Refactor downloads into a ProfileKeyedService.

content/browser/download/download_manager_delegate.h

Index: content/browser/download/download_manager_delegate.h
diff --git a/content/browser/download/download_manager_delegate.h b/content/browser/download/download_manager_delegate.h
index 55d25f6a5f469021a3737e40af31f18b346d51fd..f0b0f7fadf1e13abeec5385699a0e9c0f2b9361f 100644
--- a/content/browser/download/download_manager_delegate.h
+++ b/content/browser/download/download_manager_delegate.h
@@ -11,6 +11,7 @@
 #include "base/time.h"
 
 class DownloadItem;
+class DownloadManager;
 class FilePath;
 class TabContents;
 class SavePackage;
@@ -105,6 +106,10 @@ class DownloadManagerDelegate {
   // Informs the delegate that the progress of downloads has changed.
   virtual void DownloadProgressUpdated() = 0;
 
+  // Tell the delegate what manager it's delegating for;
+  // only called by the DownloadManager.
+  virtual void SetDownloadManager(DownloadManager* manager) = 0;

[jam, 2011/10/10 18:19:33]

this pattern of having the delegate be told by the object that it's been set is
not something we do with our other delegates. is this really necessary or can we
just do this manually? not all embedders might care to have the DownloadManager
pointer

[Randy Smith (Not in Mondays), 2011/10/11 18:24:43]

Yeah.  This is to make the test code simpler--it allows us to replace delegates
on the download manage and have the right thing happen with a single call (set
the DM delegate and have the DM tell the delegate what DM it's delegating for). 
But I agree that it's worth a fair amount of complexity in test code to make
production code simpler, so I'll change it.  It's just producing more fragile
test code.  

+
  protected:
   DownloadManagerDelegate() {}