test: runner support for flaky tests

Adding --flaky-tests option, to allow regarding flaky tests failures
as non-fatal.

Currently only observed by the TapProgressIndicator, which will
add a # TODO directive to tests classified as flaky. According to the
TAP specification, the test harness is supposed to treat failures
that have a # TODO directive as non-fatal.

PR-URL: #25686
Reviewed-By: Julien Gilli <[email protected]>

Author: orangemocha

tools/test.py

@@ -55,8 +55,9 @@
 
 class ProgressIndicator(object):
 
-  def __init__(self, cases):
+  def __init__(self, cases, flaky_tests_mode):
     self.cases = cases
+    self.flaky_tests_mode = flaky_tests_mode
     self.queue = Queue(len(cases))
     for case in cases:
       self.queue.put_nowait(case)
@@ -234,13 +235,19 @@ def HasRun(self, output):
     self._done += 1
     command = basename(output.command[-1])
     if output.UnexpectedOutput():
-      print 'not ok %i - %s' % (self._done, command)
+      status_line = 'not ok %i - %s' % (self._done, command)
+      if FLAKY in output.test.outcomes and self.flaky_tests_mode == "dontcare":
+        status_line = status_line + " # TODO : Fix flaky test"
+      print status_line
       for l in output.output.stderr.splitlines():
         print '#' + l
       for l in output.output.stdout.splitlines():
         print '#' + l
     else:
-      print 'ok %i - %s' % (self._done, command)
+      status_line = 'ok %i - %s' % (self._done, command)
+      if FLAKY in output.test.outcomes:
+        status_line = status_line + " # TODO : Fix flaky test"
+      print status_line
 
     duration = output.test.duration
 
@@ -258,8 +265,8 @@ def Done(self):
 
 class CompactProgressIndicator(ProgressIndicator):
 
-  def __init__(self, cases, templates):
-    super(CompactProgressIndicator, self).__init__(cases)
+  def __init__(self, cases, flaky_tests_mode, templates):
+    super(CompactProgressIndicator, self).__init__(cases, flaky_tests_mode)
     self.templates = templates
     self.last_status_length = 0
     self.start_time = time.time()
@@ -314,29 +321,29 @@ def PrintProgress(self, name):
 
 class ColorProgressIndicator(CompactProgressIndicator):
 
-  def __init__(self, cases):
+  def __init__(self, cases, flaky_tests_mode):
     templates = {
       'status_line': "[%(mins)02i:%(secs)02i|\033[34m%%%(remaining) 4d\033[0m|\033[32m+%(passed) 4d\033[0m|\033[31m-%(failed) 4d\033[0m]: %(test)s",
       'stdout': "\033[1m%s\033[0m",
       'stderr': "\033[31m%s\033[0m",
     }
-    super(ColorProgressIndicator, self).__init__(cases, templates)
+    super(ColorProgressIndicator, self).__init__(cases, flaky_tests_mode, templates)
 
   def ClearLine(self, last_line_length):
     print "\033[1K\r",
 
 
 class MonochromeProgressIndicator(CompactProgressIndicator):
 
-  def __init__(self, cases):
+  def __init__(self, cases, flaky_tests_mode):
     templates = {
       'status_line': "[%(mins)02i:%(secs)02i|%%%(remaining) 4d|+%(passed) 4d|-%(failed) 4d]: %(test)s",
       'stdout': '%s',
       'stderr': '%s',
       'clear': lambda last_line_length: ("\r" + (" " * last_line_length) + "\r"),
       'max_length': 78
     }
-    super(MonochromeProgressIndicator, self).__init__(cases, templates)
+    super(MonochromeProgressIndicator, self).__init__(cases, flaky_tests_mode, templates)
 
   def ClearLine(self, last_line_length):
     print ("\r" + (" " * last_line_length) + "\r"),
@@ -738,8 +745,8 @@ def GetVmFlags(self, testcase, mode):
   def GetTimeout(self, mode):
     return self.timeout * TIMEOUT_SCALEFACTOR[mode]
 
-def RunTestCases(cases_to_run, progress, tasks):
-  progress = PROGRESS_INDICATORS[progress](cases_to_run)
+def RunTestCases(cases_to_run, progress, tasks, flaky_tests_mode):
+  progress = PROGRESS_INDICATORS[progress](cases_to_run, flaky_tests_mode)
   return progress.Run(tasks)
 
 
@@ -763,6 +770,7 @@ def BuildRequirements(context, requirements, mode, scons_flags):
 TIMEOUT = 'timeout'
 CRASH = 'crash'
 SLOW = 'slow'
+FLAKY = 'flaky'
 
 
 class Expression(object):
@@ -1212,6 +1220,9 @@ def BuildOptions():
       default=False, action="store_true")
   result.add_option("--cat", help="Print the source of the tests",
       default=False, action="store_true")
+  result.add_option("--flaky-tests",
+      help="Regard tests marked as flaky (run|skip|dontcare)",
+      default="run")
   result.add_option("--warn-unused", help="Report unused rules",
       default=False, action="store_true")
   result.add_option("-j", help="The number of parallel tasks to run",
@@ -1258,6 +1269,13 @@ def ProcessOptions(options):
     options.scons_flags.append("arch=" + options.arch)
   if options.snapshot:
     options.scons_flags.append("snapshot=on")
+  def CheckTestMode(name, option):
+    if not option in ["run", "skip", "dontcare"]:
+      print "Unknown %s mode %s" % (name, option)
+      return False
+    return True
+  if not CheckTestMode("--flaky-tests", options.flaky_tests):
+    return False
   return True
 
 
@@ -1450,15 +1468,16 @@ def wrap(processor):
 
   result = None
   def DoSkip(case):
-    return SKIP in case.outcomes or SLOW in case.outcomes
+    return (SKIP in case.outcomes or SLOW in case.outcomes or 
+           (FLAKY in case.outcomes and options.flaky_tests == "skip"))
   cases_to_run = [ c for c in all_cases if not DoSkip(c) ]
   if len(cases_to_run) == 0:
     print "No tests to run."
-    return 0
+    return 1
   else:
     try:
       start = time.time()
-      if RunTestCases(cases_to_run, options.progress, options.j):
+      if RunTestCases(cases_to_run, options.progress, options.j, options.flaky_tests):
         result = 0
       else:
         result = 1