On 02/24/2012 05:17 AM, Pedro Alves wrote:
>> > @@ -4882,6 +4883,17 @@ info_static_tracepoint_markers_command (char *arg, int from_tty)
>> >    struct ui_out *uiout = current_uiout;
>> >    int i;
>> >  
>> > +  if (target_can_use_agent () == 0)
> Hmm, I don't think doing this change now is right.  If you connect GDB to a GDBserver
> that hasn't been patched, this will fail because that GDBserver doesn't report
> the QAgent feature, so it is an incompatible change of the protocol to require QAgent
> for static tracepoint listing.
> The right thing to do seems to be to just go ahead and try the target method call,
> as before.  We could at a minimum check whether the "statictracepoints" qSupported
> feature Was reported by the target, though that doesn't tell you whether the current
> inferior has the IPA or UST loaded at all, only that the frontend stub understands
> static tracepoints.  What was the motivation for these checks here?
> 

The motivation is quite simple, `since we have some conditions, it is
better to check them first'.  However, it breaks compatibility, as you
pointed out.  In this version, I get rid of these checks, and only leave
a paragraph of comment on why we don't check these conditions there.

-- 
Yao (齐尧)