Moving ScheduleNewAnalysis to break the cycle

appengine/findit/crash/findit.py

 # Copyright 2016 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 copy
 import logging
 
 from google.appengine.ext import ndb
 
 from common import appengine_util
 from common import chrome_dependency_fetcher
 from common import constants
 from common import time_util
 from crash.crash_report import CrashReport
 from crash.culprit import NullCulprit
 from model import analysis_status
 from model.crash.crash_config import CrashConfig
 
 # TODO(http://crbug.com/659346): since most of our unit tests are
 # FinditForFracas-specific, wrengr moved them to findit_for_chromecrash_test.py.
 # However, now we're missing coverage for most of this file (due to the
 # buggy way coverage is computed). Need to add a bunch of new unittests
 # to get coverage back up.
 
 # TODO: this class depends on ndb stuff, and should therefore move to
 # cr-culprit-finder/service/predator as part of the big reorganization.
 # Also, this class should be renamed to "PredatorApp" (alas, "Azalea"
 # was renamed to "Predator").
 class Findit(object):
-  def __init__(self, repository, pipeline_cls):
+  def __init__(self, repository):
     """
     Args:
       repository (Repository): the Git repository for getting CLs to classify.
-      pipeline_cls (class): the class for constructing pipelines in
-        ScheduleNewAnalysis. This will almost surely be
-        |crash.crash_pipeline.CrashWrapperPipeline|; but we must pass
-        the class in as a parameter in order to break an import cycle.
     """
     self._repository = repository
     # TODO(http://crbug.com/659354): because self.client is volatile,
     # we need some way of updating the Azelea instance whenever the
     # config changes. How to do that cleanly?
     self._azalea = None
     self._stacktrace_parser = None
-    self._pipeline_cls = pipeline_cls
 
   # This is a class method because it should be the same for all
   # instances of this class. We can in fact call class methods on
   # instances (not just on the class itself), so we could in principle
   # get by with just this method. However, a @classmethod is treated
   # syntactically like a method, thus we'd need to have the |()| at the
   # end, unlike for a @property. Thus we have both the class method and
   # the property, in order to simulate a class property.
   @classmethod
   def _ClientID(cls): # pragma: no cover
@@ skipping 130 matching lines @@
     model.regression_range = crash_data.get('regression_range', None)
 
     # Set progress properties.
     model.status = analysis_status.PENDING
     model.requested_time = time_util.GetUTCNow()
 
   @ndb.transactional
   def _NeedsNewAnalysis(self, crash_data):
     raise NotImplementedError()
 
-  # TODO(http://crbug.com/659345): try to break the cycle re pipeline_cls.
-  # TODO(http://crbug.com/659346): we don't cover anything after the
-  # call to _NeedsNewAnalysis.
-  def ScheduleNewAnalysis(self, crash_data,
-      queue_name=constants.DEFAULT_QUEUE): # pragma: no cover
-    """Create a pipeline object to perform the analysis, and start it.
-
-    If we can detect that the analysis doesn't need to be performed
-    (e.g., it was already performed, or the |crash_data| is empty so
-    there's nothig we can do), then we will skip creating the pipeline
-    at all.
-
-    Args:
-      crash_data (JSON): ??
-      queue_name (??): the name of the AppEngine queue we should start
-        the pipeline on.
-
-    Returns:
-      True if we started the pipeline; False otherwise.
-    """
-    # Check policy and tune arguments if needed.
-    crash_data = self.CheckPolicy(crash_data)
-    if crash_data is None:
-      return False
-
-    # Detect the regression range, and decide if we actually need to
-    # run a new anlaysis or not.
-    if not self._NeedsNewAnalysis(crash_data):
-      return False
-
-    crash_identifiers = crash_data['crash_identifiers']
-    # N.B., we cannot pass |self| directly to the _pipeline_cls, because
-    # it is not JSON-serializable (and there's no way to make it such,
-    # since JSON-serializability is defined by JSON-encoders rather than
-    # as methods on the objects being encoded).
-    analysis_pipeline = self._pipeline_cls(self.client_id, crash_identifiers)
-    # Attribute defined outside __init__ - pylint: disable=W0201
-    analysis_pipeline.target = appengine_util.GetTargetNameForModule(
-        constants.CRASH_BACKEND[self.client_id])
-    analysis_pipeline.start(queue_name=queue_name)
-    logging.info('New %s analysis is scheduled for %s', self.client_id,
-                 repr(crash_identifiers))
-    return True
-
   # TODO(wrengr): does the parser actually need the version, signature,
   # and platform? If not, then we should be able to just pass the string
   # to be parsed (which would make a lot more sense than passing the
   # whole model).
   # TODO(http://crbug.com/659346): coverage tests for this class, not
   # just for FinditForFracas.
   def ParseStacktrace(self, model): # pragma: no cover
     """Parse a CrashAnalysis's |stack_trace| string into a Stacktrace object.
 
     Args:
@@ skipping 30 matching lines @@
       # TODO(http://crbug.com/659359): refactor things so we don't need
       # the NullCulprit class.
       return NullCulprit()
 
     return self._azalea.FindCulprit(CrashReport(
         crashed_version = model.crashed_version,
         signature = model.signature,
         platform = model.platform,
         stacktrace = stacktrace,
         regression_range = model.regression_range))