dualfroz/seperman-deepdiff
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
aa2b15b1b32ab5abcb418166f7046e7012b46416
seperman-deepdiff/deepdiff/diff.py
T

2413 lines
114 KiB
Python
Raw Blame History

#!/usr/bin/env python
# In order to run the docstrings:
# python3 -m deepdiff.diff
# You might need to run it many times since dictionaries come in different orders
# every time you run the docstrings.
# However the docstring expects it in a specific order in order to pass!
import difflib
import logging
import types
import datetime
import uuid
from enum import Enum
from copy import deepcopy
from math import isclose as is_close
from typing import List, Dict, Callable, Union, Any, Pattern, Tuple, Optional, Set, FrozenSet, TYPE_CHECKING, Protocol, Literal, cast
from collections.abc import Mapping, Iterable, Sequence
from collections import defaultdict
from inspect import getmembers
from itertools import zip_longest
from functools import lru_cache
from deepdiff.helper import (strings, bytes_type, numbers, uuids, ListItemRemovedOrAdded, notpresent, not_found,
IndexedHash, unprocessed, add_to_frozen_set, basic_types,
convert_item_or_items_into_set_else_none, get_type,
convert_item_or_items_into_compiled_regexes_else_none,
type_is_subclass_of_type_group, type_in_type_group, get_doc,
number_to_string, datetime_normalize, KEY_TO_VAL_STR, booleans,
np_ndarray, np_floating, get_numpy_ndarray_rows, RepeatedTimer,
TEXT_VIEW, TREE_VIEW, DELTA_VIEW, COLORED_VIEW, COLORED_COMPACT_VIEW,
detailed__dict__, add_root_to_paths,
np, get_truncate_datetime, dict_, CannotCompare, ENUM_INCLUDE_KEYS,
PydanticBaseModel, Opcode, SetOrdered, ipranges,
separate_wildcard_and_exact_paths)
from deepdiff.serialization import SerializationMixin
from deepdiff.distance import DistanceMixin, logarithmic_similarity
from deepdiff.model import (
RemapDict, ResultDict, TextResult, TreeResult, DiffLevel,
DictRelationship, AttributeRelationship, REPORT_KEYS,
SubscriptableIterableRelationship, NonSubscriptableIterableRelationship,
SetRelationship, NumpyArrayRelationship, CUSTOM_FIELD,
FORCE_DEFAULT, ChildRelationship,
)
from deepdiff.deephash import DeepHash, combine_hashes_lists
from deepdiff.base import Base
from deepdiff.lfucache import LFUCache, DummyLFU
from deepdiff.colored_view import ColoredView
from deepdiff._multiprocessing import (
normalize_mp_config, compute_distances_parallel,
compute_hashes_parallel, compute_subtree_diffs_parallel,
)
if TYPE_CHECKING:
from pytz.tzinfo import BaseTzInfo
logger = logging.getLogger(__name__)
MAX_PASSES_REACHED_MSG = (
'DeepDiff has reached the max number of passes of {}. '
'You can possibly get more accurate results by increasing the max_passes parameter.')
MAX_DIFFS_REACHED_MSG = (
'DeepDiff has reached the max number of diffs of {}. '
'You can possibly get more accurate results by increasing the max_diffs parameter.')
notpresent_indexed = IndexedHash(indexes=[0], item=notpresent)
doc = get_doc('diff_doc.rst')
PROGRESS_MSG = "DeepDiff {} seconds in progress. Pass #{}, Diff #{}"
def _report_progress(_stats: Dict[str, Any], progress_logger: Callable[[str], None], duration: float) -> None:
"""
Report the progress every few seconds.
"""
progress_logger(PROGRESS_MSG.format(duration, _stats[PASSES_COUNT], _stats[DIFF_COUNT]))
DISTANCE_CACHE_HIT_COUNT = 'DISTANCE CACHE HIT COUNT'
DIFF_COUNT = 'DIFF COUNT'
PASSES_COUNT = 'PASSES COUNT'
MAX_PASS_LIMIT_REACHED = 'MAX PASS LIMIT REACHED'
MAX_DIFF_LIMIT_REACHED = 'MAX DIFF LIMIT REACHED'
DISTANCE_CACHE_ENABLED = 'DISTANCE CACHE ENABLED'
PREVIOUS_DIFF_COUNT = 'PREVIOUS DIFF COUNT'
PREVIOUS_DISTANCE_CACHE_HIT_COUNT = 'PREVIOUS DISTANCE CACHE HIT COUNT'
WORKER_DIFF_COUNT = 'WORKER DIFF COUNT'
WORKER_PASSES_COUNT = 'WORKER PASSES COUNT'
WORKER_DISTANCE_CACHE_HIT_COUNT = 'WORKER DISTANCE CACHE HIT COUNT'
WORKER_BATCH_COUNT = 'WORKER BATCH COUNT'
CANT_FIND_NUMPY_MSG = 'Unable to import numpy. This must be a bug in DeepDiff since a numpy array is detected.'
INVALID_VIEW_MSG = "view parameter must be one of 'text', 'tree', 'delta', 'colored' or 'colored_compact'. But {} was passed."
CUTOFF_RANGE_ERROR_MSG = 'cutoff_distance_for_pairs needs to be a positive float max 1.'
VERBOSE_LEVEL_RANGE_MSG = 'verbose_level should be 0, 1, or 2.'
PURGE_LEVEL_RANGE_MSG = 'cache_purge_level should be 0, 1, or 2.'
_ENABLE_CACHE_EVERY_X_DIFF = '_ENABLE_CACHE_EVERY_X_DIFF'
model_fields_set = frozenset(["model_fields_set"])
# What is the threshold to consider 2 items to be pairs. Only used when ignore_order = True.
CUTOFF_DISTANCE_FOR_PAIRS_DEFAULT = 0.3
# What is the threshold to calculate pairs of items between 2 iterables.
# For example 2 iterables that have nothing in common, do not need their pairs to be calculated.
CUTOFF_INTERSECTION_FOR_PAIRS_DEFAULT = 0.7
DEEPHASH_PARAM_KEYS = (
'exclude_types',
'exclude_paths',
'exclude_glob_paths',
'include_paths',
'include_glob_paths',
'exclude_regex_paths',
'hasher',
'significant_digits',
'number_format_notation',
'ignore_string_type_changes',
'ignore_numeric_type_changes',
'ignore_uuid_types',
'use_enum_value',
'ignore_type_in_groups',
'ignore_type_subclasses',
'ignore_string_case',
'exclude_obj_callback',
'ignore_private_variables',
'encodings',
'ignore_encoding_errors',
'default_timezone',
'custom_operators',
)
class DeepDiffProtocol(Protocol):
t1: Any
t2: Any
cutoff_distance_for_pairs: float
use_log_scale: bool
log_scale_similarity_threshold: float
view: str
math_epsilon: Optional[float]
class DeepDiff(ResultDict, SerializationMixin, DistanceMixin, DeepDiffProtocol, Base):
__doc__ = doc
CACHE_AUTO_ADJUST_THRESHOLD = 0.25
def __init__(self,
t1: Any,
t2: Any,
_original_type: Optional[Any]=None,
cache_purge_level: int=1,
cache_size: int=0,
cache_tuning_sample_size: int=0,
custom_operators: Optional[List[Any]] =None,
cutoff_distance_for_pairs: float=CUTOFF_DISTANCE_FOR_PAIRS_DEFAULT,
cutoff_intersection_for_pairs: float=CUTOFF_INTERSECTION_FOR_PAIRS_DEFAULT,
default_timezone:Union[datetime.timezone, "BaseTzInfo"]=datetime.timezone.utc,
encodings: Optional[List[str]]=None,
exclude_obj_callback: Optional[Callable]=None,
exclude_obj_callback_strict: Optional[Callable]=None,
exclude_paths: Union[str, List[str], Set[str], FrozenSet[str], None]=None,
exclude_regex_paths: Union[str, List[str], Pattern[str], List[Pattern[str]], None]=None,
exclude_types: Optional[List[type]]=None,
get_deep_distance: bool=False,
group_by: Union[str, Tuple[str, str], Callable, None]=None,
group_by_sort_key: Union[str, Callable, None]=None,
hasher: Optional[Callable]=None,
hashes: Optional[Dict[Any, Any]]=None,
ignore_encoding_errors: bool=False,
ignore_nan_inequality: bool=False,
ignore_numeric_type_changes: bool=False,
ignore_order: bool=False,
ignore_order_func: Optional[Callable]=None,
ignore_private_variables: bool=True,
ignore_string_case: bool=False,
ignore_string_type_changes: bool=False,
ignore_type_in_groups: Optional[List[Tuple[Any, ...]]]=None,
ignore_type_subclasses: bool=False,
ignore_uuid_types: bool=False,
include_obj_callback: Optional[Callable]=None,
include_obj_callback_strict: Optional[Callable]=None,
include_paths: Union[str, List[str], None]=None,
iterable_compare_func: Optional[Callable]=None,
log_frequency_in_sec: int=0,
log_scale_similarity_threshold: float=0.1,
log_stacktrace: bool=False,
math_epsilon: Optional[float]=None,
max_diffs: Optional[int]=None,
max_passes: int=10000000,
multiprocessing: bool=False,
multiprocessing_workers: Optional[int]=None,
multiprocessing_threshold: Optional[int]=None,
number_format_notation: Literal["f", "e"]="f",
number_to_string_func: Optional[Callable]=None,
progress_logger: Callable[[str], None]=logger.info,
report_repetition: bool=False,
significant_digits: Optional[int]=None,
threshold_to_diff_deeper: float = 0.33,
truncate_datetime: Optional[str]=None,
use_enum_value: bool=False,
use_log_scale: bool=False,
verbose_level: int=1,
view: str=TEXT_VIEW,
zip_ordered_iterables: bool=False,
_parameters: Optional[Dict[str, Any]]=None,
_shared_parameters: Optional[Dict[str, Any]]=None,
**kwargs):
super().__init__()
