def _similarity_score(obj1, obj2, block_cache=None): """Compute a similarity score for two objects. Args: obj1: The first object to score. obj2: The second object to score. block_cache: An optional dict of SHA to block counts to cache results between calls. Returns: The similarity score between the two objects, defined as the number of bytes in common between the two objects divided by the maximum size, scaled to the range 0-100. """ if block_cache is None: block_cache = {} if obj1.id not in block_cache: block_cache[obj1.id] = _count_blocks(obj1) if obj2.id not in block_cache: block_cache[obj2.id] = _count_blocks(obj2) common_bytes = _common_bytes(block_cache[obj1.id], block_cache[obj2.id]) max_size = max(obj1.raw_length(), obj2.raw_length()) if not max_size: return _MAX_SCORE return int(float(common_bytes) * _MAX_SCORE / max_size)
def _find_content_rename_candidates(self): candidates = self._candidates = [] # TODO: Optimizations: # - Compare object sizes before counting blocks. # - Skip if delete's S_IFMT differs from all adds. # - Skip if adds or deletes is empty. # Match C git's behavior of not attempting to find content renames if # the matrix size exceeds the threshold. if not self._should_find_content_renames(): return block_cache = {} check_paths = self._rename_threshold is not None for delete in self._deletes: if S_ISGITLINK(delete.old.mode): continue # Git links don't exist in this repo. old_sha = delete.old.sha old_obj = self._store[old_sha] block_cache[old_sha] = _count_blocks(old_obj) for add in self._adds: if stat.S_IFMT(delete.old.mode) != stat.S_IFMT(add.new.mode): continue new_obj = self._store[add.new.sha] score = _similarity_score(old_obj, new_obj, block_cache=block_cache) if score > self._rename_threshold: new_type = self._rename_type(check_paths, delete, add) rename = TreeChange(new_type, delete.old, add.new) candidates.append((-score, rename))
def _find_content_renames(self): # TODO: Optimizations: # - Compare object sizes before counting blocks. # - Skip if delete's S_IFMT differs from all adds. # - Skip if adds or deletes is empty. # Match C git's behavior of not attempting to find content renames if # the matrix size exceeds the threshold. if len(self._adds) * len(self._deletes) > self._max_files ** 2: return check_paths = self._rename_threshold is not None candidates = [] for delete in self._deletes: if S_ISGITLINK(delete.old.mode): continue # Git links don't exist in this repo. old_sha = delete.old.sha old_obj = self._store[old_sha] old_blocks = _count_blocks(old_obj) for add in self._adds: if stat.S_IFMT(delete.old.mode) != stat.S_IFMT(add.new.mode): continue new_obj = self._store[add.new.sha] score = _similarity_score(old_obj, new_obj, block_cache={old_sha: old_blocks}) if score > self._rename_threshold: if check_paths and delete.old.path == add.new.path: # If the paths match, this must be a split modify, so # make sure it comes out as a modify. new_type = CHANGE_MODIFY elif delete.type != CHANGE_DELETE: # If it's in deletes but not marked as a delete, it must # have been added due to find_copies_harder, and needs # to be marked as a copy. new_type = CHANGE_COPY else: new_type = CHANGE_RENAME rename = TreeChange(new_type, delete.old, add.new) candidates.append((-score, rename)) # Sort scores from highest to lowest, but keep names in ascending order. candidates.sort() delete_paths = set() add_paths = set() for _, change in candidates: new_path = change.new.path if new_path in add_paths: continue old_path = change.old.path orig_type = change.type if old_path in delete_paths: change = TreeChange(CHANGE_COPY, change.old, change.new) # If the candidate was originally a copy, that means it came from a # modified or unchanged path, so we don't want to prune it. if orig_type != CHANGE_COPY: delete_paths.add(old_path) add_paths.add(new_path) self._changes.append(change) self._prune(add_paths, delete_paths)
def _similarity_score(obj1, obj2, block_cache=None): """Compute a similarity score for two objects. :param obj1: The first object to score. :param obj2: The second object to score. :param block_cache: An optional dict of SHA to block counts to cache results between calls. :return: The similarity score between the two objects, defined as the number of bytes in common between the two objects divided by the maximum size, scaled to the range 0-100. """ if block_cache is None: block_cache = {} if obj1.id not in block_cache: block_cache[obj1.id] = _count_blocks(obj1) if obj2.id not in block_cache: block_cache[obj2.id] = _count_blocks(obj2) common_bytes = _common_bytes(block_cache[obj1.id], block_cache[obj2.id]) max_size = max(obj1.raw_length(), obj2.raw_length()) if not max_size: return _MAX_SCORE return int(float(common_bytes) * _MAX_SCORE / max_size)
def _find_content_renames(self): # TODO: Optimizations: # - Compare object sizes before counting blocks. # - Skip if delete's S_IFMT differs from all adds. # - Skip if adds or deletes is empty. # Match C git's behavior of not attempting to find content renames if # the matrix size exceeds the threshold. if len(self._adds) * len(self._deletes) > self._max_files**2: return check_paths = self._rename_threshold is not None candidates = [] for delete in self._deletes: if S_ISGITLINK(delete.old.mode): continue # Git links don't exist in this repo. old_sha = delete.old.sha old_obj = self._store[old_sha] old_blocks = _count_blocks(old_obj) for add in self._adds: if stat.S_IFMT(delete.old.mode) != stat.S_IFMT(add.new.mode): continue new_obj = self._store[add.new.sha] score = _similarity_score(old_obj, new_obj, block_cache={old_sha: old_blocks}) if score > self._rename_threshold: if check_paths and delete.old.path == add.new.path: # If the paths match, this must be a split modify, so # make sure it comes out as a modify. new_type = CHANGE_MODIFY elif delete.type != CHANGE_DELETE: # If it's in deletes but not marked as a delete, it must # have been added due to find_copies_harder, and needs # to be marked as a copy. new_type = CHANGE_COPY else: new_type = CHANGE_RENAME rename = TreeChange(new_type, delete.old, add.new) candidates.append((-score, rename)) # Sort scores from highest to lowest, but keep names in ascending order. candidates.sort() delete_paths = set() add_paths = set() for _, change in candidates: new_path = change.new.path if new_path in add_paths: continue old_path = change.old.path orig_type = change.type if old_path in delete_paths: change = TreeChange(CHANGE_COPY, change.old, change.new) # If the candidate was originally a copy, that means it came from a # modified or unchanged path, so we don't want to prune it. if orig_type != CHANGE_COPY: delete_paths.add(old_path) add_paths.add(new_path) self._changes.append(change) self._prune(add_paths, delete_paths)