[PATCH] D101097: [Sema] Don't set BlockDecl's DoesNotEscape bit If the block is being passed to a function taking a reference parameter

[Akira Hatanaka via Phabricator via cfe-commits]

ahatanak added inline comments.


================
Comment at: clang/lib/Sema/SemaExpr.cpp:5917
         if (auto *BE = dyn_cast<BlockExpr>(Arg->IgnoreParenNoopCasts(Context)))
           BE->getBlockDecl()->setDoesNotEscape();
 
----------------
rjmccall wrote:
> We need to be checking that the parameter type is a block pointer type.  A parameter of a type like `id` or `void*` does not have the enhanced semantics of `noescape` for blocks.
> 
> The inevitable weird C++ test case is:
> 
> ```
> struct NoescapeCtor {
>   NoescapeCtor(__attribute__((noescape)) void (^)());
> };
> struct EscapeCtor {
>   EscapeCtor(void (^)());
> };
> 
> void helper1(NoescapeCtor a);
> void test1() { helper1(^{}); } // <- should be noescape
> 
> void helper2(NoescapeCtor &&a);
> void test2() { helper2(^{}); } // <- should be noescape
> 
> void helper3(__attribute__((noescape)) EscapeCtor &&a);
> void test3() { helper3(^{}); } // <- should not be noescape
> ```
> 
> You should probably also test that calls to function templates behave according to the instantiated type of the parameter.  I expect that that should just fall out from this implementation, which I think only triggers on non-dependent calls.
I understand why the blocks should or shouldn't be `noescape` in the C++ example, but I'm not sure I understand the comment about `id` and `void*`.

Do you mean the `DoesNotEscape` bit shouldn't be set in the following example?

```
void helper(__attribute__((noescape)) id);

void test() {
  S s;
  helper(^{ (void)s; });
}
```


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D101097/new/

https://reviews.llvm.org/D101097