While trying to make the gaussian-blur plugin work with different input formats, I encountered a bug that I'm getting different result depending on the output format. E.g. I'm getting, what seems to be tile boundary problems when using RGB that I don't see when using RGBA. Currently I have verified that I'm using the "iir" algorithm and that the channel pixels sums of the buffers are identical for the two different formats.

It seems that I'm getting a fade-off effect at the edges of the tiles when doing gegl_buffer_set() into a RGB buffer that I don't get when using RGBA. Is this a known problem? Should i file a bug about it?

Thanks, Dov

Ok, I filed the bug:

https://bugzilla.gnome.org/show_bug.cgi?id=701677

I tried reproducing it without my modified gaussian-blur.c , but unfortunately it didn't work. I still hope it is concise enough.

Regards, Dov

On Wed, Jun 5, 2013 at 12:43 PM, Daniel Sabo wrote:

Not a known problem with buffer_set. There is a known issue with process() if you write to a buffer that is also a source somewhere in the graph you will start to see result chunks in your input.

But filing a bug with a test case is always good, even for known issues, otherwise they tend to get forgotten.

Thanks, - Daniel