WIP: Add auto-sheriff.appspot.com code to Blink

Tools/AutoSheriff/reasons.py

Index: Tools/AutoSheriff/reasons.py
diff --git a/Tools/AutoSheriff/reasons.py b/Tools/AutoSheriff/reasons.py
new file mode 100644
index 0000000000000000000000000000000000000000..149aac4d2fc1a83a8a451e34bfd9ababc0ea3b5d
--- /dev/null
+++ b/Tools/AutoSheriff/reasons.py
@@ -0,0 +1,317 @@
+# Copyright 2014 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.
+
+import json
+import logging
+import requests
+import sys
+import urllib
+import urlparse
+import argparse
+import re
+
+import requests_cache
+
+requests_cache.install_cache('reasons')
+
+
+# This is relative to build/scripts:
+# https://chromium.googlesource.com/chromium/tools/build/+/master/scripts
+BUILD_SCRIPTS_PATH = "/src/build/scripts"
+sys.path.append(BUILD_SCRIPTS_PATH)
+from common import gtest_utils
+
+# Python logging is stupidly verbose to configure.
+def setup_logging():
+    logger = logging.getLogger(__name__)
+    logger.setLevel(logging.DEBUG)
+    handler = logging.StreamHandler()
+    handler.setLevel(logging.DEBUG)
+    formatter = logging.Formatter('%(levelname)s: %(message)s')
+    handler.setFormatter(formatter)
+    logger.addHandler(handler)
+    return logger, handler
+
+
+log, logging_handler = setup_logging()
+
+
+def build_url(master_url, builder_name, build_number):
+  quoted_name = urllib.pathname2url(builder_name)
+  args = (master_url, quoted_name, build_number)
+  return "%s/builders/%s/builds/%s" % args
+
+
+def stdio_for_step(master_url, builder_name, build, step):
+# FIXME: Should get this from the step in some way?
+  base_url = build_url(master_url, builder_name, build['number'])
+  stdio_url = "%s/steps/%s/logs/stdio/text" % (base_url, step['name'])
+
+  try:
+    return requests.get(stdio_url).text
+  except requests.exceptions.ConnectionError, e:
+    # Some builders don't save logs for whatever reason.
+    log.error('Failed to fetch %s: %s' % (stdio_url, e))
+    return None
+
+
+# These are reason finders, more than splitters?
+class GTestSplitter(object):
+  def handles_step(self, step):
+    step_name = step['name']
+    # Silly heuristic, at least we won't bother processing
+    # stdio from gclient revert, etc.
+    if step_name.endswith('tests'):
+      return True
+
+    KNOWN_STEPS = [

[ojan, 2014/07/22 02:01:26]

I'd leave this out for now. Can always add it back in when we decide we need it.

+      # There are probably other gtest steps not named 'tests'.
+    ]
+    return step_name in KNOWN_STEPS
+
+  def split_step(self, step, build, builder_name, master_url):
+    stdio_log = stdio_for_step(master_url, builder_name, build, step)
+    # Can't split if we can't get the logs.
+    if not stdio_log:
+      return None
+
+    # Lines this fails for:
+    #[  FAILED  ] ExtensionApiTest.TabUpdate, where TypeParam =  and GetParam() =  (10907 ms)
+
+    log_parser = gtest_utils.GTestLogParser()
+    for line in stdio_log.split('\n'):
+      log_parser.ProcessLine(line)
+
+    failed_tests = log_parser.FailedTests()
+    if failed_tests:
+      return failed_tests
+    # Failed to split, just group with the general failures.
+    log.debug('First Line: %s' % stdio_log.split('\n')[0])
+    return None
+
+
+# Our Android tests produce very gtest-like output, but not
+# quite GTestLogParser-compatible (it parse the name of the
+# test as org.chromium).
+
+class JUnitSplitter(object):
+  def handles_step(self, step):
+    KNOWN_STEPS = [
+      'androidwebview_instrumentation_tests',
+      'mojotest_instrumentation_tests', # Are these always java?
+    ]
+    return step['name'] in KNOWN_STEPS
+
+  FAILED_REGEXP = re.compile('\[\s+FAILED\s+\] (?P<test_name>\S+)( \(.*\))?$')
+
+  def failed_tests_from_stdio(self, stdio):
+    failed_tests = []
+    for line in stdio.split('\n'):
+      match = self.FAILED_REGEXP.search(line)
+      if match:
+        failed_tests.append(match.group('test_name'))
+    return failed_tests
+
+  def split_step(self, step, build, builder_name, master_url):
+    stdio_log = stdio_for_step(master_url, builder_name, build, step)
+    # Can't split if we can't get the logs.
+    if not stdio_log:
+      return None
+
+    failed_tests = self.failed_tests_from_stdio(stdio_log)
+    if failed_tests:
+      return failed_tests
+    # Failed to split, just group with the general failures.
+    log.debug('First Line: %s' % stdio_log.split('\n')[0])
+    return None
+
+
+def decode_results(results, include_expected=False):
+    tests = convert_trie_to_flat_paths(results['tests'])
+    failures = {}
+    flakes = {}
+    passes = {}
+    for (test, result) in tests.iteritems():
+        if include_expected or result.get('is_unexpected'):
+            actual_results = result['actual'].split()
+            expected_results = result['expected'].split()
+            if len(actual_results) > 1:
+                if actual_results[1] in expected_results:
+                    flakes[test] = actual_results[0]