# Defaults for glob path attributes — needed for non-root instances
# that may receive _parameters without these keys.
self.exclude_glob_paths = None
self.include_glob_paths = None
if kwargs:
raise ValueError((
"The following parameter(s) are not valid: %s\n"
"The valid parameters are ignore_order, report_repetition, significant_digits, "
"number_format_notation, exclude_paths, include_paths, exclude_types, exclude_regex_paths, ignore_type_in_groups, "
"ignore_string_type_changes, ignore_numeric_type_changes, ignore_type_subclasses, ignore_uuid_types, truncate_datetime, "
"ignore_private_variables, ignore_nan_inequality, number_to_string_func, verbose_level, "
"view, hasher, hashes, max_passes, max_diffs, zip_ordered_iterables, "
"cutoff_distance_for_pairs, cutoff_intersection_for_pairs, log_frequency_in_sec, cache_size, "
"cache_tuning_sample_size, get_deep_distance, group_by, group_by_sort_key, cache_purge_level, log_stacktrace,"
"math_epsilon, iterable_compare_func, use_enum_value, _original_type, threshold_to_diff_deeper, default_timezone "
"ignore_order_func, custom_operators, encodings, ignore_encoding_errors, use_log_scale, log_scale_similarity_threshold "
"multiprocessing, multiprocessing_workers, multiprocessing_threshold, "
"_parameters and _shared_parameters.") % ', '.join(kwargs.keys()))
if _parameters:
self.__dict__.update(_parameters)
else:
self.custom_operators = custom_operators or []
self.ignore_order = ignore_order
self.ignore_order_func = ignore_order_func
ignore_type_in_groups = ignore_type_in_groups or []
if numbers == ignore_type_in_groups or numbers in ignore_type_in_groups:
ignore_numeric_type_changes = True
self.ignore_numeric_type_changes = ignore_numeric_type_changes
if strings == ignore_type_in_groups or strings in ignore_type_in_groups:
ignore_string_type_changes = True
# Handle ignore_uuid_types - check if uuid+str group is already in ignore_type_in_groups
uuid_str_group = (uuids[0], str)
if uuid_str_group == ignore_type_in_groups or uuid_str_group in ignore_type_in_groups:
ignore_uuid_types = True
self.ignore_uuid_types = ignore_uuid_types
self.use_enum_value = use_enum_value
self.log_scale_similarity_threshold = log_scale_similarity_threshold
self.use_log_scale = use_log_scale
self.default_timezone = default_timezone
self.log_stacktrace = log_stacktrace
self.threshold_to_diff_deeper = threshold_to_diff_deeper
self.ignore_string_type_changes = ignore_string_type_changes
self.ignore_type_in_groups = self.get_ignore_types_in_groups(
ignore_type_in_groups=ignore_type_in_groups,
ignore_string_type_changes=ignore_string_type_changes,
ignore_numeric_type_changes=ignore_numeric_type_changes,
ignore_type_subclasses=ignore_type_subclasses,
ignore_uuid_types=ignore_uuid_types)
self.report_repetition = report_repetition
_exclude_set = convert_item_or_items_into_set_else_none(exclude_paths)
_exclude_exact, self.exclude_glob_paths = separate_wildcard_and_exact_paths(_exclude_set)
self.exclude_paths = add_root_to_paths(_exclude_exact)
_include_set = convert_item_or_items_into_set_else_none(include_paths)
_include_exact, self.include_glob_paths = separate_wildcard_and_exact_paths(_include_set)
self.include_paths = add_root_to_paths(_include_exact)
self.exclude_regex_paths = convert_item_or_items_into_compiled_regexes_else_none(exclude_regex_paths)
self.exclude_types = set(exclude_types) if exclude_types else None
self.exclude_types_tuple = tuple(exclude_types) if exclude_types else None # we need tuple for checking isinstance
self.ignore_type_subclasses = ignore_type_subclasses
self.type_check_func = type_in_type_group if ignore_type_subclasses else type_is_subclass_of_type_group
self.ignore_string_case = ignore_string_case
self.exclude_obj_callback = exclude_obj_callback
self.exclude_obj_callback_strict = exclude_obj_callback_strict
self.include_obj_callback = include_obj_callback
self.include_obj_callback_strict = include_obj_callback_strict
self.number_to_string = number_to_string_func or number_to_string
self.iterable_compare_func = iterable_compare_func
self.zip_ordered_iterables = zip_ordered_iterables
self.ignore_private_variables = ignore_private_variables
self.ignore_nan_inequality = ignore_nan_inequality
self.hasher = hasher
self.cache_tuning_sample_size = cache_tuning_sample_size
self.group_by = group_by
if callable(group_by_sort_key):
self.group_by_sort_key = group_by_sort_key
elif group_by_sort_key:
def _group_by_sort_key(x):
return x[group_by_sort_key]
self.group_by_sort_key = _group_by_sort_key
else:
self.group_by_sort_key = None
self.encodings = encodings
self.ignore_encoding_errors = ignore_encoding_errors
self.significant_digits = self.get_significant_digits(significant_digits, ignore_numeric_type_changes)
self.math_epsilon = math_epsilon
if self.math_epsilon is not None and self.ignore_order:
logger.warning("math_epsilon in conjunction with ignore_order=True is only used for flat object comparisons. Custom math_epsilon will not have an effect when comparing nested objects.")
self.truncate_datetime = get_truncate_datetime(truncate_datetime)
self.number_format_notation = number_format_notation
if verbose_level in {0, 1, 2}:
self.verbose_level = verbose_level
else:
raise ValueError(VERBOSE_LEVEL_RANGE_MSG)
if cache_purge_level not in {0, 1, 2}:
raise ValueError(PURGE_LEVEL_RANGE_MSG)
self.view = view
# Setting up the cache for dynamic programming. One dictionary per instance of root of DeepDiff running.
self.max_passes = max_passes
self.max_diffs = max_diffs
self.cutoff_distance_for_pairs = float(cutoff_distance_for_pairs)
self.cutoff_intersection_for_pairs = float(cutoff_intersection_for_pairs)
if self.cutoff_distance_for_pairs < 0 or self.cutoff_distance_for_pairs > 1:
raise ValueError(CUTOFF_RANGE_ERROR_MSG)
# _Parameters are the clean _parameters to initialize DeepDiff with so we avoid all the above
# cleaning functionalities when running DeepDiff recursively.
# However DeepHash has its own set of _parameters that are slightly different than DeepDIff.
# DeepDiff _parameters are transformed to DeepHash _parameters via _get_deephash_params method.
self.progress_logger = progress_logger
self.cache_size = cache_size
self._mp_config = normalize_mp_config(
multiprocessing, multiprocessing_workers, multiprocessing_threshold)
_parameters = self.__dict__.copy()
_parameters['group_by'] = None # overwriting since these parameters will be passed on to other passes.
