On Fri, May 08, 2009 at 05:42:07PM +0100, Daniel P. Berrange wrote:
The patches we just applied for the VirtualBox open method still were not quite right. It would return VIR_DRV_OPEN_DECLINED when uri==NULL, but before doing so it would have set conn->uri to vbox:///session. So even though it declined the connection, all the later drivers would now ignore it. Also, it now returns DECLINED for some real errors that should be reported to the user.
Here's an alternative idea I've had for trying to address this. Some goals:
- If the user gives a URI with a vbox:/// prefix, we should always handle it, unless a 'server' is set when we leave it to the remote driver - If an invalid path is given we must give back a real error code - If after deciding the URI is for us, any initialization fails we must raise an error. - If the vbox glue layer is missing, we should still raise errors for requested URIs, so user knows their URI is correct. Looks very sensible to me and much nicer than my original suggestion to move VBoxCGlueInit() early into vboxOpen.
[..snip..]
+ _("no VirtualBox drviver path specified (try vbox:///session)")); s/drviver/driver/
Cheers, -- Guido