Give SxS distribution its own registration GUIDs.

chrome/installer/util/browser_distribution.h

Index: chrome/installer/util/browser_distribution.h
diff --git a/chrome/installer/util/browser_distribution.h b/chrome/installer/util/browser_distribution.h
index b308c04fecddd8223de6ef11c97ae9dcd7f031a3..f1ec82c0db2a4224c4b3b8b74372872a2dfa14a3 100644
--- a/chrome/installer/util/browser_distribution.h
+++ b/chrome/installer/util/browser_distribution.h
@@ -89,6 +89,14 @@ class BrowserDistribution {
   // should be suffixed in all scenarios.
   virtual string16 GetBaseAppId();
 
+  // Returns the Browser ProgId prefix (e.g. ChromeHTML, ChromiumHTM, etc...).
+  // The full id is of the form |prefix||suffix| where suffix is a 26-digit
+  // alpha-numeric identifier.

[gab, 2013/09/06 18:40:08]

|suffix| is actually a '.' followed by a 26-character alpha-numeric identifier.

Also add to this comment that this prefix must be 11 characters or less so that
|prefix||.suffix| is 38 characters or less given the restriction that the ProgId
be 39 characters or less including the NULL character (something like the
previous comment in shell_util.cc, maybe slightly less details would do here).

[zturner, 2013/09/06 20:53:16]

Done.

+  virtual string16 GetBrowserProgIdPrefix();
+
+  // Returns the Browser ProgId description.
+  virtual string16 GetBrowserProgIdDesc();
+
   virtual string16 GetInstallSubDir();
 
   virtual string16 GetPublisherName();
@@ -117,8 +125,20 @@ class BrowserDistribution {
 
   virtual string16 GetVersionKey();
 
+  // Returns true if setting the default browser is a supported operation.

[grt (UTC plus 2), 2013/09/06 02:53:39]

it seems that uninstall is the only real use of this function (the
metro-specific user data dir is history). what do you think of removing this and
changing uninstall.cc's DeleteChromeRegistrationKeys so that it does an early
exit if the distribution has an empty BrowserProgIdPrefix (and provide overrides
for GCF and the binaries so that GetBrowserProgIdPrefix() returns an empty
string for them)?

[gab, 2013/09/06 18:40:08]

Hmmm, I'm not sure about that, we usually put NOTREACHED() in methods of
distributions that shouldn't be called (i.e. GetBrowserProgIdPrefix for
binaries/CF); I find that safer than returning empty strings only as a mean to
check what this new method is trying to provide clearly.

[grt (UTC plus 2), 2013/09/09 18:39:15]

I see your point. I find these two methods (IsFooSupported and CanFoo)
confusing. I think a simpler API is preferable. I find making CanSetAsDefault a
tristate less so. How do you find that proposal?

+  virtual bool IsSetAsDefaultSupported();
+
+  // Returns true if this distribution can be set as the default browser.
+  // Note that this is not the same as IsSetAsDefaultSupported, which returns
+  // true if the browser can EVER be set as default.  It is possible for the
+  // operation to be supported but currently disabled (for example due to
+  // policy).
   virtual bool CanSetAsDefault();
 
+  // Forces CanSetAsDefault() to return true.  For distributions where
+  // IsSetAsDefaultSupported() returns false, this function is a no-op.
+  virtual void DebugEnableSetAsDefault();
+
   virtual bool CanCreateDesktopShortcuts();
 
   virtual bool GetChromeChannel(string16* channel);
@@ -126,7 +146,11 @@ class BrowserDistribution {
   // Returns true if this distribution includes a DelegateExecute verb handler,
   // and provides the CommandExecuteImpl class UUID if |handler_class_uuid| is
   // non-NULL.
-  virtual bool GetCommandExecuteImplClsid(string16* handler_class_uuid);
+  virtual bool GetCommandExecuteImplClsidString(string16* handler_class_uuid);
+
+#if defined(OS_WIN)
+  virtual bool GetCommandExecuteImplClsid(CLSID* handler_class_uuid);
+#endif
 
   // Returns true if this distribution uses app_host.exe to run platform apps.
   virtual bool AppHostIsSupported();