if log_stacktrace:
self.log_err = logger.exception
else:
self.log_err = logger.error
# Non-Root
if _shared_parameters:
self.is_root = False
self._shared_parameters = _shared_parameters
self.__dict__.update(_shared_parameters)
# We are in some pass other than root
progress_timer = None
# Root
else:
self.is_root = True
# Caching the DeepDiff results for dynamic programming
self._distance_cache = LFUCache(cache_size) if cache_size else DummyLFU()
self._stats = {
PASSES_COUNT: 0,
DIFF_COUNT: 0,
DISTANCE_CACHE_HIT_COUNT: 0,
PREVIOUS_DIFF_COUNT: 0,
PREVIOUS_DISTANCE_CACHE_HIT_COUNT: 0,
MAX_PASS_LIMIT_REACHED: False,
MAX_DIFF_LIMIT_REACHED: False,
DISTANCE_CACHE_ENABLED: bool(cache_size),
# Multiprocessing aggregates: each parallel batch sums per-worker
# _stats deltas into these keys. Parent-side counters above stay
# comparable to a serial run so existing tests are unaffected.
WORKER_DIFF_COUNT: 0,
WORKER_PASSES_COUNT: 0,
WORKER_DISTANCE_CACHE_HIT_COUNT: 0,
WORKER_BATCH_COUNT: 0,
}
self.hashes = dict_() if hashes is None else hashes
self._numpy_paths = dict_() # if _numpy_paths is None else _numpy_paths
self.group_by_keys = set() # Track keys that originated from group_by operations
self._shared_parameters = {
'hashes': self.hashes,
'_stats': self._stats,
'_distance_cache': self._distance_cache,
'group_by_keys': self.group_by_keys,
'_numpy_paths': self._numpy_paths,
_ENABLE_CACHE_EVERY_X_DIFF: self.cache_tuning_sample_size * 10,
}
if log_frequency_in_sec:
# Creating a progress log reporter that runs in a separate thread every log_frequency_in_sec seconds.
progress_timer = RepeatedTimer(log_frequency_in_sec, _report_progress, self._stats, progress_logger)
else:
progress_timer = None
self._parameters = _parameters
self.deephash_parameters = self._get_deephash_params()
self.tree = TreeResult()
self._iterable_opcodes = {}
if group_by and self.is_root:
try:
original_t1 = t1
t1 = self._group_iterable_to_dict(t1, group_by, item_name='t1')
except (KeyError, ValueError):
pass
else:
try:
t2 = self._group_iterable_to_dict(t2, group_by, item_name='t2')
except (KeyError, ValueError):
t1 = original_t1
self.t1 = t1
self.t2 = t2
try:
root = DiffLevel(t1, t2, verbose_level=self.verbose_level)
# _original_type is only used to pass the original type of the data. Currently only used for numpy arrays.
# The reason is that we convert the numpy array to python list and then later for distance calculations
# we convert only the the last dimension of it into numpy arrays.
self._diff(root, parents_ids=frozenset({id(t1)}), _original_type=_original_type)
if get_deep_distance and view in {TEXT_VIEW, TREE_VIEW}:
self.tree['deep_distance'] = self._get_rough_distance()
self.tree.remove_empty_keys()
view_results = self._get_view_results(self.view)
if isinstance(view_results, ColoredView):
self.update(view_results.tree)
self._colored_view = view_results
else:
self.update(view_results)
finally:
if self.is_root:
if cache_purge_level:
del self._distance_cache
del self.hashes
del self._shared_parameters
del self._parameters
for key in (PREVIOUS_DIFF_COUNT, PREVIOUS_DISTANCE_CACHE_HIT_COUNT,
DISTANCE_CACHE_ENABLED):
del self._stats[key]
if progress_timer:
duration = progress_timer.stop()
self._stats['DURATION SEC'] = duration
logger.info('stats {}'.format(self.get_stats()))
if cache_purge_level == 2:
self.__dict__.clear()
def _get_deephash_params(self):
result = {key: self._parameters.get(key) for key in DEEPHASH_PARAM_KEYS}
result['ignore_repetition'] = not self.report_repetition
result['number_to_string_func'] = self.number_to_string
return result
def _report_result(self, report_type, change_level, local_tree=None):
"""
Add a detected change to the reference-style result dictionary.
report_type will be added to level.
(We'll create the text-style report from there later.)
:param report_type: A well defined string key describing the type of change.
Examples: "set_item_added", "values_changed"
:param change_level: A DiffLevel object describing the objects in question in their
before-change and after-change object structure.
:local_tree: None
"""
if not self._skip_this(change_level):
if self._skip_report_for_include_glob(change_level):
return
change_level.report_type = report_type
tree = self.tree if local_tree is None else local_tree
tree[report_type].add(change_level)
def custom_report_result(self, report_type, level, extra_info=None):
"""
Add a detected change to the reference-style result dictionary.
report_type will be added to level.
(We'll create the text-style report from there later.)
:param report_type: A well defined string key describing the type of change.
Examples: "set_item_added", "values_changed"
:param parent: A DiffLevel object describing the objects in question in their
before-change and after-change object structure.
:param extra_info: A dict that describe this result
:rtype: None
"""
if not self._skip_this(level):
if self._skip_report_for_include_glob(level):
return
level.report_type = report_type
level.additional[CUSTOM_FIELD] = extra_info
self.tree[report_type].add(level)
def _skip_report_for_include_glob(self, level):
"""When include_glob_paths is set, _skip_this allows ancestors through for traversal.
This method does a stricter check at report time: only report if the path
actually matches a glob pattern or is a descendant of a matching path,
or if it already matches an exact include_path."""
if not self.include_glob_paths:
return False
level_path = level.path()
# If exact include_paths already matched, don't skip
if self.include_paths:
if level_path in self.include_paths:
return False
for prefix in self.include_paths:
if prefix in level_path:
return False
# Check glob patterns: match or descendant
for gp in self.include_glob_paths:
if gp.match_or_is_descendant(level_path):
return False
return True
@staticmethod
def _dict_from_slots(object: Any) -> Dict[str, Any]:
def unmangle(attribute: str) -> str:
if attribute.startswith('__') and attribute != '__weakref__':
return '_{type}{attribute}'.format(
type=type(object).__name__,
attribute=attribute
)
return attribute
all_slots = []
if isinstance(object, type):
mro = object.__mro__ # pragma: no cover. I have not been able to write a test for this case. But we still check for it.
else:
mro = object.__class__.__mro__
for type_in_mro in mro:
slots = getattr(type_in_mro, '__slots__', None)
if slots:
if isinstance(slots, strings):
all_slots.append(slots)
else:
all_slots.extend(slots)
return {i: getattr(object, key) for i in all_slots if hasattr(object, key := unmangle(i))}
def _diff_enum(self, level: Any, parents_ids: FrozenSet[int]=frozenset(), local_tree: Optional[Any]=None) -> None:
t1 = detailed__dict__(level.t1, include_keys=ENUM_INCLUDE_KEYS)
t2 = detailed__dict__(level.t2, include_keys=ENUM_INCLUDE_KEYS)
self._diff_dict(
level,
parents_ids,
print_as_attribute=True,
override=True,
override_t1=t1,
override_t2=t2,
local_tree=local_tree,
)
def _diff_obj(self, level: Any, parents_ids: FrozenSet[int]=frozenset(), is_namedtuple: bool=False, local_tree: Optional[Any]=None, is_pydantic_object: bool=False) -> None:
"""Difference of 2 objects"""
processing_error = False
t1: Optional[Dict[str, Any]] = None
t2: Optional[Dict[str, Any]] = None
try:
if is_namedtuple:
t1 = level.t1._asdict()
t2 = level.t2._asdict()
elif is_pydantic_object:
t1 = detailed__dict__(level.t1, ignore_private_variables=self.ignore_private_variables, ignore_keys=model_fields_set)
t2 = detailed__dict__(level.t2, ignore_private_variables=self.ignore_private_variables, ignore_keys=model_fields_set)
elif all('__dict__' in dir(t) for t in level):
t1 = detailed__dict__(level.t1, ignore_private_variables=self.ignore_private_variables)
t2 = detailed__dict__(level.t2, ignore_private_variables=self.ignore_private_variables)
elif all('__slots__' in dir(t) for t in level):
t1 = self._dict_from_slots(level.t1)
t2 = self._dict_from_slots(level.t2)
else:
t1 = {k: v for k, v in getmembers(level.t1) if not callable(v)}
t2 = {k: v for k, v in getmembers(level.t2) if not callable(v)}
except AttributeError:
processing_error = True
if processing_error is True or t1 is None or t2 is None:
self._report_result('unprocessed', level, local_tree=local_tree)
return
self._diff_dict(
level,
parents_ids,
print_as_attribute=True,
override=True,
override_t1=t1,
override_t2=t2,
local_tree=local_tree,
)
def _skip_this(self, level: Any) -> bool:
"""
Check whether this comparison should be skipped because one of the objects to compare meets exclusion criteria.
