Re: parcellfun debug

[Olaf Till]

On Wed, Jul 08, 2015 at 11:31:10PM -0700, Muhali wrote:
> Olaf Till-2 wrote
> > That the corresponding cellfun call works indicates the fault is in
> > parcellfun itself. Have you the latest version installed? On which
> > operating system do you use it? Which options did you set? The same as
> > in cellfun? It's always difficult to help someone else
> > debugging. First you can increase VerboseLevel to >1. Also, you could
> > insert commands into parcellfun which printout informative messages if
> > some steps are performed, e.g. before and after
> > 'isubp = fread (statr, 1, "double");', (parent process)
> > before and after fload() (parent process) and before
> > 'fwrite (cmdw(isubp), ijob, "double");', (parent process),
> > printing the value of ijob.
> 
> the program hangs at this step:
> 
>         isubp = fread (statr, 1, "double");
>         ## if pipe contained no more data, that's bad

If you don't want to do the debugging yourself, can you please post
the exact command calling parcellfun?

Is it possible to provide an example call to parcellfun which triggers
the problem? If possible with all arguments saved in binary format and
posted.

But maybe your parallely executed user function is too complex to post
it too. What you can do then is setting nproc to 1 and inserting a
printf() at the start of your user function, printing the arguments,
and a further printf at the end of the user function (giving the
information that the function has reached the end). So you see if the
user function always finishes, or at which arguments it hangs or gives
an error. And if the arguments are as expected from your specified
argument lists of parcellfun.

> regarding your questions/remarks:
> 
> I am using the dev Version (both of octave and parallel) under Linux and
> options are the same as for cellfun.

Then, what revision of parallel exactly? There were quite a lot of
commits in the last days.

If the options are the same as for cellfun, there should be no need to
use something different than the last release of parallel at all.

> Should I open a bug report and move
> discussion there?

I intend to follow this immediately, but if you wan't to have input
from others, the bug tracker may be better. If you post larger or
binary material, uploading it to the bug tracker is better.

Olaf

-- 
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

signature.asc
Description: Digital signature