[Findit] Initial code of findit for crash. Add scorers to apply heuristic rules.

appengine/findit/crash/results.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.
+
+_INFINITY = 1000
+
+
+class Result(object):
+
+  def __init__(self, changelog, dep_path, component,
+               confidence=None, reason=None):
+    self.changelog = changelog
+    self.dep_path = dep_path
+    self.component = component
+    self.confidence = confidence
+    self.reason = reason
+
+    self.file_to_stack_infos = {}
+
+  def ToDict(self):
+    return {
+        'url': self.changelog.commit_url,
+        'revision': self.changelog.revision,
+        'dep_path': self.dep_path,
+        'component': self.component,
+        'author': self.changelog.author_email,
+        'time': self.changelog.author_time,
+        'reason': self.reason,
+        'confidence': self.confidence,
+    }
+
+  def ToString(self):
+    if not self.file_to_stack_infos:
+      return ''
+
+    lines = []
+    for file_path, stack_infos in self.file_to_stack_infos.iteritems():
+      line_parts = []
+      for frame, _ in stack_infos:
+        line_parts.append('%s (#%d)' % (frame.function, frame.index))
+
+      lines.append('Changed file %s which crashed in %s' % (
+          file_path, ', '.join(line_parts)))
+
+    return '\n'.join(lines)
+
+  def __str__(self):
+    return self.ToString()
+
+  def __repr__(self):
+    return self.ToString()

[Martin Barbella, 2016/04/18 21:01:18]

This doesn't seem like what would usually be expected from __repr__. Maybe
repr(self.ToDict()), but even that's a stretch. Any reason why this is needed at
all?

[Sharu Jiang, 2016/04/19 20:38:38]

This is more for a debug purpose, but yeah, this may not be needed. Delete it.

+
+
+class MatchResult(Result):
+
+  def __init__(self, changelog, dep_path, component,
+               confidence=None, reason=None):
+    super(MatchResult, self).__init__(
+        changelog, dep_path, component, confidence, reason)
+
+    self.min_distance = _INFINITY
+
+  def Update(self, file_path, stack_infos, blame):
+    """Updates match result once a file_path is found both shown in
+    stacktrace and touched by the revision of this result.
+
+    Inserts the file path and its stack infos, and updates the min distance
+    if less distance is found between touched lines of this result and
+    crashed lines in the file path.
+
+    Args:
+      file_path (str): File path of the crashed file.
+      stack_infos (list): List of (StackFrame, callstack priority) tuples,
+        represents frames of this file and the callstack priorities of those
+        frames.
+      blame (Blame): Blame oject of this file.
+    """
+    self.file_to_stack_infos[file_path] = stack_infos
+
+    for region in blame:
+      if region.revision != self.changelog.revision:
+        continue
+
+      region_lines = range(region.start, region.start + region.count)
+
+      for frame, _ in stack_infos:
+        self.min_distance = min(self.min_distance, self._DistanceOfTwoRegions(
+            frame.crashed_line_numbers, region_lines))
+
+  def _DistanceOfTwoRegions(self, region1, region2):
+
+    if set(region1).intersection(set(region2)):
+      return 0
+
+    if region1[-1] < region2[0]:

[Martin Barbella, 2016/04/18 21:01:18]

I suspect there's a more compact way to express this, but it's fairly readable
as-is. Either way is fine.

[Sharu Jiang, 2016/04/19 20:38:38]

Since it's too long to fit in one line, I just keep as it is.

+      return region2[0] - region1[-1]
+
+    return region1[0] - region2[-1]
+
+
+class MatchResults(dict):
+
+  def __init__(self, ignore_cls=None):
+    super(MatchResults, self).__init__()
+    self.ignore_cls = ignore_cls
+
+  def GenerateMatchResults(self, file_path, dep_path,
+                           stack_infos, changelogs, blame):
+    """Generates match results based on newly found file path, its stack_infos,
+    and all the changelogs that touched this file in the dep in regression
+    ranges, those reverted changelogs should be ignored.
+
+    Args:
+      file_path (str): File path of the crashed file.
+      dep_path (str): Path of the dependency of the file.
+      stack_infos (list): List of (StackFrame, callstack priority) tuples,
+        represents frames of this file and the callstack priorities of those
+        frames.
+      changelogs (list): List of Changelog objects in the dep in regression
+        range which touched the file.
+      blame (Blame): Blame of the file.
+    """
+
+    for changelog in changelogs:
+      if self.ignore_cls and changelog.revision in self.ignore_cls:
+        continue
+
+      if changelog.revision not in self:
+        # TODO(katesonia): Enable component classifier later. Get it from
+        # file_path and dep_path.
+        component = ''
+        self[changelog.revision] = MatchResult(changelog, dep_path, component)
+
+      match_result = self[changelog.revision]
+
+      match_result.Update(file_path, stack_infos, blame)