[gimp] Bug 742623: Equalize operation fills buffer with nan values in floating point precision

[Thomas Manni <tmanni src gnome org>]

commit 678b61a3aca7f4f6f1c2b2ffa239d60d4d4b71e3
Author: Thomas Manni <thomas manni free fr>
Date:   Tue Feb 10 12:07:45 2015 +0100

    Bug 742623: Equalize operation fills buffer with nan values in floating point precision
    
    Use histogram bins count to process equalization instead of "256" hard-coded value.

 app/operations/gimpoperationequalize.c |   42 ++++++++++++++++++++++++++-----
 app/operations/gimpoperationequalize.h |    3 +-
 2 files changed, 37 insertions(+), 8 deletions(-)
---
diff --git a/app/operations/gimpoperationequalize.c b/app/operations/gimpoperationequalize.c
index 5e4a3da..14dce6a 100644
--- a/app/operations/gimpoperationequalize.c
+++ b/app/operations/gimpoperationequalize.c
@@ -96,6 +96,8 @@ gimp_operation_equalize_class_init (GimpOperationEqualizeClass *klass)
 static void
 gimp_operation_equalize_init (GimpOperationEqualize *self)
 {
+  self->values = NULL;
+  self->n_bins = 0;
 }
 
 static void
@@ -103,6 +105,12 @@ gimp_operation_equalize_finalize (GObject *object)
 {
   GimpOperationEqualize *self = GIMP_OPERATION_EQUALIZE (object);
 
+  if (self->values)
+    {
+      g_free (self->values);
+      self->values = NULL;
+    }
+
   if (self->histogram)
     {
       g_object_unref (self->histogram);
@@ -143,15 +151,33 @@ gimp_operation_equalize_set_property (GObject      *object,
     case PROP_HISTOGRAM:
       if (self->histogram)
         g_object_unref (self->histogram);
+
       self->histogram = g_value_dup_object (value);
+
       if (self->histogram)
         {
           gdouble pixels;
+          gint    n_bins;
           gint    max;
           gint    k;
 
+          n_bins = gimp_histogram_n_bins (self->histogram);
+
+          if ((self->values != NULL) && (self->n_bins != n_bins))
+            {
+              g_free (self->values);
+              self->values = NULL;
+            }
+
+          if (self->values == NULL)
+            {
+              self->values = g_new (gdouble, 3 * n_bins);
+            }
+
+          self->n_bins = n_bins;
+
           pixels = gimp_histogram_get_count (self->histogram,
-                                             GIMP_HISTOGRAM_VALUE, 0, 255);
+                                             GIMP_HISTOGRAM_VALUE, 0, n_bins - 1);
 
           if (gimp_histogram_n_channels (self->histogram) == 1 ||
               gimp_histogram_n_channels (self->histogram) == 2)
@@ -164,7 +190,7 @@ gimp_operation_equalize_set_property (GObject      *object,
               gdouble sum = 0;
               gint    i;
 
-             for (i = 0; i < 256; i++)
+             for (i = 0; i < n_bins; i++)
                 {
                   gdouble histi;
 
@@ -172,12 +198,12 @@ gimp_operation_equalize_set_property (GObject      *object,
 
                   sum += histi;
 
-                  self->part[k][i] = sum / pixels;
+                  self->values[k * n_bins + i] = sum / pixels;
 
                   if (max == 1)
                     {
-                      self->part[1][i] = self->part[0][i];
-                      self->part[2][i] = self->part[0][i];
+                      self->values[n_bins + i] = self->values[i];
+                      self->values[2 * n_bins + i] = self->values[i];
                     }
                 }
            }
@@ -195,9 +221,11 @@ gimp_operation_equalize_map (GimpOperationEqualize *self,
                              gint                   component,
                              gfloat                 value)
 {
-  gint index = (gint) CLAMP (value * 255.0, 0, 255);
+  gint index;
+  index = component * self->n_bins + \
+            (gint) (CLAMP (value * (self->n_bins - 1), 0.0, self->n_bins - 1));
 
-  return self->part[component][index];
+  return self->values[index];
 }
 
 static gboolean
diff --git a/app/operations/gimpoperationequalize.h b/app/operations/gimpoperationequalize.h
index 30348a8..c6a7886 100644
--- a/app/operations/gimpoperationequalize.h
+++ b/app/operations/gimpoperationequalize.h
@@ -41,7 +41,8 @@ struct _GimpOperationEqualize
   GimpOperationPointFilter  parent_instance;
 
   GimpHistogram            *histogram;
-  gfloat                    part[5][256];
+  gdouble                  *values;
+  gint                      n_bins;
 };
 
 struct _GimpOperationEqualizeClass