Re: SparseMatrix constructor

[c.]

On 19 Oct 2013, at 19:24, Daniel J Sebald <address@hidden> wrote:

> On 10/19/2013 03:28 AM, c. wrote:
>> 
>> On 18 Oct 2013, at 18:22, c.<address@hidden>  wrote:
>> 
>>> Hi,
>>> 
>>> I have some code[1] that uses the constructor
>>> 
>>>          SparseMatrix::SparseMatrix (Array<double>  V, 
>>> Array<octave_idx_type>  I,
>>>                                      Array<octave_idx_type>  J, 
>>> octave_idx_type nr,
>>>                                      octave_idx_type nc, bool add);
>>> 
>>> where I and J are 0-based arrays of indices, which works in Octave 3.6.x
>>> but fails with development sources with the error:
>>> 
>>>  error: op_u_v: subscript indices must be either positive integers less 
>>> than 2^31 or logicals
>>> 
>>> have there been changes in the SparseMatrix constructors that can explain 
>>> this change?
>>> Can anyone point me to the changeset that caused the different behaviour?
>>> 
>>> c.
>>> 
>>> [1] 
>>> http://sourceforge.net/p/geopdes/code/HEAD/tree/trunk/geopdes_base/src/op_u_v.cc
>>> [2]         
>> 
>> Hi,
>> 
>> Just an update, I'm trying to find the changeset that introduced this 
>> modification.
>> 
>> I am doing this by bisect and my current "good" release is the following:
>> 
>>   changeset:   17465:77bec442a35a
>>   user:        Rik<address@hidden>
>>   date:        Wed Sep 18 10:32:27 2013 -0700
>>   summary:     Overhaul stem family of plot functions.
>> 
>> Bisecting is quite cumbersome in this case as I have to install each time in 
>> order to use
>> "mkoctfile" and "pkg install" so if anyone is aware of a changeset in the 
>> last month that
>> might have caused this problem I'd really appreciate hints.
>> 
>> Marco, I am CC-ing you because I believe this problem might affect your code 
>> as well,
>> could you please test your package with a recent build of Octave development 
>> sources?
> 
> One should be able to narrow this down by using the source control diff 
> utilities of the repository.  Mercurial's standard front end server browser 
> isn't geared for this, but it is possible to advance backward through the 
> parents of a file and do diffs...at least those two features exist, but 
> perhaps not conveniently combined.  The constructor is in :
> 
> [sebald@ octave]$ grep SparseMatrix::SparseMatrix * -r
> [snip the ChangeLog entries]
> liboctave/array/dSparse.cc:SparseMatrix::SparseMatrix (const SparseBoolMatrix 
> &a)
> liboctave/array/dSparse.cc:SparseMatrix::SparseMatrix (const DiagMatrix& a)
> 
> and here is that source file in the repository:
> 
> http://hg.savannah.gnu.org/hgweb/octave/file/9bb5d3f63cdd/liboctave/array/dSparse.cc
> 
> Click on the "diff" link and click on the "parents" link in various ways and 
> I believed you'll be able to trace the recent history of dSparse.cc with just 
> the differences.  There probably is a way to do this at the command line on 
> the repository on your computer, but at least the HTML version looks a little 
> better if perhaps more clumsy.
> 
> Dan

Dan, 

Thanks, but in the meanwhile the bisect process has completed and now I see why 
I was not able
to see what had been changed about sparse indexing ...

It's because nothing had actually changed ;)

It seems the changeset that broke my code is this one:

  changeset:   17542:d0fa3568eb5d
  user:        John W. Eaton <address@hidden>
  date:        Thu Sep 26 15:28:33 2013 -0400
  summary:     update links in Help menu

which apparently removed the definition of the macro OCTAVE_API_VERSION_NUMBER 
which was used in my code
to maintain compatibility with Octave versions older than 3.4:

  
http://sourceforge.net/p/geopdes/code/HEAD/tree/trunk/geopdes_base/src/op_u_v.cc#l65

Rafael, I think the easiest way to fix geopdes is to just drop compatibility 
with the 3.2 version ;)

Marco this also explains why your "assemble" method is unaffected, sorry for 
making you worry about nothing ...

c.