From bug-request at octave dot org Wed Dec 29 11:52:22 2004
Subject: Waiting for gnuplot to write file.
From: Peter JENSEN <peterjensen at agere dot com>
To: bug at octave dot org
Date: Wed, 29 Dec 2004 11:17:37 -0600

Hi,

On my system at work (with two processors) I see
a problem with legend. The problem is intermittent
however I managed to track it down, and I think I
understand the underlying reason.

The chain of events are as follows.

Firstly legend.m asks gnuplot
to save the original plotting command
into a temp file.

This is done via :

   tmpfilename=tmpnam;
   command=["save \"",tmpfilename,"\"\n"];
   graw(command);

Then legend waits for file to appear
and immediately after starts a
shell command that processes the temp file.

The code looks like.

   while (isempty(stat(tmpfilename)))
     if (++attempt > 20) error("gnuplot is not responding"); endif
     usleep(1e5);
   end
   plot_cmd = split(system(shell_cmd),"\n");

The problem is that the file is empty at
the time when the system call is made.

The problem is due to that the code only
checks if the file exists. However at
the time where the system call is made
the file has actually not been written
by gnuplot.

I verified this by modifying legend.m
as follows :

   file_test = stat(tmpfilename) ;
   while (isempty(file_test))
     if (++attempt > 20) error("gnuplot is not responding"); endif
     usleep(1e5);
     file_test = stat(tmpfilename) ;
   end

   file_test

When the code fails the size of the file is zero,
and the code prints.

file_test =
{
   atime = 1104339140
   blksize = 32768
   blocks = 0
   ctime = 1104339140
   dev = 21
   gid = 2100
   ino = 12074898
   modestr = -rw-rw-r--
   mtime = 1104339140
   nlink = 1
   rdev = 0
   size = 0
   uid = 23737
}

And when the code (occasionally works ) the file
size is different from zero, and the code prints.

file_test =
{
   atime = 1104339203
   blksize = 32768
   blocks = 16
   ctime = 1104339203
   dev = 21
   gid = 2100
   ino = 12074898
   modestr = -rw-rw-r--
   mtime = 1104339203
   nlink = 1
   rdev = 0
   size = 4316
   uid = 23737
}

On way to solve the problem is to look at the file
written by gnuplot, and see if it is complete
(it is not sufficient to look at the size of the
file.....).

Gnuplot indicated that the file has been
completed by writing "#    EOF" to the last line.
We can therefore test this via a "tail -1 file_name".

We could also find out if the file is open for writing,
as I assume that gnuplot will close the file when it has
been fully written, however I don't know how to do this.

I have enclosed the code that I used for testing.

Peter Jensen

=======================================
#!/opt/sbin/octave -q
more("off") ;
clear
closeplot

level = [ 0 -1 -2 -3 -4 -5 -6 ] ;

plot_legend = {} ;

for i=1:length(level)
   signal{i} = sin(0:0.01:10)*10**(level(i)/10) ;
   plot_legend{i}  = strcat("L = ",num2str(level(i))) ;
endfor

clearplot ;

## Plot signals
axis([ 1 length(signal{1}) ])
title("Test") ;

plot(signal{1}) ;
hold on ;

for k=2:length(level)
   plot(signal{k});
endfor

legend(plot_legend,-1) ;
==========================================

-- 
Peter Kabell Jensen                     E-mail  : peterjensen at agere dot com
Agere Systems                           Tel     : +44 (0) 1344 86 5849
Microelectronics House                  Fax     : +44 (0) 1344 86 5990
Kings Ride, Ascot, SL5 8AD  England



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------