parsing: Fix auto-delete feature in escaped_split

escaped_split() was not handling empty matches correctly. It only
checked the first matching group (containing the text before the split
sequence), but ignored the second one, that consumes escaped escapes.
So if only the first group was empty, the algorithm decides the
complete entry is empty, but it needed to check the second group also.
This is now fixed with the help of a temporary string.

coalib/parsing/StringProcessing.py

@@ -95,17 +95,22 @@ def escaped_split(pattern,
     last_pos = 0
     # Process each returned MatchObject.
     for item in matches:
-        if (not remove_empty_matches or len(item.group(1)) != 0):
-            # Return the first matching group. The pattern from parameter can't
-            # change the group order.
-            match_strings.append(item.group(1))
-            # Concat it with the second group, that contains all escapes that
-            # are escaped and would get consumed.
-            if (item.group(2) is not None):
-                match_strings[-1] += item.group(2)
+        # Use a temporary concatenation string to check at last if it's empty.
+        concat_string = item.group(1)
 
-            # Update the end position.
-            last_pos = item.end()
+        if (item.group(2) is not None):
+            # If escaped escapes were consumed from the second group, append
+            # them too.
+            concat_string += item.group(2)
+
+        # If our temporary concatenation string is empty and the
+        # remove_empty_matches flag is specified, don't append it to the
+        # result.
+        if (not remove_empty_matches or len(concat_string) != 0):
+            match_strings.append(concat_string)
+
+        # Update the end position.
+        last_pos = item.end()
 
     # Append the rest of the string, since it's not in the result list (only
     # matches are captured that have a leading separator).