[GRASS5] r.thin lseek problems for large files

[Andrew Danner]

Thanks for the quick commit.  Yes, I suppose if you have more than
MAX_INT rows, bad things will happen, but with MAX_INT being over a
billion (sizeof(int)=4), I don't see this being a problem in the near
future. You could have a raster of the entire earth at 4cm resolution
before this becomes a problem. I think there will be other things that
break in GRASS and many other GIS packages before we hit this limit. 

I have rasters of about 30,000 rows and columns or larger and I'm
hitting the 4GB barrier pretty regularly and the lseeks are breaking. A
lot of them have been fixed already and a lot of lseeks just seek to the
beginning and do not need to be fixed, but there are still a few bad
ones around. 

examples of some suspicious looking lseeks are below. I have yet to use
any of these programs, so I don't know if the fix is needed or not
 r.drain/main.c:      lseek(fd,list->row*ncol*sizeof(CELL),SEEK_SET);
 r.drain/resolve.c:      lseek(fd,(nl-1)*bnd->sz,SEEK_SET);
 r.fill.dir/resolve.c:      lseek(fd,(nl-1)*bnd->sz,SEEK_SET);
 r.fill.dir/wtrshed.c:         lseek(fm,bas[i].offset,SEEK_SET);
 r.mfilter/getrow.c:     lseek (fd, (long) row*len, 0);

-Andy

On Wed, 2005-09-07 at 19:16 -0700, Brad Douglas wrote:
> On Wed, 2005-09-07 at 19:31 -0400, Andrew Danner wrote:
> > r.thin is segfaulting on large files because of a bad type in a couple
> > of lseeks.
> > 
> >  grass61.cvs/raster/r.thin/io.c
> >  
> >  lines 71 and 77 read
> > 
> >  lseek(file,(long) row * buf_len,0);
> > 
> >  but should be
> > 
> >  lseek(file, ((off_t)row) * buf_len,0);
> > 
> > can some make the change in CVS?
> 
> Done, but I really don't see it helping.
> 
> What happens if a map exceeds a number of rows greater than the max of
> type int?
> 
> Does it exit gracefully, or is it a disaster waiting to happen?
> lib/rowio is not designed to handle very large rows, AFAICT.
> 
> 
> PS -
> Take my words lightly.  My ammonia levels are in the triple digits of
> late and I am not thinking clearly.
> 
>