Re: tracing broken if target doesn't do disconnected tracing

[Stan Shebs <stan@codesourcery.com>]

Pedro Alves wrote:
> On Wednesday 07 April 2010 23:04:38, Stan Shebs wrote:
>   
>> Pedro Alves wrote:
>>
>>     
>>> I suggest we change that to simply:
>>>
>>>   /* If running interactively, warn the user a trace run is ongoing.
>>>      She may want to cancel detaching instead.  */
>>>     if (current_trace_status ()->running && from_tty)
>>>       {
>>>         if (current_trace_status ()->disconnected_tracing)
>>>           {
>>>             if (!query (_("Trace is running and will continue after detach; detach anyway? ")))
>>>               error (_("Not confirmed."));
>>>           }
>>>         else
>>>           {
>>>             if (!query (_("Trace is running but will stop on detach; detach anyway? ")))
>>>              error (_("Not confirmed."));
>>>           }
>>>       }
>>>
>>>   - simpler, more coherent, less explaining, less confusing.
>>>   
>>>       
>> The downside of this design is that if you did want to shut tracing 
>> down, you have to cancel the detach, do a tstop, then redo the detach.  
>> It's not crucial perhaps, but it seems a bit pedantic for GDB to have 
>> the power to choose whether to keep the trace running, but not to 
>> exercise it, and to insist that you have cancel and type the command 
>> yourself.  Perhaps the crux of the confusion is that this is really a 
>> three-way choice - trace/detach, tstop/detach, cancel - and a pair of 
>> yes/no questions is not a good way to model it.
>>     
>
> That's just begging for:
>
> The downside of your current design is that if you did want to detach
> and leave tracing running, you have to cancel the detach, do a "set
> disconnected-tracing on", then redo the detach.
> It's not crucial perhaps, but it seems a bit pedantic for GDB to have 
> the power to choose whether to keep the trace running, but not to 
> exercise it, and to insist that you have cancel and type the command 
> yourself.
>
> :-)
>   

Um, I must be missing the joke, it looks like you cut-n-pasted verbatim?

> I'd say that this use case would be much more common, and if
> gdb gave you the option of turning disconnected tracing _on_
> would be more helpful than the other way around (having trace
> running and wanting it to stop on detach).  After having used
> tracing for a while (granted, not really in the field), I believe
> that this "do you want to stop tracing" query will get old soon.  But,
> I can see that I can't manage to convince you of anything, so
> please, let's not get stuck on this issue, and let's move forward
> with whatever you think is right.
>   

My mental model is that users will tend to have disconnected tracing on 
by default if possible, because it ensures that the tracing run isn't 
prematurely terminated by GDB crash or loss of network connection.  But 
yeah, without much user feedback yet, we are just speculating on what 
users will prefer.  So I'll check in my version, and if people complain 
about it, then you'll get that delicious feeling of vindication. :-)

Stan