Make a valgrind tool in the same vein as the purify one.

Currently supports the following commands:
  ./chrome_tests.sh --test base
  ./chrome_tests.sh --test net
  ./chrome_tests.sh --test ipc
  ./chrome_tests.sh --test unit

The valgrind_*.py files are my first real big chunks of python, so any suggestions would be greatly appreciated.

Review URL: http://codereview.chromium.org/16583


git-svn-id: svn://svn.chromium.org/chrome/trunk/src@7796 0039d316-1c4b-4281-b951-d872f2087c98

5 files changed, 863 insertions, 0 deletions

diff --git a/tools/valgrind/chrome_tests.py b/tools/valgrind/chrome_tests.py
new file mode 100755
index 0000000..d393353
--- /dev/null
+++ b/tools/valgrind/chrome_tests.py
@@ -0,0 +1,346 @@
+#!/usr/bin/python
+# Copyright (c) 2006-2008 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+# chrome_tests.py
+
+''' Runs various chrome tests through valgrind_test.py.
+
+This file is a copy of ../purify/chrome_tests.py. Eventually, it would be nice
+to merge these two files. For now, I'm leaving it here with sections that
+aren't supported commented out as this is more of a work in progress.
+'''
+
+import glob
+import logging
+import optparse
+import os
+import stat
+import sys
+
+import google.logging_utils
+import google.path_utils
+# Import the platform_utils up in the layout tests which have been modified to
+# work under non-Windows platforms instead of the ones that are in the
+# tools/python/google directory. (See chrome_tests.sh which sets PYTHONPATH
+# correctly.)
+#
+# TODO(erg): Copy/Move the relevant functions from the layout_package version
+# of platform_utils back up to google.platform_utils
+# package. http://crbug.com/6164
+import layout_package.platform_utils
+
+import common
+
+
+class TestNotFound(Exception): pass
+
+
+class ChromeTests:
+  '''This class is derived from the chrome_tests.py file in ../purify/.
+
+  TODO(erg): Finish implementing this. I've commented out all the parts that I
+  don't have working yet. We still need to deal with layout tests, and long
+  term, the UI tests.
+  '''
+
+  def __init__(self, options, args, test):
+    # the known list of tests
+    self._test_list = {
+#      "test_shell": self.TestTestShell,
+      "unit": self.TestUnit,
+      "net": self.TestNet,
+      "ipc": self.TestIpc,
+      "base": self.TestBase,
+#      "layout": self.TestLayout,
+#      "layout_all": self.TestLayoutAll,
+#      "ui": self.TestUI
+    }
+
+    if test not in self._test_list:
+      raise TestNotFound("Unknown test: %s" % test)
+
+    self._options = options
+    self._args = args
+    self._test = test
+
+    script_dir = google.path_utils.ScriptDir()
+    utility = layout_package.platform_utils.PlatformUtility(script_dir)
+    # Compute the top of the tree (the "source dir") from the script dir (where
+    # this script lives).  We assume that the script dir is in tools/valgrind/
+    # relative to the top of the tree.
+    self._source_dir = os.path.dirname(os.path.dirname(script_dir))
+    # since this path is used for string matching, make sure it's always
+    # an absolute Windows-style path
+    self._source_dir = utility.GetAbsolutePath(self._source_dir)
+    valgrind_test = os.path.join(script_dir, "valgrind_test.py")
+    self._command_preamble = ["python", valgrind_test, "--echo_to_stdout",
+                              "--source_dir=%s" % (self._source_dir)]
+
+  def _DefaultCommand(self, module, exe=None):
+    '''Generates the default command array that most tests will use.'''
+    module_dir = os.path.join(self._source_dir, module)
+
+    # For now, the suppressions files are all the same across all modules. Copy
+    # the code in the purify version of chrome_tests.py if we ever need
+    # per-module suppressions again...
