The format of the string data used in the imageop module is described

as "This is the same format as used by gl.lrectwrite() and the imgfile
module."  This implies a certain byte order in multi-byte pixel
formats.  However, the code was originally written on an SGI
(big-endian) and *uses* the fact that bytes are stored in a particular
order in ints.  This means that the code uses and produces different
byte order on little-endian systems.

This fix adds a module-level flag "backward_compatible" (default not
set, and if not set, behaves as if set to 1--i.e. backward compatible)
that can be used on a little-endian system to use the same byte order
as the SGI.  Using this flag it is then possible to prepare
SGI-compatible images on a little-endian system.

This patch is the result of a (small) discussion on python-dev and was
submitted to SourceForge as patch #874358.

1 files changed, 12 insertions, 0 deletions

diff --git a/Doc/lib/libimageop.tex b/Doc/lib/libimageop.tex
index 0f611e7..27ed056 100644
--- a/Doc/lib/libimageop.tex
+++ b/Doc/lib/libimageop.tex
@@ -86,3 +86,15 @@ Convert a 4-bit greyscale image to an 8-bit greyscale image.
 \begin{funcdesc}{grey22grey}{image, width, height}
 Convert a 2-bit greyscale image to an 8-bit greyscale image.
 \end{funcdesc}
+
+\begin{datadesc}{backward_compatible}
+If set to 0, the functions in this module use a non-backward
+compatible way of representing multi-byte pixels on little-endian
+systems.  The SGI for which this module was originally written is a
+big-endian system, so setting this variable will have no effect.
+However, the code wasn't originally intended to run on anything else,
+so it made assumptions about byte order which are not universal.
+Setting this variable to 0 will cause the byte order to be reversed on
+little-endian systems, so that it then is the same as on big-endian
+systems.
+\end{datadesc}