Pedro Alves wrote:
> On Wednesday 17 June 2009 18:19:10, Aleksandar Ristovski wrote:
> 
>> Here is a patch that does what you described.
> 
> Thanks.  This is still not complete, though.
> 
> You need to make sure that the "multi_process" flag stays turned off
> if the backend doesn't support MP, otherwise, things like
> remote-utils.c:write_ptid will still misbehave (gdb will realize the
> target doesn't support multi-process, but the remote target will
> be replying with multi-process packets, how odd would that be?!).
> That flag is currently bound to GDB's multi-process support,
> because current gdbserver is assuming all backends support mp.
> 
> 	    /* Record if GDB knows about multiprocess support.  */
> 	    if (strcmp (p, "multiprocess+") == 0)
> 	      multi_process = 1;
> 
> 
> To see this in action, you can connect with
> "target extended-remote :foo"/"set debug remote 1", and make sure that
> the packets flowing in both directions don't use multi-process thread
> id extensions (e.g., 'p1.1' vs '1').
> 

Ok, then something like this?

-- 
Aleksandar Ristovski
QNX Software Systems


* target.h (supports_multiprocess): New function.
* server.c (handle_query): Use supports_multiprocess and output
multiprocess accordingly.
* linux-low.c (linux_supports_multiprocess): New function.
(linux_target_ops): Set supports_multiprocess.