:rtype: bool
"""
level_path = level.path()
skip = False
if self.exclude_paths and level_path in self.exclude_paths:
skip = True
elif self.exclude_glob_paths and any(gp.match(level_path) for gp in self.exclude_glob_paths):
skip = True
if not skip and (self.include_paths or self.include_glob_paths) and level_path != 'root':
skip = True
if self.include_paths:
if level_path in self.include_paths:
skip = False
else:
for prefix in self.include_paths:
if prefix in level_path or level_path in prefix:
skip = False
break
if skip and self.include_glob_paths:
for gp in self.include_glob_paths:
if gp.match_or_is_ancestor(level_path):
skip = False
break
elif self.exclude_regex_paths and any(
[exclude_regex_path.search(level_path) for exclude_regex_path in self.exclude_regex_paths]):
skip = True
elif self.exclude_types_tuple and \
(isinstance(level.t1, self.exclude_types_tuple) or isinstance(level.t2, self.exclude_types_tuple)):
skip = True
elif self.exclude_obj_callback and \
(self.exclude_obj_callback(level.t1, level_path) or self.exclude_obj_callback(level.t2, level_path)):
skip = True
elif self.exclude_obj_callback_strict and \
(self.exclude_obj_callback_strict(level.t1, level_path) and
self.exclude_obj_callback_strict(level.t2, level_path)):
skip = True
elif self.include_obj_callback and level_path != 'root':
skip = True
if (self.include_obj_callback(level.t1, level_path) or self.include_obj_callback(level.t2, level_path)):
skip = False
elif self.include_obj_callback_strict and level_path != 'root':
skip = True
if (self.include_obj_callback_strict(level.t1, level_path) and
self.include_obj_callback_strict(level.t2, level_path)):
skip = False
return skip
def _skip_this_key(self, level: Any, key: Any) -> bool:
# if include_paths is not set, than treet every path as included
if self.include_paths is None and self.include_glob_paths is None:
return False
key_path = "{}['{}']".format(level.path(), key)
if self.include_paths:
if key_path in self.include_paths:
return False
if level.path() in self.include_paths:
# matches e.g. level+key root['foo']['bar']['veg'] include_paths ["root['foo']['bar']"]
return False
for prefix in self.include_paths:
if key_path in prefix:
# matches as long the prefix is longer than this object key
# eg.: level+key root['foo']['bar'] matches prefix root['foo']['bar'] from include paths
# level+key root['foo'] matches prefix root['foo']['bar'] from include_paths
# level+key root['foo']['bar'] DOES NOT match root['foo'] from include_paths This needs to be handled afterwards
return False
# check if a higher level is included as a whole (=without any sublevels specified)
# matches e.g. level+key root['foo']['bar']['veg'] include_paths ["root['foo']"]
# but does not match, if it is level+key root['foo']['bar']['veg'] include_paths ["root['foo']['bar']['fruits']"]
up = level.up
while up is not None:
if up.path() in self.include_paths:
return False
up = up.up
if self.include_glob_paths:
for gp in self.include_glob_paths:
if gp.match_or_is_ancestor(key_path):
return False
return True
def _get_clean_to_keys_mapping(self, keys: Any, level: Any) -> Dict[Any, Any]:
"""
Get a dictionary of cleaned value of keys to the keys themselves.
This is mainly used to transform the keys when the type changes of keys should be ignored.
TODO: needs also some key conversion for groups of types other than the built-in strings and numbers.
"""
result = dict_()
for key in keys:
if self.ignore_string_type_changes and isinstance(key, bytes):
clean_key = key.decode('utf-8')
elif self.ignore_string_type_changes and isinstance(key, memoryview):
clean_key = key.tobytes().decode('utf-8')
elif self.use_enum_value and isinstance(key, Enum):
clean_key = key.value
elif isinstance(key, numbers):
# Skip type prefixing for keys that originated from group_by operations
if hasattr(self, 'group_by_keys') and key in self.group_by_keys:
if self.significant_digits is None:
clean_key = key
else:
clean_key = self.number_to_string(key, significant_digits=self.significant_digits,
number_format_notation=self.number_format_notation) # type: ignore # type: ignore
else:
type_ = "number" if self.ignore_numeric_type_changes else key.__class__.__name__
if self.significant_digits is None:
clean_key = key
else:
clean_key = self.number_to_string(key, significant_digits=self.significant_digits,
number_format_notation=self.number_format_notation) # type: ignore # type: ignore
clean_key = KEY_TO_VAL_STR.format(type_, clean_key)
else:
clean_key = key
if self.ignore_string_case and isinstance(clean_key, str):
clean_key = clean_key.lower()
if clean_key in result:
logger.warning(('{} and {} in {} become the same key when ignore_numeric_type_changes'
'or ignore_numeric_type_changes are set to be true.').format(
key, result[clean_key], level.path()))
else:
result[clean_key] = key
return result
def _diff_dict(
self,
level: Any,
parents_ids: FrozenSet[int]=frozenset([]),
print_as_attribute: bool=False,
override: bool=False,
override_t1: Any=None,
override_t2: Any=None,
local_tree: Optional[Any]=None,
) -> None:
"""Difference of 2 dictionaries"""
if override:
# for special stuff like custom objects and named tuples we receive preprocessed t1 and t2
# but must not spoil the chain (=level) with it
t1 = override_t1
t2 = override_t2
else:
t1 = level.t1
t2 = level.t2
if print_as_attribute:
item_added_key = "attribute_added"
item_removed_key = "attribute_removed"
rel_class = AttributeRelationship
else:
item_added_key = "dictionary_item_added"
item_removed_key = "dictionary_item_removed"
rel_class = DictRelationship
if self.ignore_private_variables:
t1_keys = SetOrdered([key for key in t1 if not(isinstance(key, str) and key.startswith('__')) and not self._skip_this_key(level, key)])
t2_keys = SetOrdered([key for key in t2 if not(isinstance(key, str) and key.startswith('__')) and not self._skip_this_key(level, key)])
else:
t1_keys = SetOrdered([key for key in t1 if not self._skip_this_key(level, key)])
t2_keys = SetOrdered([key for key in t2 if not self._skip_this_key(level, key)])
if self.ignore_string_type_changes or self.ignore_numeric_type_changes or self.ignore_string_case:
t1_clean_to_keys = self._get_clean_to_keys_mapping(keys=t1_keys, level=level)
t2_clean_to_keys = self._get_clean_to_keys_mapping(keys=t2_keys, level=level)
t1_keys = SetOrdered(t1_clean_to_keys.keys())
t2_keys = SetOrdered(t2_clean_to_keys.keys())
else:
t1_clean_to_keys = t2_clean_to_keys = None
t_keys_intersect = t2_keys & t1_keys
t_keys_added = t2_keys - t_keys_intersect
t_keys_removed = t1_keys - t_keys_intersect
if self.threshold_to_diff_deeper:
if self.exclude_paths or self.exclude_glob_paths:
t_keys_union = {f"{level.path()}[{repr(key)}]" for key in (t2_keys | t1_keys)}
if self.exclude_paths:
t_keys_union -= self.exclude_paths
if self.exclude_glob_paths:
t_keys_union = {k for k in t_keys_union
if not any(gp.match(k) for gp in self.exclude_glob_paths)}
t_keys_union_len = len(t_keys_union)
else:
t_keys_union_len = len(t2_keys | t1_keys)
if t_keys_union_len > 1 and len(t_keys_intersect) / t_keys_union_len < self.threshold_to_diff_deeper:
self._report_result('values_changed', level, local_tree=local_tree)
return
for key in t_keys_added:
if self._count_diff() is StopIteration:
return
key = t2_clean_to_keys[key] if t2_clean_to_keys else key
change_level = level.branch_deeper(
notpresent,
t2[key],
child_relationship_class=rel_class,
child_relationship_param=key,
child_relationship_param2=key,
)
self._report_result(item_added_key, change_level, local_tree=local_tree)
for key in t_keys_removed:
if self._count_diff() is StopIteration:
return # pragma: no cover. This is already covered for addition.
key = t1_clean_to_keys[key] if t1_clean_to_keys else key
change_level = level.branch_deeper(
t1[key],
notpresent,
child_relationship_class=rel_class,
child_relationship_param=key,
child_relationship_param2=key,
)
self._report_result(item_removed_key, change_level, local_tree=local_tree)
for key in t_keys_intersect: # key present in both dicts - need to compare values
if self._count_diff() is StopIteration:
return # pragma: no cover. This is already covered for addition.
