REGRESSION: form events don't fire after back/forward navigation, due to inconsistent load state

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Geoffrey Garen]

Check out the attached reduction. 

The patch for 5518 caused this. I'm surprised Mitz didn't catch it; he must be slipping... :)

[Geoffrey Garen]

Created attachment 5414 [details]
reduction

[Geoffrey Garen]

Sorry, not 5518; rdar://problem/4268278.

[Geoffrey Garen]

Created attachment 5562 [details]
Fix

Darin had a look at a previous incarnation of this patch, but I think I've
found a better way that is more in tune with the loader's natural behavior.

[Geoffrey Garen]

Comment on attachment 5562 [details]
Fix

The key to this new patch is:

+    } else if ([_private->subresourceLoaders count] > 0) {
+	 // The main resource loader already finished loading. Set the
cancelled error on the 
+	 // document and let the subresourceLoaders send individual cancelled
messages below.
+	 [self _setMainDocumentError:[self _cancelledError]];
     } else {
-	 // Main handle is already done. Set the cancelled error.
-	 NSError *cancelledError = [NSError
_webKitErrorWithDomain:NSURLErrorDomain
-							     
code:NSURLErrorCancelled
-							       URL:[self
_URL]];
-	 [self _setMainDocumentError:cancelledError];
+	 // If there are no resource loaders, we need to manufacture a
cancelled message.
+	 // (A back/forward navigation has no resource loaders because its
resources are cached.)
+	 [[self _webView] _mainReceivedError:[self _cancelledError]
+			      fromDataSource:self
+				    complete:YES];
     }

[Darin Adler]

Comment on attachment 5562 [details]
Fix

I would write "Cancels the DataSource's pending loads" as "Cancels the data
source's pending loads" (not important, just a comment in passing).

This looks better than the earlier fix you had me review last week.

r=me

[Geoffrey Garen]

Landed as revision 11963.