+    self._data_dir = google.path_utils.ScriptDir()
+
+    if not self._options.build_dir:
+      dir_chrome = os.path.join(self._source_dir, "chrome", "Hammer")
+      dir_module = os.path.join(module_dir, "Hammer")
+      if exe:
+        exe_chrome = os.path.join(dir_chrome, exe)
+        exe_module = os.path.join(dir_module, exe)
+        if os.path.isfile(exe_chrome) and not os.path.isfile(exe_module):
+          self._options.build_dir = dir_chrome
+        elif os.path.isfile(exe_module) and not os.path.isfile(exe_chrome):
+          self._options.build_dir = dir_module
+        elif (os.stat(exe_module)[stat.ST_MTIME] >
+              os.stat(exe_chrome)[stat.ST_MTIME]):
+          self._options.build_dir = dir_module
+        else:
+          self._options.build_dir = dir_chrome
+      else:
+        if os.path.isdir(dir_chrome) and not os.path.isdir(dir_module):
+          self._options.build_dir = dir_chrome
+        elif os.path.isdir(dir_module) and not os.path.isdir(dir_chrome):
+          self._options.build_dir = dir_module
+        elif (os.stat(dir_module)[stat.ST_MTIME] >
+              os.stat(dir_chrome)[stat.ST_MTIME]):
+          self._options.build_dir = dir_module
+        else:
+          self._options.build_dir = dir_chrome
+
+    cmd = list(self._command_preamble)
+    cmd.append("--data_dir=%s" % self._data_dir)
+    if self._options.baseline:
+      cmd.append("--baseline")
+    if self._options.verbose:
+      cmd.append("--verbose")
+    if self._options.generate_suppressions:
+      cmd.append("--generate_suppressions")
+    if exe:
+      cmd.append(os.path.join(self._options.build_dir, exe))
+    return cmd
+
+  def Run(self):
+    ''' Runs the test specified by command-line argument --test '''
+    logging.info("running test %s" % (self._test))
+    return self._test_list[self._test]()
+
+  def _ReadGtestFilterFile(self, name, cmd):
+    '''Read a file which is a list of tests to filter out with --gtest_filter
+    and append the command-line option to cmd.
+    '''
+    filters = []
+    filename = os.path.join(self._data_dir, name + ".gtest.txt")
+    if os.path.exists(filename):
+      f = open(filename, 'r')
+      for line in f.readlines():
+        if line.startswith("#") or line.startswith("//") or line.isspace():
+          continue
+        line = line.rstrip()
+        filters.append(line)
+    gtest_filter = self._options.gtest_filter
+    if len(filters):
+      if gtest_filter:
+        gtest_filter += ":"
+        if gtest_filter.find("-") < 0:
+          gtest_filter += "-"
+      else:
+        gtest_filter = "-"
+      gtest_filter += ":".join(filters)
+    if gtest_filter:
+      cmd.append("--gtest_filter=%s" % gtest_filter)
+
+  def SimpleTest(self, module, name):
+    cmd = self._DefaultCommand(module, name)
+    self._ReadGtestFilterFile(name, cmd)
+    return common.RunSubprocess(cmd, 0)
+
+  def ScriptedTest(self, module, exe, name, script, multi=False, cmd_args=None,
+                   out_dir_extra=None):
+    '''Valgrind a target binary, which will be executed one or more times via a
+       script or driver program.
+    Args:
+      module - which top level component this test is from (webkit, base, etc.)
+      exe - the name of the exe (it's assumed to exist in build_dir)
+      name - the name of this test (used to name output files)
+      script - the driver program or script.  If it's python.exe, we use
+        search-path behavior to execute, otherwise we assume that it is in
+        build_dir.
