Re: [patch] remote_set_trace_notes

[Pedro Alves <palves@redhat.com>]

On 03/01/2013 05:43 PM, Abid, Hafiz wrote:
> 
> Hi All,
> During review of trace-buffer-size patch, I was asked to move a change into a separate patch. This is simple change in remote_set_trace_notes that will make this function return if all arguments are NULL.

> 
> 2012-03-01  Stan Shebs  <stan@codesourcery.com>
>         Hafiz Abid Qadeer  <abidh@codesourcery.com>
> 
>     gdb/
>     * remote.c (remote_set_trace_notes): Handle no-op case better.

Should this really be a no-op?  I'm not sure this is correct.

Say you connect to a target that had been tracing
before (disconnected tracing), and had its trace
notes/user/etc. set.

GDB connects, the user sets up a new trace session,
but doesn't set notes/user/etc.  IOW, the (trace_user/trace_notes/etc.)
GDB variables are left with their NULL defaults.

"show trace-notes" shows:

  (gdb) show trace-notes
  The notes string to use for current and future trace runs is "".

The user starts the trace session with tstart.

With the patch, GDB skips sending "QTNotes:...".

The target reuses the user/notes from the previous session.

That can't be right.

Do the same experiment again, but do (literally)
"set trace-notes" before tstart.  This sets the trace_notes
variable to "" instead of NULL.  This time, "tstart"
actually clears the previous run's trace notes on the
target.

Seems to me there's actually a bug here, and NULL and ""
should be treated as equivalent by remote_set_trace_notes.

-- 
Pedro Alves