Update tests for conditional rvalue bounds #422
Conversation
| @@ -514,7 +514,7 @@ extern void check_condexpr(int val) { | |||
| array_ptr<int> t5 = val ? r : 0; | |||
| array_ptr<int> t5a = val ? v : 0; | |||
| array_ptr<int> t6 = val ? 0 : r; | |||
| nt_array_ptr<int> t6a = val ? 0 : v; | |||
| nt_array_ptr<int> t6a = val ? 0 : v; // expected-warning {{cannot prove declared bounds for 't6a' are valid after initialization}} | |||
There was a problem hiding this comment.
I am surprised by this. I would have expected 0 to have bounds(any), the bounds for v to be count(5), and the overall bounds for the two arms to be count(5). What are the actual bounds the compiler is inferring for the two arms?
There was a problem hiding this comment.
The compiler infers bounds(any) for 0, bounds(v, v + 5) for v, and bounds(v, v + 5) as the overall bounds for the two arms. v and t6a are not known to be equal, so the compiler cannot prove that bounds(v, v + 5) imply bounds(t6a, t6a + 0).
There was a problem hiding this comment.
Created checkedc-clang/934 to track this follow-up work.
There was a problem hiding this comment.
Thanks for the explanation and opening the issue to track this. I agree that this should be handled as a separate follow-up issue.
This PR adds expected warnings and errors to tests that result from inferring rvalue bounds for conditional expressions (in checkedc-clang/924).