On 09/14/2011 11:42 AM, Eric Blake wrote:
I've confirmed that we have a regression :(. Talking to an 0.8.8 server gives VIR_ERR_NO_SUPPORT when issuing the "unknown procedure: 243" message; it is only 0.9.3 and newer where we switched to the new rpc handling where we get VIR_ERR_RPC instead. I'm looking into how best to repair this regression, so that clients need not check for VIR_ERR_RPC everywhere.
More info from gdb. In both cases, I tested virsh client from latest libvirt.git, and had a breakpoint at virNetClientProgramDispatchError. With 0.8.8 server, err looks like: (gdb) p err $5 = {code = 39, domain = 13, message = 0x657cb0, level = 2, dom = 0x0, str1 = 0x657de0, str2 = 0x0, str3 = 0x0, int1 = 0, int2 = 0, net = 0x0} but with 0.9.4, it looks like: (gdb) p err $2 = {code = 39, domain = 7, message = 0x657d50, level = 2, dom = 0x0, str1 = 0x657e70, str2 = 0x657f90, str3 = 0x0, int1 = -1, int2 = -1, net = 0x0} That difference in domain (VIR_FROM_REMOTE vs. VIR_FROM_RPC) is sufficient to bypass our rewriting rules later in the function which convert VIR_ERR_RPC back to VIR_ERR_NO_SUPPORT. -- Eric Blake eblake@redhat.com +1-801-349-2682 Libvirt virtualization library http://libvirt.org