key1 = t1_clean_to_keys[key] if t1_clean_to_keys else key
key2 = t2_clean_to_keys[key] if t2_clean_to_keys else key
item_id = id(t1[key1])
if parents_ids and item_id in parents_ids:
continue
parents_ids_added = add_to_frozen_set(parents_ids, item_id)
# Go one level deeper
next_level = level.branch_deeper(
t1[key1],
t2[key2],
child_relationship_class=rel_class,
child_relationship_param=key,
child_relationship_param2=key,
)
self._diff(next_level, parents_ids_added, local_tree=local_tree)
def _diff_set(self, level: Any, local_tree: Optional[Any]=None) -> None:
"""Difference of sets"""
t1_hashtable = self._create_hashtable(level, 't1')
t2_hashtable = self._create_hashtable(level, 't2')
t1_hashes = set(t1_hashtable.keys())
t2_hashes = set(t2_hashtable.keys())
hashes_added = t2_hashes - t1_hashes
hashes_removed = t1_hashes - t2_hashes
items_added = [t2_hashtable[i].item for i in hashes_added]
items_removed = [t1_hashtable[i].item for i in hashes_removed]
for item in items_added:
if self._count_diff() is StopIteration:
return # pragma: no cover. This is already covered for addition.
change_level = level.branch_deeper(
notpresent, item, child_relationship_class=SetRelationship)
self._report_result('set_item_added', change_level, local_tree=local_tree)
for item in items_removed:
if self._count_diff() is StopIteration:
return # pragma: no cover. This is already covered for addition.
change_level = level.branch_deeper(
item, notpresent, child_relationship_class=SetRelationship)
self._report_result('set_item_removed', change_level, local_tree=local_tree)
@staticmethod
def _iterables_subscriptable(t1: Any, t2: Any) -> bool:
try:
if getattr(t1, '__getitem__') and getattr(t2, '__getitem__'):
return True
else: # pragma: no cover
return False # should never happen
except AttributeError:
return False
def _diff_iterable(self, level: Any, parents_ids: FrozenSet[int]=frozenset(), _original_type: Optional[type]=None, local_tree: Optional[Any]=None) -> None:
"""Difference of iterables"""
if (self.ignore_order_func and self.ignore_order_func(level)) or self.ignore_order:
self._diff_iterable_with_deephash(level, parents_ids, _original_type=_original_type, local_tree=local_tree)
else:
self._diff_iterable_in_order(level, parents_ids, _original_type=_original_type, local_tree=local_tree)
def _compare_in_order(
self, level,
t1_from_index: Optional[int]=None, t1_to_index: Optional[int]=None,
t2_from_index: Optional[int]=None, t2_to_index: Optional[int]=None
) -> List[Tuple[Tuple[int, int], Tuple[Any, Any]]]:
"""
Default compare if `iterable_compare_func` is not provided.
This will compare in sequence order.
"""
if t1_from_index is None or t2_from_index is None:
return [((i, i), (x, y)) for i, (x, y) in enumerate(
zip_longest(
level.t1, level.t2, fillvalue=ListItemRemovedOrAdded))]
else:
t1_chunk = level.t1[t1_from_index:t1_to_index]
t2_chunk = level.t2[t2_from_index:t2_to_index]
return [((i + t1_from_index, i + t2_from_index), (x, y)) for i, (x, y) in enumerate(
zip_longest(
t1_chunk, t2_chunk, fillvalue=ListItemRemovedOrAdded))]
def _get_matching_pairs(
self, level,
t1_from_index=None, t1_to_index=None,
t2_from_index=None, t2_to_index=None
) -> List[Tuple[Tuple[int, int], Tuple[Any, Any]]]:
"""
Given a level get matching pairs. This returns list of two tuples in the form:
[
(t1 index, t2 index), (t1 item, t2 item)
]
This will compare using the passed in `iterable_compare_func` if available.
Default it to compare in order
"""
if self.iterable_compare_func is None:
# Match in order if there is no compare function provided
return self._compare_in_order(
level,
t1_from_index=t1_from_index, t1_to_index=t1_to_index,
t2_from_index=t2_from_index, t2_to_index=t2_to_index,
)
try:
matches = []
y_matched = set()
y_index_matched = set()
for i, x in enumerate(level.t1):
x_found = False
for j, y in enumerate(level.t2):
if(j in y_index_matched):
# This ensures a one-to-one relationship of matches from t1 to t2.
# If y this index in t2 has already been matched to another x
# it cannot have another match, so just continue.
continue
if(self.iterable_compare_func(x, y, level)):
deep_hash = DeepHash(y,
hashes=self.hashes,
apply_hash=True,
**self.deephash_parameters,
)
y_index_matched.add(j)
y_matched.add(deep_hash[y])
matches.append(((i, j), (x, y)))
x_found = True
break
if(not x_found):
matches.append(((i, -1), (x, ListItemRemovedOrAdded)))
for j, y in enumerate(level.t2):
deep_hash = DeepHash(y,
hashes=self.hashes,
apply_hash=True,
**self.deephash_parameters,
)
if(deep_hash[y] not in y_matched):
matches.append(((-1, j), (ListItemRemovedOrAdded, y)))
return matches
except CannotCompare:
return self._compare_in_order(
level,
t1_from_index=t1_from_index, t1_to_index=t1_to_index,
t2_from_index=t2_from_index, t2_to_index=t2_to_index
)
def _diff_iterable_in_order(self, level, parents_ids=frozenset(), _original_type=None, local_tree=None):
# We're handling both subscriptable and non-subscriptable iterables. Which one is it?
subscriptable = self._iterables_subscriptable(level.t1, level.t2)
if subscriptable:
child_relationship_class = SubscriptableIterableRelationship
else:
child_relationship_class = NonSubscriptableIterableRelationship
if (
not self.zip_ordered_iterables
and isinstance(level.t1, Sequence)
and isinstance(level.t2, Sequence)
and self._all_values_basic_hashable(level.t1)
and self._all_values_basic_hashable(level.t2)
and self.iterable_compare_func is None
):
local_tree_pass = TreeResult()
opcodes_with_values = self._diff_ordered_iterable_by_difflib(
level,
parents_ids=parents_ids,
_original_type=_original_type,
child_relationship_class=child_relationship_class,
local_tree=local_tree_pass,
)
# Sometimes DeepDiff's old iterable diff does a better job than DeepDiff
if len(local_tree_pass) > 1:
local_tree_pass2 = TreeResult()
self._diff_by_forming_pairs_and_comparing_one_by_one(
level,
parents_ids=parents_ids,
_original_type=_original_type,
child_relationship_class=child_relationship_class,
local_tree=local_tree_pass2,
)
if len(local_tree_pass) >= len(local_tree_pass2):
local_tree_pass = local_tree_pass2
else:
self._iterable_opcodes[level.path(force=FORCE_DEFAULT)] = opcodes_with_values
for report_type, levels in local_tree_pass.items():
if levels:
self.tree[report_type] |= levels
else:
self._diff_by_forming_pairs_and_comparing_one_by_one(
level,
parents_ids=parents_ids,
_original_type=_original_type,
child_relationship_class=child_relationship_class,
local_tree=local_tree,
)
def _all_values_basic_hashable(self, iterable: Iterable[Any]) -> bool:
"""
Are all items basic hashable types?
Or there are custom types too?
"""
# We don't want to exhaust a generator
if isinstance(iterable, types.GeneratorType):
return False
for item in iterable:
if not isinstance(item, basic_types):
return False
return True
def _diff_by_forming_pairs_and_comparing_one_by_one(
self, level, local_tree, parents_ids=frozenset(),
_original_type=None, child_relationship_class=None,
t1_from_index=None, t1_to_index=None,
t2_from_index=None, t2_to_index=None,
):
for (i, j), (x, y) in self._get_matching_pairs(
level,
t1_from_index=t1_from_index, t1_to_index=t1_to_index,
t2_from_index=t2_from_index, t2_to_index=t2_to_index
):
if self._count_diff() is StopIteration:
return # pragma: no cover. This is already covered for addition.
