Reduce the emitting of object definitions in ps and svg

jkseppan · jkseppan · commit ef2c9fbb871a · 2014-09-21T17:46:34.000+03:00
The cost calculations are very rough, back-of-the-envelope. In the example from #3345 I get significant file size reductions: 2.3M before.svg 1.8M after.svg 916K before.ps 672K after.ps
diff --git a/lib/matplotlib/backends/backend_ps.py b/lib/matplotlib/backends/backend_ps.py
@@ -637,6 +637,23 @@ def draw_path_collection(self, gc, master_transform, paths, all_transforms,
                              offsets, offsetTrans, facecolors, edgecolors,
                              linewidths, linestyles, antialiaseds, urls,
                              offset_position):
+        # Is the optimization worth it? Rough calculation:
+        # cost of emitting a path in-line is
+        #     (len_path + 2) * uses_per_path
+        # cost of definition+use is
+        #     (len_path + 3) + 3 * uses_per_path
+        len_path = len(paths[0].vertices) if len(paths) > 0 else 0
+        uses_per_path = self._iter_collection_uses_per_path(
+            paths, all_transforms, offsets, facecolors, edgecolors)
+        should_do_optimization = \
+            len_path + 3 * uses_per_path + 3 < (len_path + 2) * uses_per_path
+        if not should_do_optimization:
+            return RendererBase.draw_path_collection(
+                self, gc, master_transform, paths, all_transforms,
+                offsets, offsetTrans, facecolors, edgecolors,
+                linewidths, linestyles, antialiaseds, urls,
+                offset_position)
+
         write = self._pswriter.write
 
         path_codes = []
diff --git a/lib/matplotlib/backends/backend_svg.py b/lib/matplotlib/backends/backend_svg.py
@@ -607,6 +607,23 @@ def draw_path_collection(self, gc, master_transform, paths, all_transforms,
                              offsets, offsetTrans, facecolors, edgecolors,
                              linewidths, linestyles, antialiaseds, urls,
                              offset_position):
+        # Is the optimization worth it? Rough calculation:
+        # cost of emitting a path in-line is
+        #    (len_path + 5) * uses_per_path
+        # cost of definition+use is
+        #    (len_path + 3) + 9 * uses_per_path
+        len_path = len(paths[0].vertices) if len(paths) > 0 else 0
+        uses_per_path = self._iter_collection_uses_per_path(
+            paths, all_transforms, offsets, facecolors, edgecolors)
+        should_do_optimization = \
+            len_path + 9 * uses_per_path + 3 < (len_path + 5) * uses_per_path
+        if not should_do_optimization:
+            return RendererBase.draw_path_collection(
+                self, gc, master_transform, paths, all_transforms,
+                offsets, offsetTrans, facecolors, edgecolors,
+                linewidths, linestyles, antialiaseds, urls,
+                offset_position)
+
         writer = self.writer
         path_codes = []
         writer.start('defs')
