From bug-octave-request at bevo dot che dot wisc dot edu  Tue Aug 11 23:45:11 1998
Subject: Octave dumps core
From: "John W. Eaton" <jwe at bevo dot che dot wisc dot edu>
To: Francesco Potorti` <F dot Potorti at cnuce dot cnr dot it>
Cc: Octave bugs list <bug-octave at bevo dot che dot wisc dot edu>
Date: Tue, 11 Aug 1998 23:44:19 -0500 (CDT)

On 11-Aug-1998, Francesco Potorti` <F dot Potorti at cnuce dot cnr dot it> wrote:

| Bug report for Octave 2.0.13 configured for i386-pc-linux-gnu
| 
| I append  .m file that  makes octave crash, and a  data file bzipped and
| uuencoded, which is read  by the .m  file.  To reproduce, just call  the
| script as shown:

| [...]

| ---------------- start of satelliteatt.m -----------------
| # Read the "att" file containing two column vectors att20 and att30 listing
| # the measured attenuation.  They are 179580 samples long.
| load -force att
| L=length(att20);
| 
| # Build the matrix att composed of the two column vectors
| att=[att20 att30];
| disp("[min(att); max(att)]:");
| disp([min(att); max(att)]);
| 
| # Build the attbin20, attbin30 vectors and the attbin matrix, containing the
| # number od the bin of the attenuation samples.  Bins are 2dB-wide for
| # uplink (30 GHz) and 1dB-wide for downlink (20GHz).  The first bin has
| # number 0. 
| attbin=fix([att20 att30/2]);
| disp("[min(attbin);max(attbin)]:");
| disp([min(attbin);max(attbin)]);
| 
| # Build attstring, a column vector desribing the strings of samples internal
| # to the same bin. Four columns: row where the string begins, column (1 for
| # att20, 2 for att30), length of the string, number of the bin.
| attseq=[[1 1]; diff(attbin)];
| attseq=(attseq!=0);
| [a,b]=find(attseq);
| d=attbin(a,b);

This is a memory allocation problem.  Unfortunately, it appears that
on Linux systems malloc (and therefore the C++ operator new) is not
reporting a problem, and then the failure happens later when Octave
tries to use memory that it shouldn't.  At least on a DEC Alpha
running Digital Unix, I get a memory exhausted error message.  I
believe the problem really needs to be fixed in the Linux malloc.
See http://www.che.wisc.edu/octave/mailing-lists/bug-octave/1998/215
for some more information about this problem.

In your code, I don't think the lines

  [a,b]=find(attseq);
  d=attbin(a,b);

are doing what you want.  For your data, the a and b index vectors are
both 32294 elements long, so when you do the indexing operation, you
are attempting to create a result matrix with 1042902436 elements,
which requires approximately 7GB of storage.

I'm not exactly sure what you want your code to do.  I'm guessing that
it is something more like

  t = attbin(:)';
  d = t (find (attseq));


jwe