[ojan, 2014/07/22 02:01:26]

Can we just store result['actual'] here? That way the consuming tool can decide
that it wants to do with the result. Also, some test suites retry multiple
times.

+                else:
+                    # We report the first failure type back, even if the second
+                    # was more severe.
+                    failures[test] = actual_results[0]

[ojan, 2014/07/22 02:01:25]

Ditto

+            elif actual_results[0] == 'PASS':
+                passes[test] = result
+            else:
+                failures[test] = actual_results[0]
+
+    return (passes, failures, flakes)
+
+
+def convert_trie_to_flat_paths(trie, prefix=None):
+    # Cloned from webkitpy.layout_tests.layout_package.json_results_generator
+    # so that this code can stand alone.
+    result = {}
+    for name, data in trie.iteritems():
+        if prefix:
+            name = prefix + "/" + name
+
+        if len(data) and not "actual" in data and not "expected" in data:
+            result.update(convert_trie_to_flat_paths(data, name))
+        else:
+            result[name] = data
+
+    return result
+
+
+class LayoutTestsSplitter(object):
+  def handles_step(self, step):
+    return step['name'] == 'webkit_tests'
+
+  def split_step(self, step, build, builder_name, master_url):
+    # WTF?  The android bots call it archive_webkit_results and the rest call it archive_webkit_tests_results?
+    archive_names = ['archive_webkit_results', 'archive_webkit_tests_results']
+    archive_step = next((step for step in build['steps'] if step['name'] in archive_names), None)
+    url_to_build = build_url(master_url, builder_name, build['number'])
+
+    if not archive_step:
+      log.warn('No archive step in %s' % url_to_build)
+      # print json.dumps(build['steps'], indent=1)

[ojan, 2014/07/22 02:01:26]

Delete dead code?

+      return None
+
+    html_results_url = archive_step['urls'].get('layout test results')
+    # FIXME: Here again, Android is a special snowflake.
+    if not html_results_url:
+      html_results_url = archive_step['urls'].get('results')
+
+    if not html_results_url:
+      webkit_tests_step = next((step for step in build['steps'] if step['name'] == 'webkit_tests'), None)
+      # Common cause of this is an exception in the webkit_tests step.
+      if webkit_tests_step['results'][0] != 5:
+        log.warn('No results url for archive step in %s' % url_to_build)
+      # print json.dumps(archive_step, indent=1)

[ojan, 2014/07/22 02:01:25]

Ditto

+      return None
+
+    # !@?#!$^&$% WTF HOW DO URLS HAVE \r in them!?!

[ojan, 2014/07/22 02:01:26]

Wat

+    html_results_url = html_results_url.replace('\r', '')
+
+    jsonp_url = urlparse.urljoin(html_results_url, 'failing_results.json')
+    # FIXME: Silly that this is still JSONP.

[ojan, 2014/07/22 02:01:26]

This is because failing_results.json is used by results.html to show you the
failures when you run locally and chrome doesn't let you XHR files from file
urls. Open to suggestions to how to do it better. We could output two files from
run-webkit-tests I guess. FWIW, full_results.json is pure JSON. No P.

