Count lines using git objects instead of filesystem.

This seemed to be roughly three times slower than a direct "wc -l" from working
tree, but this way we are not limited to full & clean working trees, but can
also handle bare repositories.

gitstats

@@ -323,15 +323,19 @@ class GitDataCollector(DataCollector):
 
 		# extensions
 		self.extensions = {} # extension -> files, lines
-		lines = getpipeoutput(['git ls-files']).split('\n')
+		lines = getpipeoutput(['git ls-tree -r -z HEAD']).split('\000')
 		self.total_files = len(lines)
 		for line in lines:
-			base = os.path.basename(line)
-			# Ignore extensionless (including .hidden files)
-			if base.find('.') == -1 or base.rfind('.') == 0:
+			if len(line) == 0:
+				continue
+			parts = re.split('\s+', line, 4)
+			sha1 = parts[2]
+			filename = parts[3]
+
+			if filename.find('.') == -1 or filename.rfind('.') == 0:
 				ext = ''
 			else:
-				ext = base[(base.rfind('.') + 1):]
+				ext = filename[(filename.rfind('.') + 1):]
 			if len(ext) > MAX_EXT_LENGTH:
 				ext = ''
 
@@ -340,8 +344,7 @@ class GitDataCollector(DataCollector):
 
 			self.extensions[ext]['files'] += 1
 			try:
-				# Escaping could probably be improved here
-				self.extensions[ext]['lines'] += int(getpipeoutput(['wc -l "%s"' % line]).split()[0])
+				self.extensions[ext]['lines'] += int(getpipeoutput(['git cat-file blob %s' % sha1, 'wc -l']).split()[0])
 			except:
 				print 'Warning: Could not count lines for file "%s"' % line