Re: [Patch, Fortran] Add STAT_STOPPED_IMAGE to SYNC ALL / SYNC IMAGES

[Tobias Burnus <burnus at net-b dot de>]

Hi Daniel,

Daniel Carrera wrote:
> I agree, except that I don't know what va_{list,start,end} do. But 
> other than that, I agree that a "runtime_error" function would be good.

See http://en.wikipedia.org/wiki/Stdarg.h

> > As a side effect, it fixes the potentially wrong wording of
> > "ERROR: Image %d is stopped", caf_this_image
> > While that's currently the case, when moving to a true multi-image
> > implementation, the stopped image is usually not the current image but
> > some other remote image.
>
> I'm probably confused. I was thinking that caf_is_finalized would only 
> be 1 if the current image has stopped, in which case, the error would 
> be correct. Am I wrong?

Well, that's definitely the case with the current code. But for a truly 
multi-image code, I would expect that a finalizing image tells the other 
images that it is about to close down - before it enters the message 
processing loop, waiting for the other images to confirm that it can 
stop now (and meanwhile processing remote requests).

Thus, a finalizing image can still process all requests, but it won't be 
able to participate in any collective action such as SYNC ALL or 
ALLOCATE/DEALLOCATE, which have to be executed on all images.

While it is not impossible that one calls allocate/deallocate/sync when 
the *current* image is already stopped, it is difficult to do so in 
practice as the finalize function is automatically only called for STOP 
or when the main program ends - and it is difficult to do something 
afterwards. (One way is mult-language programming, where one manually 
calls finalize.)

On the other hand, the case that a remote image stopped can happen much 
easier - though it should be most of the time (always?) a programming 
bug of the user.

Tobias

PS: For the final patch, please remember to CC gcc-patches@