[cfe-dev] LValueExprEvaluator::VisitArraySubscriptExpr assertion

[Stephan Bergmann]

On 03/25/2015 06:22 PM, Richard Smith wrote:
> On Wed, Mar 25, 2015 at 2:48 AM, Stephan Bergmann <sbergman at redhat.com
> <mailto:sbergman at redhat.com>> wrote:
>
>     In LibreOffice we have a Clang plugin calling
>     Expr::__EvaluateAsBooleanCondition, and at least with a recent trunk
>     Clang build with assertions enabled that triggers the assert in
>
>         static bool EvaluatePointer(const Expr* E, LValue& Result,
>         EvalInfo &Info) {
>            assert(E->isRValue() &&
>         E->getType()->__hasPointerRepresentation());
>            return PointerExprEvaluator(Info, Result).Visit(E);
>         }
>
>
>     (in lib/AST/ExprConstant.cpp) on expressions like
>
>        m_Strings[nEntry]
>
>     where m_Strings is of std::vector type.
>
>     What happens is that Expr::__EvaluateAsBooleanCondition goes into
>     LValueExprEvaluator::__VisitArraySubscriptExpr, which calls
>     EvaluatePointer, which asserts because m_Strings doesn't have array
>     or pointer type.
>
>     It's not clear to me where the error lies; should
>     LValueExprEvaluator::__VisitArraySubscriptExpr even be reached for
>     uses of user-defined operator[] overloads?
>
>
> No. Are you calling EvaluateAsBooleanCondition on a value-dependent
> expression, perhaps? We should probably add an assert for that.

Turned out to be called on an Expr for which isTypeDependent() is true 
(and the problem went away after changing the plugin to no longer call 
EvaluateAsBooleanCondition on any type-dependent expressions).

However, I must admit that even after consulting [temp.dep.expr] it is 
still unclear to me why Clang considers the expression s[0] in

> struct S { void operator [](int); };
> template<typename> struct T {
>     S s;
>     void f() { s[0]; }
> };

as type-dependent.