Re: [TCLCORE] CFV: TIP #435

[Cyan O. <cya...@gm...>]

On Thu, Jul 23, 2015 at 7:23 PM, Jan Nijtmans <jan...@gm...>
wrote:

> 2015-07-23 5:39 GMT+02:00 Gustaf Neumann <ne...@wu...>:
> >
> > Am 15.07.15 um 09:29 schrieb Harald Oehlmann:
> >> No, it worked perfectly, if I remember right.
> >
> > So, the preferred pattern is: if an application needs a notifier
> > thread running after a fork, it should call Tcl_InitNotifier()
> > after the fork (like in rivet in the ap_hook_child_init() callback)
>
> I think this has been discussed before. Prefered is that all
> applications just can use fork() out of the box, while a
> forked interpreter works the same as the original
> interpreter. Since the vwait command expects events
> to be handled, and for this events to work a notifier
> thread is necessary, it means that any application
> should simply assume that the notifier works, no
> matter Tcl is compiled with or without threads.


What is the performance impact of creating a notifier thread just to throw
it away for the [exec] case?

It seems that anything measurable would be a poor trade-off - slow down an
extremely common case to enable an extremely rare case.

Cyan