+    jsonp_string = requests.get(jsonp_url).text
+    if "The specified key does not exist" in jsonp_string:
+      log.warn('%s missing for %s' % (jsonp_url, url_to_build))
+      return None
+
+    json_string = jsonp_string[len('ADD_RESULTS('):-len(');')]
+    try:
+      results = json.loads(json_string)
+      passes, failures, flakes = decode_results(results)
+      if failures:
+        return ['%s:%s' % (name, types) for name, types in failures.items()]
+    except ValueError, e:
+      print archive_step['urls']
+      print html_results_url
+      print "Failed %s, %s at decode of: %s" % (jsonp_url, e, jsonp_string)
+
+    # Failed to split, just group with the general failures.
+    return None
+
+
+class CompileSplitter(object):
+  def handles_step(self, step):
+    return step['name'] == 'compile'
+
+# Compile example:

[ojan, 2014/07/22 02:01:26]

Even better than examples in comments is examples in tests. :)

+# FAILED: /mnt/data/b/build/goma/gomacc ...
+# ../../v8/src/base/platform/time.cc:590:7: error: use of undeclared identifier 'close'
+
+# Linker example:
+# FAILED: /b/build/goma/gomacc ...
+# obj/chrome/browser/extensions/interactive_ui_tests.extension_commands_global_registry_apitest.o:extension_commands_global_registry_apitest.cc:function extensions::SendNativeKeyEventToXDisplay(ui::KeyboardCode, bool, bool, bool): error: undefined reference to 'gfx::GetXDisplay()'
+
+  def split_step(self, step, build, builder_name, master_url):
+    stdio = stdio_for_step(master_url, builder_name, build, step)
+    # Can't split if we can't get the logs.
+    if not stdio:

[ojan, 2014/07/22 02:01:26]

Should we log an error or something here?

+      return None
+
+    compile_regexp = re.compile(r'(?P<path>.*):(?P<line>\d+):(?P<column>\d+): error:')
+
+    # FIXME: I'm sure there is a cleaner way to do this.
+    next_line_is_failure = False
+    for line in stdio.split('\n'):
+      if not next_line_is_failure:
+        if line.startswith('FAILED: '):
+          next_line_is_failure = True
+        continue
+
+      match = compile_regexp.match(line)
+      if match:
+        return ['%s:%s' % (match.group('path'), match.group('line'))]
+      break
+
+    return None
+
+
+# This is a hack I wrote because all the perf bots are failing with:
+# E    0.009s Main  File not found /b/build/slave/Android_GN_Perf/build/src/out/step_results/dromaeo.jslibstyleprototype
+# and it's nice to group them by something at least!
+# Often just hits:
+# 2 new files were left in c:\users\chrome~1.per\appdata\local\temp: Fix the tests to clean up themselves.
+# so disabled for now.
+class GenericRunTests(object):
+  def handles_step(self, step):
+    return True
+
+  def split_step(self, step, build, builder_name, master_url):
+    stdio = stdio_for_step(master_url, builder_name, build, step)
+    # Can't split if we can't get the logs.
+    if not stdio:
+      return None
+
+    last_line = None
+    for line in stdio.split('\n'):
+      if last_line and line.startswith('exit code (as seen by runtest.py):'):
+        return [last_line]
+      last_line = line
+
+
+STEP_SPLITTERS = [
+  CompileSplitter(),
+  LayoutTestsSplitter(),
+  JUnitSplitter(),
+  GTestSplitter(),
+  # GenericRunTests(),

[ojan, 2014/07/22 02:01:26]

Should we delete this if we're going to comment it out?

+]
+
+
+# For testing:
+def main(args):
+  parser = argparse.ArgumentParser()
+  parser.add_argument('stdio_url', action='store')
+  args = parser.parse_args(args)
+
+  # https://build.chromium.org/p/chromium.win/builders/XP%20Tests%20(1)/builds/31886/steps/browser_tests/logs/stdio
+  url_regexp = re.compile('(?P<master_url>.*)/builders/(?P<builder_name>.*)/builds/(?P<build_number>.*)/steps/(?P<step_name>.*)/logs/stdio')
+  match = url_regexp.match(args.stdio_url)
+  if not match:
+    print "Failed to parse URL: %s" % args.stdio_url
+    sys.exit(1)
+
+  step = {
+    'name': match.group('step_name'),
+  }
+  build = {
+    'number': match.group('build_number'),
+  }
+  splitter = next((splitter for splitter in STEP_SPLITTERS if splitter.handles_step(step)), None)
+  builder_name = urllib.unquote_plus(match.group('builder_name'))
+  master_url = match.group('master_url')
+  print splitter.split_step(step, build, builder_name, master_url)
+
+
+if __name__ == '__main__':
+  sys.exit(main(sys.argv[1:]))