sort max/min results by y and x coordinate

previously the list of maxima or minima found by max/min were in found
order, so essentially random

they now sort equal values by y then x coordinate, so the results are
consistent between runs, provided there are fewer maxima / minima than
are being tracked

ChangeLog

@@ -29,6 +29,7 @@
 - fix return from C++ assignment operator overloads (+=, -= etc) 
 - add @max_slope to vips_hist_local() to implement CLAHE, thanks hunter-87
 - vips_gaussnoise() pixels are reproducible on recalc, thanks MvGulik
+- max/min sort values by y and x coordinate
 
 8/12/16 started 8.4.5
 - allow libgsf-1.14.26 to help centos, thanks tdiprima

libvips/arithmetic/max.c

@@ -23,6 +23,8 @@
  * 	- allow +/- INFINITY as a result
  * 4/12/12
  * 	- track and return top n values
+ * 24/1/17
+ * 	- sort equal values by y then x to make order more consistent
  */
 
 /*
@@ -143,10 +145,20 @@ vips_values_add( VipsValues *values, double v, int x, int y )
 
 	/* Find insertion point.
 	 */
-	for( i = 0; i < values->n; i++ )
-		if( v <= values->value[i] ) 
+	for( i = 0; i < values->n; i++ ) {
+		if( v < values->value[i] ) 
 			break;
 
+		if( v == values->value[i] ) {
+			if( y < values->y_pos[i] )
+				break;
+
+			if( y == values->y_pos[i] )
+				if( x <= values->x_pos[i] )
+					break;
+		}
+	}
+
 	/* Array full? 
 	 */
 	if( values->n == values->size ) {
@@ -498,6 +510,7 @@ vips_max_init( VipsMax *max )
  * By default it finds the single largest value. If @size is set >1, it will 
  * find the @size largest values. It will stop searching early if has found 
  * enough values. 
+ * Equal values will be sorted by y then x.
  *
  * It operates on all 
  * bands of the input image: use vips_stats() if you need to find an 
@@ -511,8 +524,7 @@ vips_max_init( VipsMax *max )
  * largest to smallest.
  *
  * If there are more than @size maxima, the maxima returned will be a random
- * selection of the maxima in the image. Equal maxima will be returned in a
- * random order.
+ * selection of the maxima in the image. 
  *
  * See also: vips_min(), vips_stats().
  *

libvips/arithmetic/min.c

@@ -145,10 +145,20 @@ vips_values_add( VipsValues *values, double v, int x, int y )
 
 	/* Find insertion point.
 	 */
-	for( i = 0; i < values->n; i++ )
-		if( v >= values->value[i] ) 
+	for( i = 0; i < values->n; i++ ) {
+		if( v > values->value[i] ) 
 			break;
 
+		if( v == values->value[i] ) {
+			if( y < values->y_pos[i] )
+				break;
+
+			if( y == values->y_pos[i] )
+				if( x <= values->x_pos[i] )
+					break;
+		}
+	}
+
 	/* Array full? 
 	 */
 	if( values->n == values->size ) {
@@ -501,6 +511,7 @@ vips_min_init( VipsMin *min )
  * By default it finds the single smallest value. If @size is set >1, it will 
  * find the @size smallest values. It will stop searching early if has found 
  * enough values. 
+ * Equal values will be sorted by y then x.
  *
  * It operates on all 
  * bands of the input image: use vips_stats() if you need to find an 
@@ -515,8 +526,7 @@ vips_min_init( VipsMin *min )
  * smallest to largest.
  *
  * If there are more than @size minima, the minima returned will be a random
- * selection of the minima in the image. Equal minima will be returned in a
- * random order.
+ * selection of the minima in the image. 
  *
  * See also: vips_min(), vips_stats().
  *

libvips/arithmetic/statistic.c

@@ -168,6 +168,7 @@ vips_statistic_class_init( VipsStatisticClass *class )
 		_( "Input image" ),
 		VIPS_ARGUMENT_REQUIRED_INPUT,
 		G_STRUCT_OFFSET( VipsStatistic, in ) );
+
 }
 
 static void