From bug-octave-request at bevo dot che dot wisc dot edu  Fri Apr 10 11:15:06 1998
Subject: Re: making a string global malfunctions
From: "John W. Eaton" <jwe at bevo dot che dot wisc dot edu>
To: Dirk Eddelbuettel <edd at debian dot org>
Cc: Michael Smolsky <fnsiguc at siputer dot weizmann dot ac dot il>,         bug-octave@bevo.che.wisc.edu
Date: Fri, 10 Apr 1998 11:14:42 -0500 (CDT)

On 10-Apr-1998, Dirk Eddelbuettel <edd at debian dot org> wrote:

| You discovered a misfeature of 2.0.10 which was fixed in 2.0.11. If you
| upgrade to Octave 2.0.11, you will see that it behaves as expected: 

Are you sure?  I can still duplicate the problem with --traditional,
but I don't think it's really a bug.  The problem is that now with
--traditional, global variables are initialized to `[]', and
implicit_str_to_num_ok is initialized to 1.  This combination causes 

  global a

to initialize a to an empty matrix, and since

  a([1,2],:)=str2mat('red','yellow');

is assigning strings to a numeric matrix using indexing, Octave
attempts to convert the RHS to numeric values, and that succeeds
because of the setting of implicit_str_to_num_ok.

But hey, it's Matlab-compatible (except that Matlab's deblank function
just warns if you try to deblank a numeric value instead of making it
an error).

So, I don't plan to change the behavior here (beyond whatever changes
happened from 2.0.10 to 2.0.11).  If you want it to work as you
expect, I suggest you use some other set of preference variables.

jwe