[webcrypto] Implement structured clone of keys (blink-side).

public/platform/WebCryptoKey.h

Index: public/platform/WebCryptoKey.h
diff --git a/public/platform/WebCryptoKey.h b/public/platform/WebCryptoKey.h
index 8a48e6b8902a9d21ab82bd62b7d71db8c208be81..e1b12a36e9e89b9839f86dbaf79c87249850567c 100644
--- a/public/platform/WebCryptoKey.h
+++ b/public/platform/WebCryptoKey.h
@@ -33,6 +33,7 @@
 
 #include "WebCommon.h"
 #include "WebPrivatePtr.h"
+#include "WebVector.h"
 
 namespace blink {
 
@@ -43,13 +44,13 @@ enum WebCryptoKeyType {
 };
 
 enum WebCryptoKeyUsage {
-    WebCryptoKeyUsageEncrypt = 1 << 0,
-    WebCryptoKeyUsageDecrypt = 1 << 1,
-    WebCryptoKeyUsageSign = 1 << 2,
-    WebCryptoKeyUsageVerify = 1 << 3,
-    WebCryptoKeyUsageDeriveKey = 1 << 4,
-    WebCryptoKeyUsageWrapKey = 1 << 5,
-    WebCryptoKeyUsageUnwrapKey = 1 << 6,
+    WebCryptoKeyUsageEncrypt = 1 << 1,

[jsbell, 2014/03/13 20:16:07]

Since this isn't directly serialized, there's no need to bump these up to make
room for Extractability. Is there any other reason to change them?

[eroman, 2014/03/14 18:28:09]

Correct there is no direct need to bump it.

I can remove this change if you prefer, the reason I did it was fear.

I added some compile asserts to make sure that the serialization routine must be
updated whenever a new usage is added, which looks like this:

COMPILE_ASSERT(blink::EndOfWebCryptoKeyUsage == (1 << 7) + 1, UpdateMe);

(Since I can't rely on a switch statement with non-default handler here).

My worry was that if it instead read:

COMPILE_ASSERT(blink::EndOfWebCryptoKeyUsage == (1 << 6) + 1, UpdateMe);

And yet the final value in SerializedScriptValue was (1 <<7), one might get
confused and introduce an off-by-one-error.

Maybe I am just being too cowardly, but I didn't see a downside in changing the
public enum values :)

(Also when we compile the serialization code, it should basically flatten into
nothing for these conversions).

[jsbell, 2014/03/14 18:39:27]

I have no preference. Maybe add a comment explaining why it doesn't start at
0x0001, though.

[eroman, 2014/03/14 19:32:51]

I reverted the change, so no longer a need for the comment.

+    WebCryptoKeyUsageDecrypt = 1 << 2,
+    WebCryptoKeyUsageSign = 1 << 3,
+    WebCryptoKeyUsageVerify = 1 << 4,
+    WebCryptoKeyUsageDeriveKey = 1 << 5,
+    WebCryptoKeyUsageWrapKey = 1 << 6,
+    WebCryptoKeyUsageUnwrapKey = 1 << 7,
 #if INSIDE_BLINK
     EndOfWebCryptoKeyUsage,
 #endif
@@ -86,11 +87,6 @@ class WebCryptoKeyHandle;
 //
 // If WebCryptoKey "isNull()" then it is invalid to call any of the other
 // methods on it (other than destruction, assignment, or isNull()).
-//
-// FIXME: Define the interface to use for structured clone.
-//        Cloning across a process boundary will need serialization,
-//        however cloning for in-process workers could just share the same
-//        (threadsafe) handle.
 class WebCryptoKey {
 public:
     ~WebCryptoKey() { reset(); }