RasterIO reporting corrupt blocks on AIG datasets

[grovduck]

When I do a simple RasterIO read over various ArcInfo? grids (using the AIG driver), I am getting sporadic "Corrupt block" errors from line 636 in gridlib.c (r17432) on grids that I think should be fine. This relevant chunk of code in gridlib.c is this:

/* The 4 is to be safe... In theory for a very small block (2x2 or 
 less), and nMinSize < 4, nDataSize could be smaller... */ 

if (nDataSize <  2 + 4) {
  CPLError( CE_Failure, CPLE_AppDefined, "Corrupt block");
  CPLFree( pabyRaw );
  return CE_Failure;
}

I'm erroring out with nDataSize of 4. I've posted a sample dataset which gives this error at:

 ftp://ftp.fsl.orst.edu/pub/gregorym/gdal/corrupt.zip (~108 MB)

[rouault]

Fixed in r17435. Works fine with corrupt.zip and also tested by hacking with an hexadecimal editor the abc3x1 in autotest/gdrivers/data to test small block sizes