reference_param1 = i
reference_param2 = j
if y is ListItemRemovedOrAdded: # item removed completely
change_level = level.branch_deeper(
x,
notpresent,
child_relationship_class=child_relationship_class,
child_relationship_param=reference_param1,
child_relationship_param2=reference_param2,
)
self._report_result('iterable_item_removed', change_level, local_tree=local_tree)
elif x is ListItemRemovedOrAdded: # new item added
change_level = level.branch_deeper(
notpresent,
y,
child_relationship_class=child_relationship_class,
child_relationship_param=reference_param1,
child_relationship_param2=reference_param2,
)
self._report_result('iterable_item_added', change_level, local_tree=local_tree)
else: # check if item value has changed
if (i != j and ((x == y) or self.iterable_compare_func)):
# Item moved
change_level = level.branch_deeper(
x,
y,
child_relationship_class=child_relationship_class,
child_relationship_param=reference_param1,
child_relationship_param2=reference_param2
)
self._report_result('iterable_item_moved', change_level, local_tree=local_tree)
if self.iterable_compare_func:
# Mark additional context denoting that we have moved an item.
# This will allow for correctly setting paths relative to t2 when using an iterable_compare_func
level.additional["moved"] = True
else:
continue
item_id = id(x)
if parents_ids and item_id in parents_ids:
continue
parents_ids_added = add_to_frozen_set(parents_ids, item_id)
# Go one level deeper
next_level = level.branch_deeper(
x,
y,
child_relationship_class=child_relationship_class,
child_relationship_param=reference_param1,
child_relationship_param2=reference_param2
)
self._diff(next_level, parents_ids_added, local_tree=local_tree)
def _diff_ordered_iterable_by_difflib(
self, level, local_tree, parents_ids=frozenset(), _original_type=None, child_relationship_class=None,
):
seq = difflib.SequenceMatcher(isjunk=None, a=level.t1, b=level.t2, autojunk=False)
opcodes = seq.get_opcodes()
opcodes_with_values = []
# TODO: this logic should be revisted so we detect reverse operations
# like when a replacement happens at index X and a reverse replacement happens at index Y
# in those cases we have a "iterable_item_moved" operation.
for tag, t1_from_index, t1_to_index, t2_from_index, t2_to_index in opcodes:
if tag == 'equal':
opcodes_with_values.append(Opcode(
tag, t1_from_index, t1_to_index, t2_from_index, t2_to_index,
))
continue
# print('{:7} t1[{}:{}] --> t2[{}:{}] {!r:>8} --> {!r}'.format(
# tag, t1_from_index, t1_to_index, t2_from_index, t2_to_index, level.t1[t1_from_index:t1_to_index], level.t2[t2_from_index:t2_to_index]))
opcodes_with_values.append(Opcode(
tag, t1_from_index, t1_to_index, t2_from_index, t2_to_index,
old_values = level.t1[t1_from_index: t1_to_index],
new_values = level.t2[t2_from_index: t2_to_index],
))
if tag == 'replace':
self._diff_by_forming_pairs_and_comparing_one_by_one(
level, local_tree=local_tree, parents_ids=parents_ids,
_original_type=_original_type, child_relationship_class=child_relationship_class,
t1_from_index=t1_from_index, t1_to_index=t1_to_index,
t2_from_index=t2_from_index, t2_to_index=t2_to_index,
)
elif tag == 'delete':
for index, x in enumerate(level.t1[t1_from_index:t1_to_index]):
change_level = level.branch_deeper(
x,
notpresent,
child_relationship_class=child_relationship_class,
child_relationship_param=index + t1_from_index,
child_relationship_param2=index + t1_from_index,
)
self._report_result('iterable_item_removed', change_level, local_tree=local_tree)
elif tag == 'insert':
for index, y in enumerate(level.t2[t2_from_index:t2_to_index]):
change_level = level.branch_deeper(
notpresent,
y,
child_relationship_class=child_relationship_class,
child_relationship_param=index + t2_from_index,
child_relationship_param2=index + t2_from_index,
)
self._report_result('iterable_item_added', change_level, local_tree=local_tree)
return opcodes_with_values
def _diff_str(self, level, local_tree=None):
"""Compare strings"""
if self.ignore_string_case:
level.t1 = level.t1.lower()
level.t2 = level.t2.lower()
if type(level.t1) == type(level.t2) and level.t1 == level.t2: # NOQA
return
# do we add a diff for convenience?
do_diff = True
t1_str = level.t1
t2_str = level.t2
if isinstance(level.t1, memoryview):
try:
t1_str = level.t1.tobytes().decode('ascii')
except UnicodeDecodeError:
do_diff = False
elif isinstance(level.t1, bytes_type):
try:
t1_str = level.t1.decode('ascii')
except UnicodeDecodeError:
do_diff = False
if isinstance(level.t2, memoryview):
try:
t2_str = level.t2.tobytes().decode('ascii')
except UnicodeDecodeError:
do_diff = False
elif isinstance(level.t2, bytes_type):
try:
t2_str = level.t2.decode('ascii')
except UnicodeDecodeError:
do_diff = False
if isinstance(level.t1, Enum):
t1_str = level.t1.value
if isinstance(level.t2, Enum):
t2_str = level.t2.value
if t1_str == t2_str:
return
if do_diff:
if '\n' in t1_str or isinstance(t2_str, str) and '\n' in t2_str:
diff = difflib.unified_diff(
t1_str.splitlines(), t2_str.splitlines(), lineterm='')
diff = list(diff)
if diff:
level.additional['diff'] = '\n'.join(diff)
self._report_result('values_changed', level, local_tree=local_tree)
def _diff_tuple(self, level, parents_ids, local_tree=None):
# Checking to see if it has _fields. Which probably means it is a named
# tuple.
try:
level.t1._asdict
# It must be a normal tuple
except AttributeError:
self._diff_iterable(level, parents_ids, local_tree=local_tree)
# We assume it is a namedtuple then
else:
self._diff_obj(level, parents_ids, is_namedtuple=True, local_tree=local_tree)
def _add_hash(self, hashes, item_hash, item, i):
if item_hash in hashes:
hashes[item_hash].indexes.append(i)
else:
hashes[item_hash] = IndexedHash(indexes=[i], item=item)
def _maybe_compute_hashes_parallel(self, level, obj):
"""Optionally hash iterable items in worker processes.
Returns a list of per-index ``item_hash`` values (or ``None`` for
items the worker could not process), aligned to ``enumerate(obj)``
order. Returns ``None`` when the section ran serially (no
``_mp_config``, below threshold, generator without ``__len__``,
unsafe inputs).
Iteration order is captured here once via ``list(obj)`` so the
parent loop and the worker job list see the same items even for
order-sensitive iterables like sets.
"""
mp_config = getattr(self, '_mp_config', None)
if mp_config is None or not mp_config.enabled:
return None, None
try:
n = len(obj)
except TypeError:
# Generators / unsized iterables: serial fallback. Materializing
# would change semantics (single-pass consumption).
return None, None
if not mp_config.should_parallelize(n):
return None, None
items = list(obj)
parent_base = level.path()
jobs = [
(item, "{}[{}]".format(parent_base, i))
for i, item in enumerate(items)
]
hashes = compute_hashes_parallel(
jobs=jobs,
deephash_parameters=self.deephash_parameters,
config=mp_config,
)
if hashes is None:
return None, None
return hashes, items
def _create_hashtable(self, level, t):
"""Create hashtable of {item_hash: (indexes, item)}"""
obj = getattr(level, t)
# Optionally precompute item hashes in worker processes. Workers
# operate on serial-order job indices and the parent merges back
# in that same order, so output is independent of worker
# completion order. ``items`` is the materialized iterable when
# parallel ran (set/dict iteration is deterministic per run but we
# need a single pass we can re-walk here).
parallel_hashes, materialized_items = self._maybe_compute_hashes_parallel(level, obj)
iterator = enumerate(materialized_items) if materialized_items is not None else enumerate(obj)
local_hashes = dict_()
for (i, item) in iterator:
try:
parent = "{}[{}]".format(level.path(), i)
if parallel_hashes is not None:
item_hash = parallel_hashes[i]
if item_hash is None:
# Worker could not process this item (KeyError or
# unprocessed marker). Mirror the serial pass:
# log once, skip.
self.log_err("Item %s was not processed while hashing "
"thus not counting this object." %
level.path())
continue
self._add_hash(hashes=local_hashes, item_hash=item_hash, item=item, i=i)
continue
# Note: in the DeepDiff we only calculate the hash of items when we have to.
# So self.hashes does not include hashes of all objects in t1 and t2.
# It only includes the ones needed when comparing iterables.
# The self.hashes dictionary gets shared between different runs of DeepHash
# So that any object that is already calculated to have a hash is not re-calculated.
deep_hash = DeepHash(
item,
hashes=self.hashes,
parent=parent,
apply_hash=True,
**self.deephash_parameters,
)
except UnicodeDecodeError as err:
err.reason = f"Can not produce a hash for {level.path()}: {err.reason}"
raise
except NotImplementedError:
raise
# except Exception as e: # pragma: no cover
# logger.error("Can not produce a hash for %s."
# "Not counting this object.\n %s" %
# (level.path(), e))
else:
try:
item_hash = deep_hash[item]
except KeyError:
pass
else:
if item_hash is unprocessed: # pragma: no cover
self.log_err("Item %s was not processed while hashing "
"thus not counting this object." %
level.path())
else:
self._add_hash(hashes=local_hashes, item_hash=item_hash, item=item, i=i)