+      multi - a boolean hint that the exe will be run multiple times, generating
+        multiple output files (without this option, only the last run will be
+        recorded and analyzed)
+      cmd_args - extra arguments to pass to the valgrind_test.py script
+    '''
+    cmd = self._DefaultCommand(module)
+    exe = os.path.join(self._options.build_dir, exe)
+    cmd.append("--exe=%s" % exe)
+    cmd.append("--name=%s" % name)
+    if multi:
+      out = os.path.join(google.path_utils.ScriptDir(),
+                         "latest")
+      if out_dir_extra:
+        out = os.path.join(out, out_dir_extra)
+        if os.path.exists(out):
+          old_files = glob.glob(os.path.join(out, "*.txt"))
+          for f in old_files:
+            os.remove(f)
+        else:
+          os.makedirs(out)
+      out = os.path.join(out, "%s%%5d.txt" % name)
+      cmd.append("--out_file=%s" % out)
+    if cmd_args:
+      cmd.extend(cmd_args)
+    if script[0] != "python.exe" and not os.path.exists(script[0]):
+      script[0] = os.path.join(self._options.build_dir, script[0])
+    cmd.extend(script)
+    self._ReadGtestFilterFile(name, cmd)
+    return common.RunSubprocess(cmd, 0)
+
+  def TestBase(self):
+    return self.SimpleTest("base", "base_unittests")
+
+  def TestIpc(self):
+    return self.SimpleTest("chrome", "ipc_tests")
+
+  def TestNet(self):
+    return self.SimpleTest("net", "net_unittests")
+
+  def TestTestShell(self):
+    return self.SimpleTest("webkit", "test_shell_tests")
+
+  def TestUnit(self):
+    return self.SimpleTest("chrome", "unit_tests")
+
+#   def TestLayoutAll(self):
+#     return self.TestLayout(run_all=True)
+
+#   def TestLayout(self, run_all=False):
+#     # A "chunk file" is maintained in the local directory so that each test
+#     # runs a slice of the layout tests of size chunk_size that increments with
+#     # each run.  Since tests can be added and removed from the layout tests at
+#     # any time, this is not going to give exact coverage, but it will allow us
+#     # to continuously run small slices of the layout tests under purify rather
+#     # than having to run all of them in one shot.
+#     chunk_num = 0
+#     # Tests currently seem to take about 20-30s each.
+#     chunk_size = 120  # so about 40-60 minutes per run
+#     chunk_file = os.path.join(os.environ["TEMP"], "purify_layout_chunk.txt")
+#     if not run_all:
+#       try:
+#         f = open(chunk_file)
+#         if f:
+#           str = f.read()
+#           if len(str):
+#             chunk_num = int(str)
+#           # This should be enough so that we have a couple of complete runs
+#           # of test data stored in the archive (although note that when we loop
+#           # that we almost guaranteed won't be at the end of the test list)
+#           if chunk_num > 10000:
+#             chunk_num = 0
+#           f.close()
+#       except IOError, (errno, strerror):
+#         logging.error("error reading from file %s (%d, %s)" % (chunk_file,
+#                       errno, strerror))
+
+#     script = os.path.join(self._source_dir, "webkit", "tools", "layout_tests",
+#                           "run_webkit_tests.py")
+#     script_cmd = ["python.exe", script, "--run-singly", "-v",
+#                   "--noshow-results", "--time-out-ms=200000",
+#                   "--nocheck-sys-deps"]
+#     if not run_all:
+#       script_cmd.append("--run-chunk=%d:%d" % (chunk_num, chunk_size))
+
+#     if len(self._args):
+#       # if the arg is a txt file, then treat it as a list of tests
+#       if os.path.isfile(self._args[0]) and self._args[0][-4:] == ".txt":
+#         script_cmd.append("--test-list=%s" % self._args[0])
+#       else:
+#         script_cmd.extend(self._args)
+
+#     if run_all:
+#       ret = self.ScriptedTest("webkit", "test_shell.exe", "layout",
+#                               script_cmd, multi=True, cmd_args=["--timeout=0"])
+#       return ret
+
+#     # store each chunk in its own directory so that we can find the data later
+#     chunk_dir = os.path.join("layout", "chunk_%05d" % chunk_num)
+#     ret = self.ScriptedTest("webkit", "test_shell.exe", "layout",
+#                             script_cmd, multi=True, cmd_args=["--timeout=0"],
+#                             out_dir_extra=chunk_dir)
+
+#     # Wait until after the test runs to completion to write out the new chunk
+#     # number.  This way, if the bot is killed, we'll start running again from
+#     # the current chunk rather than skipping it.
+#     try:
+#       f = open(chunk_file, "w")
+#       chunk_num += 1
+#       f.write("%d" % chunk_num)
+#       f.close()
+#     except IOError, (errno, strerror):
+#       logging.error("error writing to file %s (%d, %s)" % (chunk_file, errno,
+#                     strerror))
+#     # Since we're running small chunks of the layout tests, it's important to
+#     # mark the ones that have errors in them.  These won't be visible in the
+#     # summary list for long, but will be useful for someone reviewing this bot.
