Work around problems with isCXX11ConstantExpr in template code

> template<size_t Size>
> bool checkOutput(ScDocument* pDoc, const ScRange& aOutRange, const char* aOutputCheck[][Size], const char* pCaption)
> {
>     ...
>             const char* p = aOutputCheck[nRow][nCol];

in sc/qa/unit/helper/qahelper.hxx caused

>    assert(E->isRValue() && E->getType()->hasPointerRepresentation());

in Clang's EvaluatePointer (lib/AST/ExprConstant.cpp) to fire.  In the template
definition itself, Clang doesn't introduce ArrayToPointerDecay ImplicitCastExpr
into the subscripting operations (while in any implicit specializations that it
instantiates, it does).

This is interesting:  Up to C++11, [expr.sub] requires the operator to have
pointer type (so array-to-pointer decay is clearly asked for).  In C++14
(CWG1213), the operator can also be of array type but it is not explicitly
specified whether array-to-pointer decay is to be performed.  In upcoming C++17
(P0135R1), it specifies further that an operator of array type must be a glvalue
but still does not explicitly specify whether array-to-pointer decay is to be
performed.  Maybe the definition of the subscripting operation in terms of
*((E1)+(E2)) is meant to imply that, however.

Change-Id: I67c7b0f34002387dbf746288630371877c6261ef

compilerplugins/clang/conststringvar.cxx

@@ -104,6 +104,10 @@ public:
         if (init == nullptr) {
             return true;
         }
+        if (init->isInstantiationDependent()) {
+            // avoid problems with isCXX11ConstantExpr in template code
+            return true;
+        }
         APValue v;
         if (!init->isCXX11ConstantExpr(compiler.getASTContext(), &v)) {
             return true;