[Qgis-user] Re: Sextante plugin works except with GRASS

[Victor Olaya]

> This module don't start if you don't have a raster in the Qgis TOC with an
> error like "Wrong or missing parameter values". This raster is required if
> you want a mask layer to be used but not always is the case and grass let
> you decide if use it or not.

> Then
> I only made some test with a raster used as mask but the module don't load
> output layers because of some cross validation problems..
> All these parameters seem to be optional in Grass but I think that sextante
> consider them obligatory.

I have just made the layer optional and uploaded the new description
to the SVN. Note that if you find errors like that, you can modify the
description of GRASS algorithms yourself and help us improve the
SEXTANTE-GRASS link. Read this to know how to do it:

http://sextante.googlecode.com/svn/trunk/soft/bindings/qgis-plugin/src/sextante/grass/grass.txt


> Then
> Grass module let you choose if you want to use layer 0 (i.e. 3D vector) or
> layer n (1 i.e. z column). In sextante it is assumed that you use layer 1
> and of course you have to specify z column name.

Do QGIS support 3D vector? I was not sure about that, so I assumed
that most likely for most users, z values would be in a separate field

> Then
> In Grass there are 10 output files, 6 raster and 4 vectorial and you can
> choose if you want them or not - at least one (in my case elev).
> Sextante elaborate each one of them (If I'm not wrong) and then let you
> choose if you want to add to the TOC.
> Would it be better if you can choose which one to elaborate (could be time
> spending) and then if you want to add it to the TOC?

That would be easy to do for the toolbox...but very hard for the
modeler :-( The modeler needs a well-defined sematic of the algorithm.
If the number or outputs depends on the user...that makes things
really hard to control...which means more work on the developer side.
I have decided to first strive for a robust modeler, which means taht
algorithms are more restricted (or, if they are not restricted, they
can be left out of the modeler, but I think in this case it is worth
to have that process available in the modeler)

By now, the only solution is to not load layers, which is just a
cosmetic trick, but does not add extra performance if a layer is not
added, because is generated anyway

> And, if you don't specify the name the layer is given a number as name, but
> it could be confusing if you let sextante work on more output than one more
> without specify each name.. So may be it would be better to concatenate a
> random number with the parameter name?

If you want names instead of filenames (those number are a temporary
filename), you can set it in the general settings, unckecking the "use
filename as layer name" box. Doing that, it will use the output name
(in this case, the GRASS output name)

> At the end I couldn't obtain any output, I attach some logs

should work now. Please, check the SVN version. There are now two
algorithms, one for the interpolation and one for the cross validation
(v.surf.rst.cvdev). I have, however, removed the "devi" output, since
it blocks grass when i use it... I have also removed the derived
outputs (slope, etc), to keep things more simple. All those can later
be extracted with another algorithm in SEXTNATE.

As I said, there is much work to do in polishing algorithm
descriptions. It is easy, but boring...so we are slowly advancing and
hopefully will soon have SEXTANTE working fine with all these external
applications.

Regards