+#     return ret
+
+#   def TestUI(self):
+#     if not self._options.no_reinstrument:
+#       instrumentation_error = self.InstrumentDll()
+#       if instrumentation_error:
+#         return instrumentation_error
+#     return self.ScriptedTest("chrome", "chrome.exe", "ui_tests",
+#                              ["ui_tests.exe",
+#                               "--single-process",
+#                               "--ui-test-timeout=120000",
+#                               "--ui-test-action-timeout=80000",
+#                               "--ui-test-action-max-timeout=180000"],
+#                              multi=True)
+
+
+def _main(argv):
+  parser = optparse.OptionParser("usage: %prog -b <dir> -t <test> "
+                                 "[-t <test> ...]")
+  parser.disable_interspersed_args()
+  parser.add_option("-b", "--build_dir",
+                    help="the location of the output of the compiler output")
+  parser.add_option("-t", "--test", action="append",
+                    help="which test to run")
+  parser.add_option("", "--baseline", action="store_true", default=False,
+                    help="generate baseline data instead of validating")
+  parser.add_option("", "--gtest_filter",
+                    help="additional arguments to --gtest_filter")
+  parser.add_option("-v", "--verbose", action="store_true", default=False,
+                    help="verbose output - enable debug log messages")
+  parser.add_option("", "--no-reinstrument", action="store_true", default=False,
+                    help="Don't force a re-instrumentation for ui_tests")
+  parser.add_option("", "--generate_suppressions", action="store_true",
+                    default=False,
+                    help="Skip analysis and generate suppressions")
+
+  options, args = parser.parse_args()
+
+  if options.verbose:
+    google.logging_utils.config_root(logging.DEBUG)
+  else:
+    google.logging_utils.config_root()
+
+  if not options.test or not len(options.test):
+    parser.error("--test not specified")
+
+  for t in options.test:
+    tests = ChromeTests(options, args, t)
+    ret = tests.Run()
+    if ret: return ret
+  return 0
+
+
+if __name__ == "__main__":
+  ret = _main(sys.argv)
+  sys.exit(ret)
+
diff --git a/tools/valgrind/chrome_tests.sh b/tools/valgrind/chrome_tests.sh
new file mode 100755
index 0000000..e854252
--- /dev/null
+++ b/tools/valgrind/chrome_tests.sh
@@ -0,0 +1,4 @@
+#!/bin/sh
+
+export THISDIR=`dirname $0`
+PYTHONPATH=$THISDIR/../../webkit/tools/layout_tests/:$THISDIR/../purify:$THISDIR/../python "./chrome_tests.py" "$@"
diff --git a/tools/valgrind/suppressions.txt b/tools/valgrind/suppressions.txt
new file mode 100644
index 0000000..d142686
--- /dev/null
+++ b/tools/valgrind/suppressions.txt
@@ -0,0 +1,183 @@
+{
+   Uninitialized value in deflate
+   Memcheck:Cond
+   fun:longest_match
+   fun:deflate_slow
+   fun:MOZ_Z_deflate
+}
+{
+   dlopen leak; not our problem.
+   Memcheck:Leak
+   fun:malloc
+   fun:_dl_map_object_from_fd
+   fun:_dl_map_object
+   fun:dl_open_worker
+   fun:_dl_catch_error
+   fun:_dl_open
+   fun:dlopen_doit
+   fun:_dl_catch_error
+   fun:_dlerror_run
+   fun:dlopen@@GLIBC_2.1
+   fun:PR_LoadLibraryWithFlags
+   obj:*
+}
+{
+   Leak inside GTK?
+   Memcheck:Leak
+   fun:calloc
+   fun:g_malloc0
+   obj:/usr/lib32/libgobject-2.0.so.0.1600.3
+   obj:/usr/lib32/libgobject-2.0.so.0.1600.3
+   fun:g_type_register_fundamental
+   obj:/usr/lib32/libgobject-2.0.so.0.1600.3
+   fun:g_type_init_with_debug_flags
+   fun:g_type_init
+   fun:gdk_pre_parse_libgtk_only
+   obj:/usr/lib32/libgtk-x11-2.0.so.0.1200.9
+   fun:g_option_context_parse
+   fun:gtk_parse_args
+}
+{
+   Leak inside GTK?