# Also we hash the iterables themselves too so that we can later create cache keys from those hashes.
# When the per-item loop ran in parallel, child hashes were not merged into ``self.hashes``
# (cross-process id keys would not match). The iterable-level pass therefore re-hashes
# children serially; this is intentional — correctness over cache reuse for now.
DeepHash(
obj,
hashes=self.hashes,
parent=level.path(),
apply_hash=True,
**self.deephash_parameters,
)
return local_hashes
@staticmethod
@lru_cache(maxsize=2028)
def _get_distance_cache_key(added_hash, removed_hash):
key1, key2 = (added_hash, removed_hash) if added_hash > removed_hash else (removed_hash, added_hash)
if isinstance(key1, int):
# If the hash function produces integers we convert them to hex values.
# This was used when the default hash function was Murmur3 128bit which produces integers.
key1 = hex(key1).encode('utf-8')
key2 = hex(key2).encode('utf-8')
elif isinstance(key1, str):
key1 = key1.encode('utf-8')
key2 = key2.encode('utf-8')
return key1 + b'--' + key2 + b'dc'
def _get_rough_distance_of_hashed_objs(
self, added_hash, removed_hash, added_hash_obj, removed_hash_obj, _original_type=None):
# We need the rough distance between the 2 objects to see if they qualify to be pairs or not
_distance = cache_key = None
if self._stats[DISTANCE_CACHE_ENABLED]:
cache_key = self._get_distance_cache_key(added_hash, removed_hash)
cached_distance = self._distance_cache.get(cache_key)
if cached_distance is not_found:
_distance = None
else:
self._stats[DISTANCE_CACHE_HIT_COUNT] += 1
_distance = cast(float, cached_distance)
if _distance is None:
# We can only cache the rough distance and not the actual diff result for reuse.
# The reason is that we have modified the parameters explicitly so they are different and can't
# be used for diff reporting
diff = DeepDiff(
removed_hash_obj.item, added_hash_obj.item,
_parameters=self._parameters,
_shared_parameters=self._shared_parameters,
view=DELTA_VIEW,
_original_type=_original_type,
iterable_compare_func=self.iterable_compare_func,
)
_distance = diff._get_rough_distance()
if cache_key and self._stats[DISTANCE_CACHE_ENABLED]:
self._distance_cache.set(cache_key, value=_distance)
return _distance
def _maybe_compute_pair_distances_parallel(
self, hashes_added, hashes_removed, t1_hashtable, t2_hashtable,
parents_ids, _original_type, pre_calced_distances):
"""Optionally run distance computation for non-cached pairs in workers.
Returns a dict ``{(added_hash, removed_hash): distance}`` for pairs
whose distance was computed in parallel, or ``None`` if the section
ran serially (below threshold, unsafe inputs, no _mp_config, etc.).
The job list is built in the exact order of the serial nested loop
so the parent merge order is identical regardless of how many workers
run or which one finishes first.
"""
mp_config = getattr(self, '_mp_config', None)
if mp_config is None or not mp_config.enabled:
return None
# Build candidate job list in stable nested-loop order. We skip pairs
# that the serial loop also skips (loop detection, pre-calculated
# distance, distance cache hit) so workers only get real work.
jobs = []
cache_enabled = self._stats[DISTANCE_CACHE_ENABLED]
for added_hash in hashes_added:
for removed_hash in hashes_removed:
added_hash_obj = t2_hashtable[added_hash]
removed_hash_obj = t1_hashtable[removed_hash]
if id(removed_hash_obj.item) in parents_ids:
continue
if pre_calced_distances and pre_calced_distances.get(
"{}--{}".format(added_hash, removed_hash)) is not None:
continue
if cache_enabled:
cache_key = self._get_distance_cache_key(added_hash, removed_hash)
if self._distance_cache.get(cache_key) is not not_found:
# Serial path will pull this from cache; no worker
# needed and we keep cache-hit accounting in the
# parent.
continue
jobs.append((added_hash, removed_hash, added_hash_obj.item, removed_hash_obj.item))
if not mp_config.should_parallelize(len(jobs)):
return None
result = compute_distances_parallel(
jobs=jobs,
parameters=self._parameters,
original_type=_original_type,
iterable_compare_func=self.iterable_compare_func,
config=mp_config,
)
if result is None:
return None
distances, worker_stats = result
self._merge_worker_stats(worker_stats)
return distances
def _merge_worker_stats(self, worker_stats):
"""Aggregate one parallel-batch's worker ``_stats`` delta into self._stats.
Counters (DIFF / PASSES / DISTANCE CACHE HIT) sum into the matching
``WORKER_*`` keys; limit flags OR-merge into the parent's existing
MAX_*_LIMIT_REACHED flags so any worker hitting a guard surfaces the
same warning state on the public ``get_stats()`` output.
"""
if not worker_stats:
return
self._stats[WORKER_DIFF_COUNT] += int(worker_stats.get('DIFF COUNT', 0) or 0)
self._stats[WORKER_PASSES_COUNT] += int(worker_stats.get('PASSES COUNT', 0) or 0)
self._stats[WORKER_DISTANCE_CACHE_HIT_COUNT] += int(
worker_stats.get('DISTANCE CACHE HIT COUNT', 0) or 0)
self._stats[WORKER_BATCH_COUNT] += 1
if worker_stats.get(MAX_PASS_LIMIT_REACHED):
self._stats[MAX_PASS_LIMIT_REACHED] = True
if worker_stats.get(MAX_DIFF_LIMIT_REACHED):
self._stats[MAX_DIFF_LIMIT_REACHED] = True
def _get_most_in_common_pairs_in_iterables(
self, hashes_added, hashes_removed, t1_hashtable, t2_hashtable, parents_ids, _original_type):
"""
Get the closest pairs between items that are removed and items that are added.
returns a dictionary of hashes that are closest to each other.
The dictionary is going to be symmetrical so any key will be a value too and otherwise.
Note that due to the current reporting structure in DeepDiff, we don't compare an item that
was added to an item that is in both t1 and t2.
For example
[{1, 2}, {4, 5, 6}]
[{1, 2}, {1, 2, 3}]
is only compared between {4, 5, 6} and {1, 2, 3} even though technically {1, 2, 3} is
just one item different than {1, 2}
Perhaps in future we can have a report key that is item duplicated and modified instead of just added.
"""
cache_key = None
if self._stats[DISTANCE_CACHE_ENABLED]:
cache_key = combine_hashes_lists(items=[hashes_added, hashes_removed], prefix='pairs_cache')
cached_pairs = self._distance_cache.get(cache_key)
if cached_pairs is not_found:
cached_pairs = None
else:
return cast(dict, cached_pairs).copy()
# A dictionary of hashes to distances and each distance to an ordered set of hashes.
# It tells us about the distance of each object from other objects.
# And the objects with the same distances are grouped together in an ordered set.
# It also includes a "max" key that is just the value of the biggest current distance in the
# most_in_common_pairs dictionary.
def defaultdict_orderedset():
return defaultdict(SetOrdered)
most_in_common_pairs = defaultdict(defaultdict_orderedset)
pairs = dict_()
pre_calced_distances = None
if hashes_added and hashes_removed and np and len(hashes_added) > 1 and len(hashes_removed) > 1:
# pre-calculates distances ONLY for 1D arrays whether an _original_type
# was explicitly passed or a homogeneous array is detected.
# Numpy is needed for this optimization.
pre_calced_distances = self._precalculate_numpy_arrays_distance(
hashes_added, hashes_removed, t1_hashtable, t2_hashtable, _original_type)
if hashes_added and hashes_removed \
and self.iterable_compare_func \
and len(hashes_added) > 0 and len(hashes_removed) > 0:
pre_calced_distances = self._precalculate_distance_by_custom_compare_func(
hashes_added, hashes_removed, t1_hashtable, t2_hashtable, _original_type)
