Implement File constructor.

Source/core/fileapi/BlobBuilder.cpp

 /*
  * Copyright (C) 2010 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 76 matching lines @@
     if (!arrayBufferView)
         return;
 
     appendBytesData(arrayBufferView->baseAddress(), arrayBufferView->byteLength());
 }
 
 void BlobBuilder::append(Blob* blob)
 {
     if (!blob)
         return;
-    if (blob->isFile()) {
+    if (blob->isOnFilesystem()) {
         File* file = toFile(blob);
         // If the blob is file that is not snapshoted, capture the snapshot now.
         // FIXME: This involves synchronous file operation. We need to figure out how to make it asynchronous.
         long long snapshotSize;
         double snapshotModificationTime;
         file->captureSnapshot(snapshotSize, snapshotModificationTime);
 
         m_size += snapshotSize;
         if (!file->fileSystemURL().isEmpty())
             m_items.append(BlobDataItem(file->fileSystemURL(), 0, snapshotSize, snapshotModificationTime));
@@ skipping 19 matching lines @@
     OwnPtr<BlobData> blobData = BlobData::create();
     blobData->setContentType(contentType);
     blobData->swapItems(m_items);
 
     RefPtr<Blob> blob = Blob::create(BlobDataHandle::create(blobData.release(), m_size));
 
     // After creating a blob from the current blob data, we do not need to keep the data around any more.
     // Instead, we only need to keep a reference to the blob data just created.
     m_items.append(BlobDataItem(blob->blobDataHandle(), 0, m_size));
 
-    return blob;
+    return blob.release();
+}
+
+PassRefPtr<File> BlobBuilder::getFile(const String& contentType, const String& fileName, double modificationTime)

[esprehn, 2013/11/04 19:28:08]

createFile, method that create objects shouldn't be called get*()

[pwnall-personal, 2013/11/04 23:01:21]

Actually, I think this whole class should be moved to platform. It is here
because it implements the deprecated BlobBuilder spec.
https://developer.mozilla.org/en-US/docs/Web/API/BlobBuilder

I used "getFile" because it is in the spec, and the signature roughly matches.
Do you still want me to rename it?

+{
+    OwnPtr<BlobData> blobData = BlobData::create();
+    blobData->setContentType(contentType);
+    blobData->swapItems(m_items);
+
+    RefPtr<File> file = File::create(fileName, modificationTime, BlobDataHandle::create(blobData.release(), m_size));
+
+    // After creating a file from the current blob data, we do not need to keep the data around any more.
+    // Instead, we only need to keep a reference to the blob data just created.
+    m_items.append(BlobDataItem(file->blobDataHandle(), 0, m_size));

[esprehn, 2013/11/04 19:28:08]

What cleans this up to prevent leaks?

[pwnall-personal, 2013/11/04 23:01:21]

I borrowed this code from "getBlob", so I don't know too much about it. Here is
why I suspect it won't leak.

m_items is an empty Vector<BlobDataItem> after the blobData->swapItems() call.
The BlobDataItem is appended to it here, and will be cleaned up in the
auto-generated BlobBuilder destructor that will call the Vector<BlobDataItem>
destructor.

blobDataHandle() returns a PassRefPtr<BlobDataHandle>, and the BlobDataItem()
constructor takes one as an argument. If I understand the WebKit blog post
correctly, PassRefPtr acts a bit like C++ 11's move constructor, so the original
BlobDataHandle* is ref()'d in blobDataHandle() and then is passed straight into
BlobDataItem without ref()/deref() churn.

I'll read the whole source code for RefPtr and PassRefPtr and re-analyze. This
is what I think based on the bits I read so far, and on the WebKit blog post.
http://www.webkit.org/coding/RefPtr.html

+
+    return file.release();
 }
 
 } // namespace WebCore