+   Memcheck:Leak
+   fun:memalign
+   fun:posix_memalign
+   obj:/usr/lib32/libglib-2.0.so.0.1600.3
+   fun:g_slice_alloc
+   fun:g_slice_alloc0
+   fun:g_type_create_instance
+   obj:/usr/lib32/libgobject-2.0.so.0.1600.3
+   fun:g_object_newv
+   fun:g_object_new_valist
+   fun:g_object_new
+   fun:gdk_display_open
+   fun:gdk_display_open_default_libgtk_only
+}
+{
+   Fontconfig leak?
+   Memcheck:Leak
+   fun:malloc
+   obj:/usr/lib32/libfontconfig.so.1.3.0
+   obj:/usr/lib32/libfontconfig.so.1.3.0
+   obj:/usr/lib32/libfontconfig.so.1.3.0
+   obj:/usr/lib32/libfontconfig.so.1.3.0
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   fun:XML_ParseBuffer
+   fun:FcConfigParseAndLoad
+   fun:FcConfigParseAndLoad
+   obj:/usr/lib32/libfontconfig.so.1.3.0
+   obj:/usr/lib32/libfontconfig.so.1.3.0
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   fun:XML_ParseBuffer
+   fun:FcConfigParseAndLoad
+   fun:FcInitLoadConfig
+   fun:FcInitLoadConfigAndFonts
+   fun:FcInit
+}
+{
+   Fontconfig Leak?
+   Memcheck:Leak
+   fun:malloc
+   fun:FcStrCopy
+   obj:/usr/lib32/libfontconfig.so.1.3.0
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   fun:XML_ParseBuffer
+   fun:FcConfigParseAndLoad
+   fun:FcConfigParseAndLoad
+   obj:/usr/lib32/libfontconfig.so.1.3.0
+   obj:/usr/lib32/libfontconfig.so.1.3.0
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   obj:/usr/lib32/libexpat.so.1.5.2
+   fun:XML_ParseBuffer
+   fun:FcConfigParseAndLoad
+   fun:FcInitLoadConfig
+   fun:FcInitLoadConfigAndFonts
+   fun:FcInit
+}
+{
+   SECMOD leak
+   Memcheck:Leak
+   fun:calloc
+   fun:PR_Calloc
+   obj:/usr/lib32/nss/libnssckbi.so
+   obj:/usr/lib32/nss/libnssckbi.so
+   obj:/usr/lib32/nss/libnssckbi.so
+   obj:/usr/lib32/nss/libnssckbi.so
+   obj:/usr/lib32/nss/libnssckbi.so
+   obj:/usr/lib32/nss/libnssckbi.so
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   fun:SECMOD_LoadModule
+   fun:SECMOD_LoadUserModule
+}
+{
+   SECMOD leak?
+   Memcheck:Leak
+   fun:malloc
+   fun:PR_Malloc
+   fun:PL_ArenaAllocate
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   fun:SECMOD_LoadUserModule
+}
+{
+   NSS leak part 3.
+   Memcheck:Leak
+   fun:calloc
+   fun:PR_Calloc
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   obj:/usr/lib32/libnss3.so.1d
+   fun:NSS_NoDB_Init
+}
+{
+   GTK leak?
