[GRASS-dev] Grass SVN in Android, display issue

[Glynn Clements]

Sören Gebbert wrote:

> > The ideal solution would be to store runs of >= 254^2 as multiple
> > runs. That would maintain backward compatibility while avoiding the
> > bug. But that would require re-writing Rast3d_rle_encode() and
> > Rast3d_rle_count_only(); there's no way that G_rle_codeLength and
> > rle_length2code can be fixed, as they assume that any length can be
> > encoded.
> 
> So this bug is not fixable at all?

The bug is fixable, but the fix isn't trivial.

> >> However, i do not understand the advantage of having RLE first and a
> >> LZ77 or LZ78 compression afterwards.
> >> Shouldn't be the LZ77/78 compression sufficient for compression of
> >> float/double values?
> >
> > For sparse data, using RLE first can often provide a noticeable
> > improvement.
> 
> I would prefer to disable RLE and keep the code only for backward compatibility.

That's certainly simpler, but you should first check how much
difference it makes to a sparse (e.g. all-zeros) map.

-- 
Glynn Clements <glynn at gclements.plus.com>