Adds a feature to MHTML serialization that omits subframes and subresources marked no-store.

third_party/WebKit/Source/core/frame/FrameSerializer.cpp

 /*
  * Copyright (C) 2011 Google Inc. All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are
  * met:
  *
  *     * Redistributions of source code must retain the above copyright
  * notice, this list of conditions and the following disclaimer.
  *     * Redistributions in binary form must reproduce the above
@@ skipping 368 matching lines @@
         break;
 
     default:
         ASSERT_NOT_REACHED();
     }
 }
 
 bool FrameSerializer::shouldAddURL(const KURL& url)
 {
     return url.isValid() && !m_resourceURLs.contains(url) && !url.protocolIsData()
-        && !m_delegate.shouldSkipResource(url);
+        && !m_delegate.shouldSkipResourceWithURL(url);
 }
 
 void FrameSerializer::addToResources(Resource* resource, PassRefPtr<SharedBuffer> data, const KURL& url)
 {
+    if (m_delegate.shouldSkipResource(resource))

[Łukasz Anforowicz, 2016/05/18 17:27:25]

I wonder if we can skip processing a resource before a data copy takes place
(i.e. by tweaking addFontToResources so that it calls a [new] shouldAddResource
instead of shouldAddURL - this would return before copying data from
font->resourceBuffer into SharedBuffer).

[dewittj, 2016/05/19 04:13:17]

I am not sure how we could get the URL (or indeed the Resource) without fetching
a remote font.  Can I look into this in a future patch?

[Łukasz Anforowicz, 2016/05/19 17:00:13]

You're right - fetching is unavoidable, but we can avoid serialization work
(which for images/fonts involves currently a memory allocation + copy, but in
theory could become more expensive in the future).
Hmmm... how about a relatively small change to:

1. Introduce a new method:
   bool FrameSerializer::shouldProcessResource(const Resource& resource)
   {
       return shouldAddURL(resource.url()) &&
              !m_delegate.shouldSkipResource(resource);
   }

2. Call this method instead of calling shouldAddURL inside:
   - FrameSerializer::addImageToResources
   - FrameSerializer::addFontToResources
    

[dewittj, 2016/05/19 18:18:35]

Does this actually cause a copy?  It looks like it's a reference to an existing
SharedBuffer.

Also your solution doesn't simplify much since many calls to addImageToResources
specify a url from the src tag, which I am not sure is the same as
ImageResource.url().

[Łukasz Anforowicz, 2016/05/19 18:50:17]

My bad - you're right, this is just a ref-count increase.
Ok - sounds like the current code is the correct one to use.  Thanks for looking
into this.

+        return;
+
     if (!data) {
         DLOG(ERROR) << "No data for resource " << url.getString();
         return;
     }
 
     String mimeType = resource->response().mimeType();
     m_resources->append(SerializedResource(url, mimeType, data));
     m_resourceURLs.add(url);
 }
 
@@ skipping 72 matching lines @@
             continue;
         }
         emitsMinus = ch == '-';
         builder.append(ch);
     }
     CString escapedUrl = builder.toString().ascii();
     return String::format("saved from url=(%04d)%s", static_cast<int>(escapedUrl.length()), escapedUrl.data());
 }
 
 } // namespace blink