Re: Re; sparse merge

[David Bateman]

Per Persson wrote:

> On Mar 1, 2005, at 09:59, David Bateman wrote:
>
> > You have to put NOINSTALL in the octave-forge sparse directory. This 
> > is probably the cause of the other problems as well...
>
>
> Not all of them it seems..., breaks like before.
> Attaching log.
>
> /Per
Per,

I have little time to fully debug this at the moment, but what it seems 
like is that the function maybe_compress is deallocating memory, or 
accessing memory that wasn't allocated. This might be that either nnz, 
or the array c of SparseRep aren't correctly initialized, or the size of 
d and r of SparseRep don't correspond to the value of nnz. The way 
forward is to identify exactly the test in sptest.m that causes the 
problems, use gdb to step through maybe_delete and identify the line 
that causes the problems, identify the size of d, c and r and the value 
of nnz of SpareRep, then try and follow back from there once the value 
that is set incorrectly is identify, to where the value is set incorrectly.

The fact that this runs correctly on an x86 platform makes me wonder 
whether this is an endianness issue, but I don't see where that might be 
the case in the code...

Cheers
David