On Aug 31, 2007, at 2:24 PM, Anyang Ren wrote:

In kjs_window.cpp, Window::isSafeScript(ExecState *exec), we have:

KURL actURL = activeFrame->loader()->url(); WebCore::String actDomain = actURL.host(); ... KURL thisURL = frame->loader()->url(); ... WebCore::String thisDomain = thisURL.host();

if (actDomain == thisDomain && actURL.protocol() == thisURL.protocol() && actURL.port() == thisURL.port()) return true;

if (Interpreter::shouldPrintExceptions()) { printf("Unsafe JavaScript attempt to access frame with URL %s from frame with URL %s. Domains, protocols and ports must match.\n", thisDocument->URL().latin1(), actDocument-

...

URL().latin1()); } String message = String::format("Unsafe JavaScript attempt to access frame with URL %s from frame with URL %s. Domains, protocols and ports must match.\n", thisDocument->URL().latin1(), actDocument- URL().latin1());

Since thisURL and actURL are the URLs used in the test, why are we using thisDocument->URL() and actDocument->URL() in the error messages?

I think perhaps the message should mention both. thisURL and actURL (not such great names!) may identify the frame that is considered to own this frame in cases like about:blank or javascript: frame source, but they are not the true URL of the frame, so may not be very helpful in the error message. I think the format should be something like: "Unsafe JavaScript attempt to access frame with URL %s (owned by %s) from frame with URL %s (owned by %s). Domains, protocols and ports must match.\n" But the "owned by" part should probably be dropped entirely in the common case where the URLs are the same. We'd take a patch to improve this.

In fact, actDocument could be NULL. Earlier in this function, we have:

WebCore::Document* actDocument = activeFrame->document();

if (actDocument) { if (thisDocument->domainWasSetInDOM() && actDocument-

...

domainWasSetInDOM()) { if (thisDocument->domain() == actDocument->domain()) return true; } }

The if (actDocument) test suggests that actDocument could be NULL.

I'm not sure it's possible for a frame to be executing script with a null document, so I don't think this is actually possible. In fact, in current trunk, I don't think it is ever possible for any frame to have a null document. Regards, Maciej