Add option to command tester for running a test multiple times. Use that

tools/process_perf_combined.py

+#!/usr/bin/python
+# Copyright (c) 2011 The Native Client Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import re
+import sys
+
+# Combine chromium-style perf log output from multiple runs.
+# Input is text *containing* the chrome buildbot perf format (may contain more)
+# but the output is only the merged perf data (throws away the rest).
+
+def ListToString(l):
+  return '[%s]' % (','.join(l))
+
+
+def Main():
+  usage = 'usage: %prog < stdin\n'
+  if len(sys.argv) != 1:
+    sys.stderr.write(usage)
+    sys.stderr.write('Instead, argv was %s\n' % str(sys.argv))
+    return 1
+  accumulated_times = {}
+  result_matcher = re.compile(r'^RESULT (.*): (.*)= (.*) (.*)$')
+  for line in sys.stdin.readlines():
+    match = result_matcher.match(line)
+    if match:
+      graph, trace, value, unit = match.groups()
+      key = (graph, trace)
+      value_list, old_unit = accumulated_times.get(key, ([], None))
+      if old_unit is not None:
+        assert(unit == old_unit)

[Nick Bray, 2011/11/11 22:42:39]

assert(unit == old_unit), (unit, old_unit) is more programmer-friendly because
it helps understand why the assert failed.

[jvoung - send to chromium..., 2011/11/11 23:28:27]

Done

+      if type(value) == list:

[Nick Bray, 2011/11/11 22:42:39]

isinstance(value, list)

[jvoung - send to chromium..., 2011/11/11 23:28:27]

Done.

+        value_list += value
+      else:
+        value_list += [value]
+      accumulated_times[key] = (value_list, unit)
+  for ((graph, trace), (values, unit)) in accumulated_times.iteritems():
+    sys.stdout.write('RESULT %s: %s= %s %s\n' %
+                     (graph, trace, ListToString(values), unit))
+  return 0
+
+
+if __name__ == '__main__':
+  sys.exit(Main())