Suppress the clang warning "-Wdeprecated-declarations" for CSSM API calls.

crypto/cssm_init.h

Index: crypto/cssm_init.h
diff --git a/crypto/cssm_init.h b/crypto/cssm_init.h
index 3eece1b9c3be12a788a2d2603614782b01d23703..87525d0c6819207ce681e4743ba0e52fab0a4194 100644
--- a/crypto/cssm_init.h
+++ b/crypto/cssm_init.h
@@ -46,11 +46,16 @@ class ScopedCSSMData {
  public:
   ScopedCSSMData();
   ~ScopedCSSMData();
+// CSSM functions are deprecated as of OSX 10.7, but have no replacement.

[Ryan Sleevi, 2016/03/01 15:59:58]

A newline between 48/49 would be helpful for readability.

More generally, though, why doesn't this deprecation wrapping need to be moved
to line 19/20? The CSSM_CSP_HANDLE / CSSM_API_MEMORY_FUNCS / CSSM_RETURN are all
part of the same deprecated state. I'm assuming it's because OS X doesn't
support deprecating typedefs?

[erikchen, 2016/03/02 02:02:59]

It looks like C++14 introduces a deprecated attribute, which Clang has already
implemented. I wouldn't be surprised if Apple deprecated the typedefs sometime
in the future.
http://josephmansfield.uk/articles/marking-deprecated-c++14.html

I went ahead and moved the wrapping to line 14, which includes one extra
function, but happens to be immediately after a namespace declaration. I also
moved the #pragma pop right before the end of the namespace scope. This seems
cleaner to me. (Given that the scope of the #clang diagnostics is entirely
independent of C++ semantic scoping, it still seems reasonable to attempt to
match the two together).

+// https://bugs.chromium.org/p/chromium/issues/detail?id=590914#c1
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
   operator CSSM_DATA*() { return &data_; }
   CSSM_DATA* operator ->() { return &data_; }
 
  private:
   CSSM_DATA data_;
+#pragma clang diagnostic pop
 
   DISALLOW_COPY_AND_ASSIGN(ScopedCSSMData);
 };