+   Memcheck:Leak
+   fun:calloc
+   fun:g_malloc0
+   obj:/usr/lib32/libgobject-2.0.so.0.1600.3
+   obj:/usr/lib32/libgobject-2.0.so.0.1600.3
+   fun:g_type_register_fundamental
+   obj:/usr/lib32/libgobject-2.0.so.0.1600.3
+   fun:g_type_init_with_debug_flags
+   fun:g_type_init
+   fun:gdk_pre_parse_libgtk_only
+   obj:/usr/lib32/libgtk-x11-2.0.so.0.1200.9
+   fun:g_option_context_parse
+   fun:gtk_parse_args
+   fun:gtk_init_check
+}
+{
+   (Probable memory leak in Skia)
+   Memcheck:Leak
+   fun:_Znwj
+   fun:_ZN10SkFontHost15ResolveTypefaceEj
+   fun:_ZN24SkScalerContext_FreeType9setupSizeEv
+   fun:_ZN24SkScalerContext_FreeType19generateFontMetricsEPN7SkPaint11FontMetricsES2_
+   fun:_ZN15SkScalerContext14getFontMetricsEPN7SkPaint11FontMetricsES2_
+   fun:_ZN12SkGlyphCacheC1EPK12SkDescriptor
+   fun:_ZN12SkGlyphCache10VisitCacheEPK12SkDescriptorPFbPKS_PvES5_
+}
+{
+   Error in ICU
+   Memcheck:Overlap
+   fun:memcpy
+   fun:init_resb_result
+}
diff --git a/tools/valgrind/valgrind_analyze.py b/tools/valgrind/valgrind_analyze.py
new file mode 100755
index 0000000..c73f6bf
--- /dev/null
+++ b/tools/valgrind/valgrind_analyze.py
@@ -0,0 +1,176 @@
+#!/usr/bin/python
+# Copyright (c) 2006-2008 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+# valgrind_analyze.py
+
+''' Given a valgrind XML file, parses errors and uniques them.'''
+
+import logging
+import optparse
+import os
+import sys
+from xml.dom.minidom import parse
+
+# These are functions (using C++ mangled names) that we look for in stack
+# traces. We don't show stack frames while pretty printing when they are below
+# any of the following:
+_TOP_OF_STACK_POINTS = [
+  # Don't show our testing framework.
+  "testing::Test::Run()",
+  # Also don't show the internals of libc/pthread.
+  "start_thread"
+]
+
+def getTextOf(top_node, name):
+  ''' Returns all text in all DOM nodes with a certain |name| that are children
+  of |top_node|.
+  '''
+
+  text = ""
+  for nodes_named in top_node.getElementsByTagName(name):
+    text += "".join([node.data for node in nodes_named.childNodes
+                     if node.nodeType == node.TEXT_NODE])
+  return text
+
+def removeCommonRoot(source_dir, directory):
+  '''Returns a string with the string prefix |source_dir| removed from
+  |directory|.'''
+  if source_dir:
+    # Do this for safety, just in case directory is an absolute path outside of
+    # source_dir.
+    prefix = os.path.commonprefix([source_dir, directory])
+    return directory[len(prefix) + 1:]
+
+  return directory
+
+# Constants that give real names to the abbreviations in valgrind XML output.
+INSTRUCTION_POINTER = "ip"
+OBJECT_FILE = "obj"
+FUNCTION_NAME = "fn"
+SRC_FILE_DIR = "dir"
+SRC_FILE_NAME = "file"
+SRC_LINE = "line"
+
+class ValgrindError:
+  ''' Takes a <DOM Element: error> node and reads all the data from it. A
+  ValgrindError is immutable and is hashed on its pretty printed output.
+  '''
+
+  def __init__(self, source_dir, error_node):
+    ''' Copies all the relevant information out of the DOM and into object
+    properties.
+
+    Args:
+      error_node: The <error></error> DOM node we're extracting from.
+      source_dir: Prefix that should be stripped from the <dir> node.
+    '''
+
+    self._kind = getTextOf(error_node, "kind")
+    self._what = getTextOf(error_node, "what")
+
+    self._frames = []
+    stack_node = error_node.getElementsByTagName("stack")[0]
+
+    for frame in stack_node.getElementsByTagName("frame"):
+      frame_dict = {
+        INSTRUCTION_POINTER : getTextOf(frame, INSTRUCTION_POINTER),
+        OBJECT_FILE         : getTextOf(frame, OBJECT_FILE),
+        FUNCTION_NAME       : getTextOf(frame, FUNCTION_NAME),
+        SRC_FILE_DIR        : removeCommonRoot(
+            source_dir, getTextOf(frame, SRC_FILE_DIR)),
+        SRC_FILE_NAME       : getTextOf(frame, SRC_FILE_NAME),
+        SRC_LINE            : getTextOf(frame, SRC_LINE)
+      }
+
+      self._frames += [frame_dict]
+
+      if frame_dict[FUNCTION_NAME] in _TOP_OF_STACK_POINTS:
+        break
+
+  def __str__(self):
+    ''' Pretty print the type and stack frame of this specific error.'''
