADefinition() &&
-      !inTemplateInstantiation()) {
+      !var->getDeclContext()->isDependentContext()) {
----------------
erichkeane wrote:

Checking JUST the actual var decl is actually going to be a 'smaller' thing, but I'm not sure it is actually going to work?  @ahatanak : please try it though, it is quite a bit easier.

The difference here is that if we switch to ONLY a dependent type (instead of a dependent decl), we end up in a situation where we might try to do this for a non-dependently typed variable in a dependent context.  So I don't think that'll fix it?

Consider the example: 

```
template <class T>
struct A {
  struct B;
  static constexpr B b{nullptr}; // This used to crash.
};
```

IN this case, `B` is a dependent type, since it is inside the struct (so it is `A<T>::B`.  However consider this slight modification (please add this test!).


```
  struct B;
template <class T>
struct A {

  static constexpr B b{nullptr}; // This used to crash.
};
```

Note I moved `B` out of the template.  Now `b` is still going to go down this path (I think?) but is no longer a dependent type, despite being a dependent context.  I think this is wrong? 



https://github.com/llvm/llvm-project/pull/161590