Re: 2.9.11?

[Daniel J Sebald]

John W. Eaton wrote:
> On 19-Apr-2007, Daniel J Sebald wrote:
>
> | > Does it help to move the fclear before the call to sleep?
> | 
> | No, that didn't seem to change anything.  It probably doesn't matter if the 
> | fclear() is before or after the sleep().
> | 
> | I think I've narrowed this down.  I added a few lines of code to print out 
> | whenever an error == EAGAIN occurs.  After about the seventh "make check", I got 
> | a string of numbers up to 99 at which point the test assumed failure and moved 
> | on.  (I.e., if it doesn't happen in ten seconds, then fail.)
> | 
> | I looked at the associated processes and found:
> | 
> |   5313 pts/2    00:03:08 octave
> |   5508 pts/2    00:00:00 sort <defunct>
> | 
> | So, for some reason that sort process is suspending.  Not sure why such a simple 
> | routine would fail.  Some confusion in the piping and in/out streams?  Maybe the 
> | coordination of fclose(in) and fgets(out) is off.  I.e., the input stream needs 
> | to finish closing before the output stream is ready, if not maybe sort says "hey"?
>
> I don't know what to suggest.  It seems to work reliably for me.  I'd
> consider a patch.

Try the following at the system command line:

mv /bin/sort /bin/sort_tmpmove
make check

and you will get the behavior I'm getting.  The patch I sent will basically time 
out after ten seconds.  I'm guessing this is a system level problem so isn't 
Octave's fault.  But Octave should probably always be robust to system level 
problems not integrated into Octave itself.  (I.e., outside program "sort" isn't 
part of Octave.)

Dan