This regression test requires perl, bash, and some programs from
netpbm: pngtopnm, pnmdepth and pgmtoppm.

The expected outputs are stored as PNG files, which compress better
than just gzipped PxM, but the output from xcf2png is not compared
byte-to-byte with the stored files, because different libpng and zlib
versions may take different, but legitimate, decisions about IDAT
chunk division and compression. Instead pixel comparisions are
done with cmp(1) on PxM representations, which have less possibility
for variations.

The test of xcftopnm assumes that the layout of the ascii header
in the output from xcftopnm matches byte-for-byte what netpbm produces.
There is an undocumented -@ switch that does this, but if you have a
wildly different version of netpbm you may need to fiddle with the
implementation of -@ and/or insert filters in the test driver
routines.


The test asssumes that xcftools has been compiled with an
iconv() library that can expand non-ASCII characters to
approximating ASCII sequenced, possibly triggered by the
'//TRANSLIT' convention of glibc. If this is not the case
you have to do appropriate corrections when the 'AE=AE' layer
is mentioned by the test script.


There are yet no tests for correct reporting of errors.



Note: display(1) from imagemagick 6.0.6.2 apparently does not display
  PNG files with grayscale+alpha correctly - the alpha channel is
  ignored.