+    output = self._kind + "\n"
+    for frame in self._frames:
+      output += ("  " + (frame[FUNCTION_NAME] or frame[INSTRUCTION_POINTER]) +
+                 " (")
+
+      if frame[SRC_FILE_DIR] != "":
+        output += (frame[SRC_FILE_DIR] + "/" + frame[SRC_FILE_DIR] + ":" +
+                   frame[SRC_LINE])
+      else:
+        output += frame[OBJECT_FILE]
+      output += ")\n"
+
+    return output
+
+  def UniqueString(self):
+    ''' String to use for object identity. Don't print this, use str(obj)
+    instead.'''
+    rep = self._kind + " "
+    for frame in self._frames:
+      rep += frame[FUNCTION_NAME]
+
+      if frame[SRC_FILE_DIR] != "":
+        rep += frame[SRC_FILE_DIR] + "/" + frame[SRC_FILE_NAME]
+      else:
+        rep += frame[OBJECT_FILE]
+
+    return rep
+
+  def __hash__(self):
+    return hash(self.UniqueString())
+  def __eq__(self, rhs):
+    return self.UniqueString() == rhs
+
+class ValgrindAnalyze:
+  ''' Given a set of Valgrind XML files, parse all the errors out of them,
+  unique them and output the results.'''
+
+  def __init__(self, source_dir, files):
+    '''Reads in a set of files.
+
+    Args:
+      source_dir: Path to top of source tree for this build
+      files: A list of filenames.
+    '''
+
+    self._errors = set()
+    for file in files:
+      raw_errors = parse(file).getElementsByTagName("error")
+      for raw_error in raw_errors:
+        self._errors.add(ValgrindError(source_dir, raw_error))
+
+  def Report(self):
+    if self._errors:
+      logging.error("FAIL! There were %s errors: " % len(self._errors))
+
+      for error in self._errors:
+        logging.error(error)
+
+      return -1
+
+    logging.info("PASS! No errors found!")
+    return 0
+
+def _main():
+  '''For testing only. The ValgrindAnalyze class should be imported instead.'''
+  retcode = 0
+  parser = optparse.OptionParser("usage: %prog [options] <files to analyze>")
+  parser.add_option("", "--source_dir",
+                    help="path to top of source tree for this build"
+                    "(used to normalize source paths in baseline)")
+
+  (options, args) = parser.parse_args()
+  if not len(args) >= 1:
+    parser.error("no filename specified")
+  filenames = args
+
+  analyzer = ValgrindAnalyze(options.source_dir, filenames)
+  retcode = analyzer.Report()
+
+  sys.exit(retcode)
+
+if __name__ == "__main__":
+  _main()
diff --git a/tools/valgrind/valgrind_test.py b/tools/valgrind/valgrind_test.py
new file mode 100755
index 0000000..1a6f363
--- /dev/null
+++ b/tools/valgrind/valgrind_test.py
@@ -0,0 +1,154 @@
+#!/usr/bin/python
+# Copyright (c) 2006-2008 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+# purify_test.py
+
+'''Runs an exe through Valgrind and puts the intermediate files in a
+directory.
