[PATCH] D50619: [clang-tidy] Handle unresolved expressions in ExprMutationAnalyzer

[Jonas Toth via Phabricator via cfe-commits]

JonasToth added a comment.

You are totally right.

Am 16.08.2018 um 02:41 schrieb Shuai Wang via Phabricator:

> shuaiwang added inline comments.
> 
> ================
>  Comment at: unittests/clang-tidy/ExprMutationAnalyzerTest.cpp:309
> 
> +TEST(ExprMutationAnalyzerTest, CallUnresolved) {
>  +  auto AST =
> 
>  ----------------
> 
> JonasToth wrote:
> 
>> I think we are missing tests for non-type template paramters (`template <size_t N>`). They should behave the same. But the following case would not be a mutation:
>> 
>>   void non_mutating(const char* Array, int size) { /* Foo */ }
>>   template <int N>
>>   struct ArrayLike {
>>     char* data[N]; // Initialized to something
>>     void SomeFunction() {
>>       non_mutating(data, N);
>>     }
>>   };
>> 
>> 
>> The difference between the 'normal' and non-type templates would be, that `N` is not mutatable at all and the semantics is clear (builtin integer-like type).
>> 
>> If the current implementation would not figure that out, you can just add a test for it and assume a mutation. Handling non-type templates later is absolutly ok.
> 
> We have to assume `data` is mutated here as well. I'll add a test case for this.
> 
>   void g(const char*, int); // <-- doesn't mutate
>   void g(char(&)[8], int); // <-- do mutate
>   
>   template <int N>
>   void f() {
>       char data[N];
>       g(data, N); // <-- we don't know which `g` will be called yet
>   }
>   
>   void h() {
>       f<8>(); // <-- f calls g(char(&)[8], int) internally
>       f<9>(); // <-- f calls g(const char*, int) internally
>   }
> 
> 
> Repository:
> 
>   rCTE Clang Tools Extra
> 
> https://reviews.llvm.org/D50619


Repository:
  rCTE Clang Tools Extra

https://reviews.llvm.org/D50619