r315811 - Re-land r315787, "[Sema] Warn about unused variables if we can constant evaluate the initializer."

[Benjamin Kramer via cfe-commits]

We should check VD->getInit()->isValueDependent() before we call
evaluateValue. I wasn't able to come up with a test case that triggers
the assert though :(

On Thu, Oct 19, 2017 at 5:45 PM, Alexander Kornienko <alexfh at google.com> wrote:
>
>
> On Sat, Oct 14, 2017 at 5:59 PM, Benjamin Kramer via cfe-commits
> <cfe-commits at lists.llvm.org> wrote:
>>
>> Author: d0k
>> Date: Sat Oct 14 08:59:34 2017
>> New Revision: 315811
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=315811&view=rev
>> Log:
>> Re-land r315787, "[Sema] Warn about unused variables if we can constant
>> evaluate the initializer."
>>
>> The warnings in libc++ tests were fixed in the meantime.
>>
>> Modified:
>>     cfe/trunk/lib/Sema/SemaDecl.cpp
>>     cfe/trunk/test/SemaCXX/warn-unused-variables.cpp
>>
>> Modified: cfe/trunk/lib/Sema/SemaDecl.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Sema/SemaDecl.cpp?rev=315811&r1=315810&r2=315811&view=diff
>>
>> ==============================================================================
>> --- cfe/trunk/lib/Sema/SemaDecl.cpp (original)
>> +++ cfe/trunk/lib/Sema/SemaDecl.cpp Sat Oct 14 08:59:34 2017
>> @@ -1723,7 +1723,8 @@ static bool ShouldDiagnoseUnusedDecl(con
>>              dyn_cast<CXXConstructExpr>(Init);
>>            if (Construct && !Construct->isElidable()) {
>>              CXXConstructorDecl *CD = Construct->getConstructor();
>> -            if (!CD->isTrivial() && !RD->hasAttr<WarnUnusedAttr>())
>> +            if (!CD->isTrivial() && !RD->hasAttr<WarnUnusedAttr>() &&
>> +                !VD->evaluateValue())
>
>
> The evaluateValue call above causes an assertion failure on
> instantiation-dependent values:
> llvm/tools/clang/lib/AST/Decl.cpp:2196 in clang::APValue
> *clang::VarDecl::evaluateValue(SmallVectorImpl<clang::PartialDiagnosticAt>
> &) const: !Init->isValueDependent()
>
> I'm not sure why evaluateValue uses an assertion instead of outputting a
> note and returning nullptr, but the assertion can be avoided on the caller
> site as well.
>
> Working on a reduced test case...
>
>>
>>                return false;
>>            }
>>          }
>>
>> Modified: cfe/trunk/test/SemaCXX/warn-unused-variables.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/cfe/trunk/test/SemaCXX/warn-unused-variables.cpp?rev=315811&r1=315810&r2=315811&view=diff
>>
>> ==============================================================================
>> --- cfe/trunk/test/SemaCXX/warn-unused-variables.cpp (original)
>> +++ cfe/trunk/test/SemaCXX/warn-unused-variables.cpp Sat Oct 14 08:59:34
>> 2017
>> @@ -1,4 +1,5 @@
>>  // RUN: %clang_cc1 -fsyntax-only -Wunused-variable -Wunused-label
>> -Wno-c++1y-extensions -verify %s
>> +// RUN: %clang_cc1 -fsyntax-only -Wunused-variable -Wunused-label
>> -Wno-c++1y-extensions -verify -std=c++11 %s
>>  template<typename T> void f() {
>>    T t;
>>    t = 17;
>> @@ -194,3 +195,35 @@ void test() {
>>  }
>>
>>  }
>> +
>> +#if __cplusplus >= 201103L
>> +namespace with_constexpr {
>> +template <typename T>
>> +struct Literal {
>> +  T i;
>> +  Literal() = default;
>> +  constexpr Literal(T i) : i(i) {}
>> +};
>> +
>> +struct NoLiteral {
>> +  int i;
>> +  NoLiteral() = default;
>> +  constexpr NoLiteral(int i) : i(i) {}
>> +  ~NoLiteral() {}
>> +};
>> +
>> +static Literal<int> gl1;          // expected-warning {{unused variable
>> 'gl1'}}
>> +static Literal<int> gl2(1);       // expected-warning {{unused variable
>> 'gl2'}}
>> +static const Literal<int> gl3(0); // expected-warning {{unused variable
>> 'gl3'}}
>> +
>> +template <typename T>
>> +void test(int i) {
>> +  Literal<int> l1;     // expected-warning {{unused variable 'l1'}}
>> +  Literal<int> l2(42); // expected-warning {{unused variable 'l2'}}
>> +  Literal<int> l3(i);  // no-warning
>> +  Literal<T> l4(0);    // no-warning
>> +  NoLiteral nl1;       // no-warning
>> +  NoLiteral nl2(42);   // no-warning
>> +}
>> +}
>> +#endif
>>
>>
>> _______________________________________________
>> cfe-commits mailing list
>> cfe-commits at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
>
>