+'''
+
+import datetime
+import glob
+import logging
+import optparse
+import os
+import re
+import shutil
+import sys
+import time
+
+import google.path_utils
+
+import common
+
+import valgrind_analyze
+
+rmtree = shutil.rmtree
+
+class Valgrind():
+  TMP_DIR = "valgrind.tmp"
+
+  def __init__(self):
+    self._data_dir = None
+
+  def CreateOptionParser(self):
+    self._parser = optparse.OptionParser("usage: %prog [options] <program to "
+                                         "test>")
+    self._parser.add_option("-e", "--echo_to_stdout",
+                      dest="echo_to_stdout", action="store_true", default=False,
+                      help="echo purify output to standard output")
+    self._parser.add_option("-t", "--timeout",
+                      dest="timeout", metavar="TIMEOUT", default=10000,
+                      help="timeout in seconds for the run (default 10000)")
+    self._parser.add_option("", "--source_dir",
+                            help="path to top of source tree for this build"
+                                 "(used to normalize source paths in baseline)")
+    self._parser.add_option("", "--data_dir", default=".",
+                            help="path to where purify data files live")
+    self._parser.add_option("", "--generate_suppressions", action="store_true",
+                            default=False,
+                            help="Skip analysis and generate suppressions")
+    self._parser.description = __doc__
+
+  def ParseArgv(self):
+    self.CreateOptionParser()
+    self._options, self._args = self._parser.parse_args()
+    self._timeout = int(self._options.timeout)
+    self._data_dir = self._options.data_dir
+    self._generate_suppressions = self._options.generate_suppressions
+    self._source_dir = self._options.source_dir
+    return True
+
+  def Setup(self):
+    return self.ParseArgv()
+
+  def Execute(self):
+    ''' Execute the app to be tested after successful instrumentation.
+    Full execution command-line provided by subclassers via proc.'''
+    logging.info("starting execution...")
+    # note that self._args begins with the exe to be run
+    proc = ["valgrind", "--smc-check=all", "--leak-check=full",
+            "--track-origins=yes", "--num-callers=30"]
+
+    # Either generate suppressions or load them.
+    if self._generate_suppressions:
+      proc += ["--gen-suppressions=all"]
+    else:
+      proc += ["--xml=yes"]
+
+    suppressions = os.path.join(self._data_dir, "suppressions.txt")
+    if os.path.exists(suppressions):
+      proc += ["--suppressions=%s" % suppressions]
+    else:
+      logging.warning("WARNING: NOT USING SUPPRESSIONS!")
+
+    proc += ["--log-file=" + self.TMP_DIR + "/valgrind.%p"] + self._args
+
+    # If we have a valgrind.tmp directory, we failed to cleanup last time.
+    if os.path.exists(self.TMP_DIR):
+      shutil.rmtree(self.TMP_DIR)
+    os.mkdir(self.TMP_DIR)
+    common.RunSubprocess(proc, self._timeout)
+
+    # Always return true, even if running the subprocess failed. We depend on
+    # Analyze to determine if the run was valid. (This behaviour copied from
+    # the purify_test.py script.)
+    return True
+
+  def Analyze(self):
+    # Glob all the files in the "valgrind.tmp" directory
+    filenames = glob.glob(self.TMP_DIR + "/valgrind.*")
+    analyzer = valgrind_analyze.ValgrindAnalyze(self._source_dir, filenames)
+    analyzer.Report()
+    return 1
+
+  def Cleanup(self):
+    # Right now, we can cleanup by deleting our temporary directory. Other
+    # cleanup is still a TODO?
+    shutil.rmtree(self.TMP_DIR)
+    return True
+
+  def RunTestsAndAnalyze(self):
+    self.Execute()
+    if self._generate_suppressions:
+      logging.info("Skipping analysis to let you look at the raw output...")
+      return 0
+
+    retcode = self.Analyze()
+    if retcode:
+      logging.error("Analyze failed.")
+      return retcode
+    logging.info("Execution and analysis completed successfully.")
+    return 0
+
+  def Main(self):
+    '''Call this to run through the whole process: Setup, Execute, Analyze'''
+    start = datetime.datetime.now()
+    retcode = -1
+    if self.Setup():
+      retcode = self.RunTestsAndAnalyze()
+
+      # Skip cleanup on generate.
+      if not self._generate_suppressions:
+        self.Cleanup()
+    else:
+      logging.error("Setup failed")
+    end = datetime.datetime.now()
+    seconds = (end - start).seconds
+    hours = seconds / 3600
+    seconds = seconds % 3600
+    minutes = seconds / 60
+    seconds = seconds % 60
+    logging.info("elapsed time: %02d:%02d:%02d" % (hours, minutes, seconds))
+    return retcode
+
+
+
+if __name__ == "__main__":
+  valgrind = Valgrind()
+  retcode = valgrind.Main()
+  sys.exit(retcode)
+
+