[R-sig-Geo] Problems freeing rgdal driver...

[Jonathan Greenberg]

A few months back I was getting some help with adapting R for doing 
tiled processing of remote sensing data, and I made some serious 
headway, but there is a lingering problem I keep scratching my head at.  
First off, a quick recap of the base algorithm:

1) Let i = input raster, f() some function to apply to i, and o the 
output raster, such that o = f(i)
2) i is read in "tiles", each tile read from the input we'll call i_tile 
(to make it simplest, assume I'm just reading i one line at a time) -- 
this process is performed using readGDAL(...offset=c(j, 0),...) where j 
is the current line I'm reading.  o_tile = f(i_tile), which is written 
using *writeBin* (not writeGDAL).
3) This produces a flat binary file of the sort that Arc binary rasters 
or ENVI rasters use.

So I produce the binary file, but the portion I'm getting a bit hung up 
on is the production of the little text header which completes the file 
(.hdr/.prj for an ESRI raster, .hdr for an ENVI raster, note that the 
.hdr is different between arc and envi).  For this example, let's try to 
make an ENVI header.  Remember i_tile is one of the readGDAL objects of 
a tiled subset of image of filename i:

***

  gp <- gridparameters(i_tile)
  cellsize <- gp$cellsize
  offset <- gp$cellcentre.offset
  dims <- gp$cells.dim
  dims[[2]]=rows
  nbands <- 1
  tempopen <- GDAL.open(i)
  file_projection <- getProjectionRef(tempopen)
  GDAL.close(tempopen)
# This was from Roger Bivand
  d.drv <- new("GDALDriver", "ENVI")
  tds.out <- new("GDALTransientDataset", driver = d.drv, rows = 
dims[2],cols = dims[1], bands = nbands, type = "Float32")
  gt <- c(offset[1] - 0.5 * cellsize[1], cellsize[1], 0.0,offset[2] + 
(dims[2] -0.5) * cellsize[2], 0.0, -cellsize[2])
  .Call("RGDAL_SetGeoTransform", tds.out, gt, PACKAGE = "rgdal")
  .Call("RGDAL_SetProject", tds.out, file_projection, PACKAGE = "rgdal")
  fn <- tempfile()
  saveDataset(tds.out, fn)
  GDAL.close(tds.out)
  GDAL.close(d.drv)
 
  temporary_dir=dirname(fn)
  outdir=dirname(outfile)
 
  fnhdr<-as.character(paste(basename(fn),outfile_hdr,sep=''))
  outhdr<-as.character(paste(basename(rasteroutput),outfile_hdr,sep=''))

  setwd(temporary_dir)

  file.rename(fnhdr,outhdr)
  file.copy(outhdr,outdir,overwrite=TRUE)

***

So this WORKS the first time I run it, but the SECOND time it is run 
(without quitting out of R) I get the following error:
 >   d.drv <- new("GDALDriver", "ENVI")
Error in .local(.Object, ...) : No driver registered with name: ENVI

It seems like there is some lingering open driver or something 
someplace, but for the life of me I can't tell where it is.  You can see 
that I did a GDAL.close(d.drv) but that doesn't seem to be sufficient.  
Thoughts?

By the way, the full algorithm would deal with RS data of an arbitrary 
number of samples, lines and bands (e.g. hyperspectral processing would 
be completely feasible).  I'm happy to send the script to the list once 
the bugs are worked out!  Thanks!

--j