Re: Patch to Octave shutdown procedures

[John Swensen]

On May 19, 2009, at 10:25 AM, Jaroslav Hajek wrote:

> On Mon, May 18, 2009 at 9:27 PM, John Swensen  
> <address@hidden> wrote:
>
> I'm not sure I understand the problem. Can't you just send "end of
> input" to the embedded Octave to make it exit from main_loop? The
> thread running Octave can then signal to other threads that main_loop
> ended in whatever manner you need, so you can synchronize, and then
> call do_octave_atexit and sysdep_cleanup. Maybe it would be nice to
> have a cleanup_only variant of cleanup_and_exit.
>
> cheers
>
> --

Having a cleanup_only variant of cleanup_and_exit was kindof what I  
was trying to do.  However, it is not enough just to have the  
cleanup_only variant, we need to allow a developer to set a flag which  
indicates this should be used in place of cleanup_and_exit.  Also, I  
don't really want to "cleanup" when an exit-Octave condition occurs, I  
simply want the Octave readline loop to end, then clean up all the IDE  
stuff that is interacting with octave, then cleanup Octave.  The  
biggest problem occurs when a user calls 'exit' or does a CTRL-C until  
Octave exits.  In this instance, exit() gets called before I have any  
indication that Octave is closing.  This means that the threads I have  
which monitor Octave continue to do so, even though all of Octave has  
been cleaned up.

My patch attempted to allow an IDE to instantiate octave by calling  
octave_main such that:
1) Any condition which would cause octave to call the libc function  
exit() (e.g. exit, CTRL-C, others) simply cause octave_main() to return
2) do_octave_atexit() would not be called before octave_main() returns

If you can think of a better way of doing this, then I will try to  
implement it.  Your suggestion of sending an "end of input" to exit  
from the main_loop doesn't quite work for 2 reasons.  First, if you  
look at the source code for the octave_main function in octave.cc, if  
main_loop returns then clean_up_and_exit() is called.   Second, if you  
look at the definition of 'quit' in toplev.cc and the signal handlers  
in sighandlers.cc you will see that clean_up_and_exit() is called from  
these without the main_loop ever returning.

My patch addresses all of these problems by simply adding a flag which  
indicates that calls to clean_up_and_exit should simply exit the  
main_loop and defer cleanup until later.  The developer using this  
"ide embedded" mode is then responsible for cleanup.

John Swensen