From maintainers-request at octave dot org Thu Mar 16 11:40:54 2006
Subject: Re: 2.9.5 and 2.1.73
From: David Bateman <David dot Bateman at motorola dot com>
To: "John W. Eaton" <jwe at bevo dot che dot wisc dot edu>
CC: David Bateman <adb014 at gmail dot com>, maintainers@octave.org
Date: Thu, 16 Mar 2006 18:36:20 +0100

John W. Eaton wrote:
> On 16-Mar-2006, David Bateman wrote:
>
> | Please find attached the patch of all of the code I have for 2.9.5. It
> | includes a number of changes a fixes for the sparse matrix type, but also an
> | attempt at a fix for a valgrind identified problem for strncmp, where I
> | introduce a means of detecting the size of the string in the OSSTREAM and
> | ensure it is longer than the length argument to strncmp. This didn't appear
> | to fix the issue for me and so I suspect the problem is due to my gcc
> | 4.0.1libc strncmp function and similar issues are identified in
> | readline.
> | However, the patch for this seems useful in any case and so I supply it. Now
> | that the test code is stabilizing its probably worthwhile taking the many
> | hours time necessary to run
> | 
> | % cd test
> | % LD_LIBRARY_PATH="`pwd`/../src:`pwd`/../liboctave:`pwd`/../libcruft"
> | valgrind --tool=memcheck ../src/octave -f -q -H -p
> | .//:../src//:.:./../scripts// ./fntests.m .
> | 
> | to get a complete list of all of the problems that valgrind can identify
> | with the current test code. However, give the above problem I have, I'm not
> | sure I'm the best to do this. I did this for the sparse test code, and only
> | identified the strncmp problem above in ls-oct-ascii.{cc,h}.
> | 
> | Other major things in this patch include
> | 
> | * Make the calculation of the condition number optional and disable it for
> | the diagonal, triangular and banded solvers as the calculation of the
> | condition number takes longer that the solution of the linear equation in
> | this case (often by orders of magnitude). The underlying code shouldn't
> | allow singular matrices to be treated in this case, so I see this as a major
> | improvement
> | * Special case sparse permutations for Sparse<T>::index function, and gain
> | several orders of
> | magnitude of speed in this case.
> | * Dulmange-Mendelsohn sparse QR solver. This should give significantly
> | better solutions for over-determined problems.
> | * Many bug fixes for rectangular solvers.
> | 
> | Please find attached the patch and a complete changelog file..
>
> I can't comment on the Sparse stuff since I don't know much about it.
> You are the expert there, so go ahead and check in those changes.
>
> I don't really want the strncmp change because it seems to be a bug in
> strncmp since that function is supposed to compare at most N
> characters.  We should not have to include checks for that ourselves.
> In any case, how about this change?  Does it also avoid the problem?
> I also see one other call to strncmp in util.cc that could be
> converted to use the std::string compare method.
>
>   
Ok, will do..

> BTW, do you think it is time that we removed the *SSTREAM macros and
> just assume that the C++ compiler has sstream?  I would expect that
> any compiler that can build Octave will have that by now.
>
>   

Your call. Though I'd suggest to leave it till after 2.9.5.

D.