[GRASS-dev] QGIS winGRASS blank space problem

[Hamish]

Markus:
> > > a friend told me that with the latest QGIS release candidate,
> > > there is a winGRASS problem:
> > >
> > > $ v.rast.stats vector=CEA_riserve raster=CEA_slope
> > > colprefix=delete c:/Programmi/Quantum: c:/Programmi/Quantum: No
> > > such file or directory
Paul:
> > I notice there are loads of places in it that paths aren't quoted so
> > that is probably the first problem
Glynn:
> That's certainly a fundamental problem; $TMP and $SQLTMP appear to be
> the most likely candidates.
> 
> Regardless of Windows-specific issues, failure to quote variables
> which contain pathnames is a bug and needs to be fixed.

done in 6.3cvs.

> AFAICT, $SQLTMP is unnecessary. You can just as easily use a subshell
> with its output piped directly to db.execute; no need for a temporary
> file.

It is used in a loop- that would make it call db.execute nine times per
vector cat instead of once. db.execute is slow to start/stop, so it is
much faster to use a single call with $SQLTMP.


r.univar is the longest step in the loop, in the version I wrote before
there was a v.rast.stats, I did the loop like:
 r.mapcalc MASK=if(v_to_rast == $cat)
 g.region zoom=MASK
 r.univar v_to_rast
 g.region region=original
 g.remove MASK

not sure how much that sped it up, IIRC if the entire row is NULL then
raster ops should skip past it very quickly, and the only gain is from
the fewer rast column cells to process. "g.region zoom=" is not so fast
though and I never did any time trials to see. Actually when I think
about it, my data was buffers around points and I calculated the tmp
region from the central coord, not zoom=. In practice r.univar of the
full MASKed region may take the same time-to-run as g.region zoom=, so
no speedup for arbitrary polygon shapes. shrug.
 

Hamish