[haiku-commits] Re: r39846 - in haiku/trunk: headers/os/kernel headers/private/debug headers/private/kernel src/kits/debug src/system/kernel ...
- From: Ingo Weinhold <ingo_weinhold@xxxxxx>
- To: haiku-commits@xxxxxxxxxxxxx
- Date: Tue, 14 Dec 2010 21:42:11 +0100
On 2010-12-14 at 21:20:39 [+0100], Rene Gollent <anevilyak@xxxxxxxxx> wrote:
> On Tue, Dec 14, 2010 at 2:39 PM, Ingo Weinhold <ingo_weinhold@xxxxxx> wrote:
> > Yeah, sorry, I'm guilty of suggesting the debugger interface extension in
> > the
> > ticket. A listener mechanism definitely is more elegant than polling, but
> > I
> > agree with Axel that name or priority changes aren't particarly debugging
> > related, nor are debuggers the only applications that might be interested
> > in
> > this info. A general listener mechanism for system information would be
> > nicer.
>
> Fair enough, what other types of system information would be of
> interest for such a listener mechanism?
For the time being I'd start with team creation/deletion/exec and thread
creation/deletion events. ATM the only programs interested in those events I
can think of are Debugger (for the team window) and the registrar (to know
when an application dies).
> > Unfortunately it's not that simple. The whole backend
> > (thread_hit_debug_event()) cannot be used. The messages must be
> > asynchronous.
> > user_debug_thread_deleted() is closer to how that should look. A major
> > problem remains, however, that in this case the calling thread should not
> > block (particularly when it belongs to another team), which cannot be
> > guaranteed, unless one uses timeouts for locking and sending the message,
> > which in turn makes those notifications uncertain (unlike all other
> > debugger
> > messages).
>
> I think that at least in the short term we can live with that
> limitation since instances of one team renaming another's threads are
> more the exception than the rule (priorities are not quite as obvious
> since e.g. processcontroller makes changing those for another team's
> threads quite trivial).
Sure. OTOH since Debugger would be the only user of the interface, a polling
approach implemented there would be just as acceptable.
CU, Ingo
Other related posts:
