[Gimp-developer] Fwd: GIMP Levels operation fix

[David Gowers]

Hi,  I sent this simple fix to mitch, however I'm too impatient to
wait for mitch to check his inbox,
So I'm now passing it onto this mailing list.
Basically it fixes over/underflow in the Levels op; with the attached
image, the bug gives the appearance of a triangle wave repeating
gradient, instead of a readjusted single gradient for the channel
being adjusted.


-- Forwarded message --
From: David Gowers [EMAIL PROTECTED]
Date: Jan 31, 2008 5:45 PM
Subject: GIMP Levels operation
To: Michael Natterer [EMAIL PROTECTED]


I've just found a bug in the Levels op -- if the selected low/hi input
endpoints of a channel are
(low - more than the lowest non-empty histogram entry; or high -less
than the highest non-empty histogram entry for the channel)
then underflow/overflow happens.
Adjusting any of the input range sliders to fall into that range will
show this behaviour up.
Using the channels dialog to disable all but one channel also helps. I
would say you have simply forgotten to clip the input values before
mapping them to output values. *pokes around* indeed, I've attached a
very short patch that performs correct clipping. With this patch, GEGL
levels op matches old levels behaviour in all tests I could contrive.
attachment: testimage.pngIndex: app/gegl/gimpoperationlevels.c
===
--- app/gegl/gimpoperationlevels.c	(revision 24754)
+++ app/gegl/gimpoperationlevels.c	(working copy)
@@ -87,6 +87,8 @@
   else
 value = (value - low_input);
 
+  value = CLAMP (value, 0.0, 1.0 );
+
   if (gamma != 0.0)
 {
   if (value = 0.0)
___
Gimp-developer mailing list
Gimp-developer@lists.XCF.Berkeley.EDU
https://lists.XCF.Berkeley.EDU/mailman/listinfo/gimp-developer

Re: [Gimp-developer] Fwd: GIMP Levels operation fix

[Sven Neumann]

Hi,

On Tue, 2008-02-05 at 17:50 +1030, David Gowers wrote:

 Basically it fixes over/underflow in the Levels op;

Your approach is not the right way to fix this. There is absolutely
nothing wrong with under- and overflows if we are using GEGL. The
problem you are seeing is that some of the float-8bit conversions in
babl are not handling colors outside the 0..1 range properly. You can
fix this by running GIMP with the environment variable BABL_ERROR set to
0.0. This will select correct, but slower, conversion routines and you
will notice that this fixes the error.

We need to fix this in babl. Conversions that don't deal correctly with
values outside the [0.0;1.0] inverval need to be assigned a large error
so that they are not being selected.


Sven


___
Gimp-developer mailing list
Gimp-developer@lists.XCF.Berkeley.EDU
https://lists.XCF.Berkeley.EDU/mailman/listinfo/gimp-developer