Post-processing python phenotypes is slow

[MihaiBabiac]

The python_filter function which converts {:, :} and {::} into newlines and indentation is slower than it has to be. It takes a significant amount of time when the operations used and the evaluation is fast. Replacing the per-char for loop with calls to builtins and a per-line for loop is faster.

See my changes below:

--- a/PonyGE2/src/utilities/representation/python_filter.py
+++ b/PonyGE2/src/utilities/representation/python_filter.py
@@ -5,20 +5,21 @@ def python_filter(txt):
     it's not possible to specify indentation correctly in a BNF
     grammar without this type of scheme."""
 
+    # Split input into lines
+    split_code = (txt
+                  .replace("{::}", "\n")
+                  .replace("{:", "{:\n")
+                  .replace(":}", ":}\n")
+                  .split("\n"))
+
+    new_line_list = []
     indent_level = 0
-    tmp = txt[:]
-    i = 0
-    while i < len(tmp):
-        tok = tmp[i:i + 2]
-        if tok == "{:":
-            indent_level += 1
-        elif tok == ":}":
-            indent_level -= 1
-        tabstr = "\n" + "  " * indent_level
-        if tok == "{:" or tok == ":}":
-            tmp = tmp.replace(tok, tabstr, 1)
-        i += 1
-    # Strip superfluous blank lines.
-    txt = "\n".join([line for line in tmp.split("\n")
-                     if line.strip() != ""])
-    return txt
+    for line in split_code:
+        if line != ":}" and line != "":
+            clean_line = line.replace("{:", "").replace(":}", "")
+            new_line_list.append("  " * indent_level + clean_line)
+        indent_level += line.count("{:")
+        indent_level -= line.count(":}")
+
+    new_code = "\n".join(new_line_list)
+    return new_code