libgimp: fixing "palette" libgimp API unit testing.

It was an equality because I think the conversion from palette format to
"RGBA float" should go through the exact same babl code path, however it
is done in the API. And that is the case on my machine where I get
indeed perfect equality. But apparently not on the CI.

Though I guess investigating further would not be a bad idea, let's
consider this a minor issue (as a general rule, comparing float with
epsilon is still a recommended software usage), and just add a comment.

libgimp/tests/test-palette.c

@@ -5,6 +5,7 @@
 #define GIMP_TEST_COLOR_R_U8   72
 #define GIMP_TEST_COLOR_G_U8   56
 #define GIMP_TEST_COLOR_B_U8   56
+#define EPSILON                1e-6f
 
 static GimpValueArray *
 gimp_c_test_run (GimpProcedure        *procedure,
@@ -74,16 +75,24 @@ gimp_c_test_run (GimpProcedure        *procedure,
   format2     = babl_format ("RGBA float");
   format2_bpp = babl_format_get_bytes_per_pixel (format2);
   fcolormap   = (gfloat *) gimp_palette_get_colormap (palette, format2, &n_colormap_colors, &n_bytes);
-  GIMP_TEST_END(fcolormap != NULL && n_colormap_colors == n_colors  &&
+  GIMP_TEST_END(fcolormap != NULL && n_colormap_colors == n_colors &&
                 format2 != format                                  &&
                 format2_bpp > format_bpp                           &&
                 n_colormap_colors * format2_bpp == n_bytes)
 
   GIMP_TEST_START("Comparing fourth palette color's RGB components from colormap in float format")
   gegl_color_get_pixel (color, format2, rgba);
-  GIMP_TEST_END(fcolormap[4 * GIMP_TEST_COLOR_IDX] == rgba[0]                  &&
+  /* XXX: whether to a colormap or to a GeglColor, I was expecting the
-                fcolormap[4 * GIMP_TEST_COLOR_IDX + 1] == rgba[1] &&
+   * conversion from palette format to target format to be done through
-                fcolormap[4 * GIMP_TEST_COLOR_IDX + 2]== rgba[2])
+   * exact same babl code path, hence equality to be perfect (i.e. no
+   * epsilon needed for float comparison). This is true on my machine,
+   * not on CI. Is it again some SSE2 inconsistency where sometimes
+   * conversion goes to SSE2, sometimes not, for the same run and same
+   * babl fish? Something else? To be further investigated.
+   */
+  GIMP_TEST_END(fabsf (fcolormap[4 * GIMP_TEST_COLOR_IDX] - rgba[0]) < EPSILON     &&
+                fabsf (fcolormap[4 * GIMP_TEST_COLOR_IDX + 1] - rgba[1]) < EPSILON &&
+                fabsf (fcolormap[4 * GIMP_TEST_COLOR_IDX + 2] - rgba[2]) < EPSILON)
 
   g_free (fcolormap);
 

libgimp/tests/test-palette.py

@@ -9,6 +9,7 @@ GIMP_TEST_COLOR_FORMAT = "R'G'B' u8"
 GIMP_TEST_COLOR_R_U8   = 72
 GIMP_TEST_COLOR_G_U8   = 56
 GIMP_TEST_COLOR_B_U8   = 56
+EPSILON                = 1e-6
 
 pal = Gimp.Palette.get_by_name(GIMP_TEST_PALETTE)
 gimp_assert('gimp_palette_get_by_name()', type(pal) == Gimp.Palette)
@@ -52,19 +53,19 @@ gimp_assert('gimp_palette_get_colormap() in "RGBA float"',
 
 
 start = f2_bpp * GIMP_TEST_COLOR_IDX
-colormap_r = struct.unpack('f', colormap[start:start + 4])
+colormap_r = struct.unpack('f', colormap[start:start + 4])[0]
-colormap_g = struct.unpack('f', colormap[start + 4:start + 8])
+colormap_g = struct.unpack('f', colormap[start + 4:start + 8])[0]
-colormap_b = struct.unpack('f', colormap[start + 8:start + 12])
+colormap_b = struct.unpack('f', colormap[start + 8:start + 12])[0]
 
 rgb_bytes = colors[GIMP_TEST_COLOR_IDX].get_bytes(f2)
 rgb = rgb_bytes.get_data()
-palette_r = struct.unpack('f', rgb[:4])
+palette_r = struct.unpack('f', rgb[:4])[0]
-palette_g = struct.unpack('f', rgb[4:8])
+palette_g = struct.unpack('f', rgb[4:8])[0]
-palette_b = struct.unpack('f', rgb[8:12])
+palette_b = struct.unpack('f', rgb[8:12])[0]
 gimp_assert("Comparing fourth palette color's RGB components from colormap in float format",
-            colormap_r == palette_r and
+            abs(colormap_r - palette_r) < EPSILON and
-            colormap_g == palette_g and
+            abs(colormap_g - palette_g) < EPSILON and
-            colormap_b == palette_b)
+            abs(colormap_b - palette_b) < EPSILON)
 
 # Run the same tests through PDB: