Tweaking WebPageSerializerImpl to emit a BOM for UTF16/32.

third_party/WebKit/Source/web/WebPageSerializerImpl.cpp

Index: third_party/WebKit/Source/web/WebPageSerializerImpl.cpp
diff --git a/third_party/WebKit/Source/web/WebPageSerializerImpl.cpp b/third_party/WebKit/Source/web/WebPageSerializerImpl.cpp
index 694c823537fd26515f275bd1652c7d9b25ae848e..dae6be2b83589f54d21dd41a566007e9bb56d3fb 100644
--- a/third_party/WebKit/Source/web/WebPageSerializerImpl.cpp
+++ b/third_party/WebKit/Source/web/WebPageSerializerImpl.cpp
@@ -93,6 +93,8 @@
 #include "core/page/PageSerializer.h"
 #include "public/platform/WebVector.h"
 #include "web/WebLocalFrameImpl.h"
+#include "wtf/ArrayBufferBuilder.h"
+#include "wtf/text/CString.h"
 #include "wtf/text/TextEncoding.h"
 
 namespace blink {
@@ -274,8 +276,26 @@ void WebPageSerializerImpl::encodeAndFlushBuffer(
 
     CString encodedContent = param->textEncoding.encode(content, WTF::EntitiesForUnencodables);
 
+    CString encodedBOM;

[jsbell, 2015/12/07 22:53:51]

Move this block up above the encodedContent initialization, as BOM comes before
content?

+    if (m_noBytesEncodedYet) {
+        encodedBOM = param->textEncoding.encodeBOMifApplicable();

[jsbell, 2015/12/07 22:53:51]

I was going to suggest some changes to encodeBOMifApplicable (and request a
comment, etc) but as I was reviewing this I ran across:

TextEncoding::isNonByteBasedEncoding() 

... which captures the semantics we want. I think that's sufficiently close to
what we want such that we should instead just call that from
WebPageSerializerImpl::encodeAndFlushBuffer and make no changes to text/wtf.
(Sorry, I should have known about this before.)

We should comment in the code something like:

// When serializing with encodings that are not ASCII-compatible
// (e.g. UTF-16) it is necessary to prefix the content with a 
// byte order mark (BOM) so that it can be parsed correctly.
if (param->textEncoding.isNonByteBasedEncoding()) { ... }

+        m_noBytesEncodedYet = false;
+    }
+
+    WebCString result;
+    if (encodedBOM.length() == 0) {
+        result = WebCString(encodedContent);
+    } else {
+        ArrayBufferBuilder resultBuilder;
+        resultBuilder.append(encodedBOM.data(), encodedBOM.length());
+        resultBuilder.append(encodedContent.data(), encodedContent.length());
+        result = WebCString(

[jsbell, 2015/12/07 22:53:51]

Don't need to wrap in blink.

+            static_cast<const char*>(resultBuilder.data()),
+            resultBuilder.byteLength());
+    }
+
     // Send result to the client.
-    m_client->didSerializeDataForFrame(WebCString(encodedContent), status);
+    m_client->didSerializeDataForFrame(result, status);
 }
 
 // TODO(yosin): We should utilize |MarkupFormatter| here to share code,
@@ -423,6 +443,7 @@ WebPageSerializerImpl::WebPageSerializerImpl(WebLocalFrame* frame,
                                              const WebString& localDirectoryName)
     : m_client(client)
     , m_localDirectoryName(localDirectoryName)
+    , m_noBytesEncodedYet(true)
     , m_htmlEntities(false)
     , m_xmlEntities(true)
 {