Re: Built-in base2dec and dec2base

[Daniel J Sebald]

On 08/20/2012 12:42 PM, Rik wrote:
> On 08/19/2012 10:41 PM, John W. Eaton wrote:
> > On 19-Aug-2012, Daniel J Sebald wrote:
> >
> > | I've gone over the builtin base2dec/dec2base code and created a second
> > | version of the patch file.
> >
> > I'm not sure that it is worth making these changes in core Octave.
> > Yes, performance is improved, but at the cost of code that is more
> > difficult to debug and maintain.
> 8/20/12
>
> Daniel,
>
> I will definitely take a look at your code, but I can only promise that
> I'll get to it before we make a new release.  I'm busy with real life
> projects and my Octave time is currently going towards the build system and
> the directory structure.

Thanks.  No rush.  That's why I put the patch on Savannah.  It's 
slightly too big to get to with limited time.


> As a further datapoint in the discussion, the m-file code I wrote is very
> specifically oriented towards performance and is reasonably opaque
> (although short).

Well, it's fairly impressive actually.  Efficiently compute from/to 
index sets for moving whitespaces to the left.  Create a table to map 
anything other than the symbols and ' ' to NAN hinging on the fact that 
math with a NAN results in NAN.  Nice.

BTW, looking at the code I have to think what is there is a wholesale 
rewrite, so that might warrant a change in the copyright notice.

The one issue with the m-script is that the cell version is so much 
slower.  I'm looking into some cell issues right now, but that's not 
such straightforward code.


>   Something in C++ might actually be clearer and more
> straightforward since one can just use a loop where a loop makes sense
> instead of employing the contortions I did to eliminate them.

Yes, that was the point I was attempting to make.

Dan