# Optionally precompute non-cached distances in worker processes.
# Returns a dict keyed by (added_hash, removed_hash). Pair selection
# below stays serial and walks the same nested loop order, so the
# public output is independent of worker completion order.
parallel_distances = self._maybe_compute_pair_distances_parallel(
hashes_added, hashes_removed, t1_hashtable, t2_hashtable,
parents_ids, _original_type, pre_calced_distances)
for added_hash in hashes_added:
for removed_hash in hashes_removed:
added_hash_obj = t2_hashtable[added_hash]
removed_hash_obj = t1_hashtable[removed_hash]
# Loop is detected
if id(removed_hash_obj.item) in parents_ids:
continue
_distance = None
if pre_calced_distances:
_distance = pre_calced_distances.get("{}--{}".format(added_hash, removed_hash))
if _distance is None and parallel_distances is not None:
_distance = parallel_distances.get((added_hash, removed_hash))
if _distance is None:
_distance = self._get_rough_distance_of_hashed_objs(
added_hash, removed_hash, added_hash_obj, removed_hash_obj, _original_type)
_distance = cast(float, _distance)
# Left for future debugging
# print(f'{Fore.RED}distance of {added_hash_obj.item} and {removed_hash_obj.item}: {_distance}{Style.RESET_ALL}')
# Discard potential pairs that are too far.
if _distance >= self.cutoff_distance_for_pairs:
continue
pairs_of_item = most_in_common_pairs[added_hash]
pairs_of_item[_distance].add(removed_hash)
used_to_hashes = set()
distances_to_from_hashes = defaultdict(SetOrdered)
for from_hash, distances_to_to_hashes in most_in_common_pairs.items():
# del distances_to_to_hashes['max']
for dist in distances_to_to_hashes:
distances_to_from_hashes[dist].add(from_hash)
for dist in sorted(distances_to_from_hashes.keys()):
from_hashes = distances_to_from_hashes[dist]
while from_hashes:
from_hash = from_hashes.pop()
if from_hash not in used_to_hashes:
to_hashes = most_in_common_pairs[from_hash][dist]
while to_hashes:
to_hash = to_hashes.pop()
if to_hash not in used_to_hashes:
used_to_hashes.add(from_hash)
used_to_hashes.add(to_hash)
# Left for future debugging:
# print(f'{bcolors.FAIL}Adding {t2_hashtable[from_hash].item} as a pairs of {t1_hashtable[to_hash].item} with distance of {dist}{bcolors.ENDC}')
pairs[from_hash] = to_hash
inverse_pairs = {v: k for k, v in pairs.items()}
pairs.update(inverse_pairs)
if cache_key and self._stats[DISTANCE_CACHE_ENABLED]:
self._distance_cache.set(cache_key, value=pairs)
return pairs.copy()
def _subtree_parallel_safe(self):
"""Return True if paired-subtree diffs in this run can be sent to workers.
Excluded features are ones whose semantics depend on the *parent's*
absolute path or on parent-process state, neither of which is visible
in a worker:
- ``custom_operators`` (per docs/multi_processing.md) can call
``custom_report_result`` and mutate the parent diff instance.
- ``exclude_obj_callback`` / ``include_obj_callback`` (and their
``_strict`` variants) receive the level path; in a worker that path
is rooted at the subtree, not the original tree, so they would fire
on the wrong paths.
- ``ignore_order_func`` is also called with the level and would see
worker-local paths.
Path-only filters (``exclude_paths`` / ``include_paths`` /
``exclude_regex_paths``) are handled by re-applying ``_skip_this``
after rebasing rather than disabling parallelism.
"""
if self.custom_operators:
return False
if self.exclude_obj_callback or self.exclude_obj_callback_strict:
return False
if self.include_obj_callback or self.include_obj_callback_strict:
return False
if self.ignore_order_func:
return False
return True
def _rebase_subtree_leaf(self, leaf, change_level):
"""Splice a worker-built leaf chain onto the parent's ``change_level``.
The worker constructed ``leaf`` inside a fresh ``DeepDiff`` whose root
DiffLevel holds the paired items themselves; that root is irrelevant
once we're back in the parent. We replace it with a *fresh copy* of
``change_level`` (so each leaf gets its own up-chain — DiffLevel.up
is shared by reference, and reusing one chain across leaves would
scramble paths).
Returns the rebased leaf. Path caches along the chain are cleared so
``leaf.path()`` recomputes against the new up-chain.
"""
# Walk up to find the worker root (up=None).
worker_root = leaf
while worker_root.up is not None:
worker_root = worker_root.up
new_cl = change_level.copy() # fresh, independent chain; new_cl is bottom
if worker_root is leaf:
# The worker reported at the very root of its diff (e.g. the two
# paired items differ at the top level — type_changes,
# values_changed). Transfer the report payload onto our fresh
# change_level copy.
new_cl.report_type = leaf.report_type
new_cl.additional = leaf.additional
cur = new_cl
while cur is not None:
cur._path = dict_()
cur = cur.up
return new_cl
first_under_root = worker_root.down
# Splice: new_cl takes worker_root's place. Setting .down auto-sets
# the opposite .up link (see DiffLevel.__setattr__).
new_cl.down = first_under_root
if worker_root.t1_child_rel is not None:
new_cl.t1_child_rel = ChildRelationship.create(
klass=worker_root.t1_child_rel.__class__,
parent=new_cl.t1, child=first_under_root.t1,
param=worker_root.t1_child_rel.param,
)
if worker_root.t2_child_rel is not None:
new_cl.t2_child_rel = ChildRelationship.create(
klass=worker_root.t2_child_rel.__class__,
parent=new_cl.t2, child=first_under_root.t2,
param=worker_root.t2_child_rel.param,
)
# Clear path cache on the entire chain so path() recomputes against
# the new up-chain.
cur = leaf
while cur is not None:
cur._path = dict_()
cur = cur.up
return leaf
def _dispatch_subtree_jobs(self, pending_jobs, _original_type, local_tree):
"""Run deferred paired-subtree diffs (parallel when allowed, else serial).
``pending_jobs`` is the list of ``(change_level, t1_item, t2_item,
parents_ids_added)`` tuples in the exact order the inline serial code
would have visited them. Parallel results are merged in that same
order regardless of worker completion order, so output is identical
to the equivalent serial run.
"""
if not pending_jobs:
return
mp_config = getattr(self, '_mp_config', None)
parallel_results = None
if (mp_config is not None and mp_config.enabled
and mp_config.should_parallelize(len(pending_jobs))):
jobs_payload = [(t1_item, t2_item) for (_, t1_item, t2_item, _) in pending_jobs]
outcome = compute_subtree_diffs_parallel(
jobs=jobs_payload,
parameters=self._parameters,
original_type=_original_type,
config=mp_config,
)
if outcome is not None:
parallel_results, worker_stats = outcome
self._merge_worker_stats(worker_stats)
if parallel_results is None:
# Below threshold or unsafe inputs — run inline-equivalent serial.
# Walking pending_jobs in order matches how inline serial would
# have run them; the parent tree fills up the same way.
for change_level, _t1_item, _t2_item, parents_ids_added in pending_jobs:
self._diff(change_level, parents_ids_added, local_tree=local_tree)
return
target_tree = self.tree if local_tree is None else local_tree
for (change_level, _t1_item, _t2_item, _parents_ids_added), entries in zip(
pending_jobs, parallel_results):
for report_type, leaf in entries:
rebased_leaf = self._rebase_subtree_leaf(leaf, change_level)
# Re-apply path-based filters in the parent — exclude_paths
# and friends were not applied correctly inside the worker
# because the worker's level paths are subtree-relative.
if self._skip_this(rebased_leaf):
continue
rebased_leaf.report_type = report_type
target_tree[report_type].add(rebased_leaf)
def _diff_iterable_with_deephash(self, level, parents_ids, _original_type=None, local_tree=None):
"""Diff of hashable or unhashable iterables. Only used when ignoring the order."""
full_t1_hashtable = self._create_hashtable(level, 't1')
full_t2_hashtable = self._create_hashtable(level, 't2')
t1_hashes = SetOrdered(full_t1_hashtable.keys())
t2_hashes = SetOrdered(full_t2_hashtable.keys())
hashes_added = t2_hashes - t1_hashes
hashes_removed = t1_hashes - t2_hashes
# Deciding whether to calculate pairs or not.
if (len(hashes_added) + len(hashes_removed)) / (len(full_t1_hashtable) + len(full_t2_hashtable) + 1) > self.cutoff_intersection_for_pairs:
get_pairs = False
else:
get_pairs = True
# reduce the size of hashtables
if self.report_repetition:
t1_hashtable = full_t1_hashtable
t2_hashtable = full_t2_hashtable
else:
t1_hashtable = {k: v for k, v in full_t1_hashtable.items() if k in hashes_removed}
t2_hashtable = {k: v for k, v in full_t2_hashtable.items() if k in hashes_added}
if self._stats[PASSES_COUNT] < self.max_passes and get_pairs:
self._stats[PASSES_COUNT] += 1
pairs = self._get_most_in_common_pairs_in_iterables(
hashes_added, hashes_removed, t1_hashtable, t2_hashtable, parents_ids, _original_type)
elif get_pairs:
if not self._stats[MAX_PASS_LIMIT_REACHED]:
self._stats[MAX_PASS_LIMIT_REACHED] = True
logger.warning(MAX_PASSES_REACHED_MSG.format(self.max_passes))
pairs = dict_()
else:
pairs = dict_()
def get_other_pair(hash_value, in_t1=True):
"""
Gets the other paired indexed hash item to the hash_value in the pairs dictionary
in_t1: are we looking for the other pair in t1 or t2?
"""
if in_t1:
hashtable = t1_hashtable
the_other_hashes = hashes_removed
else:
hashtable = t2_hashtable
the_other_hashes = hashes_added
other = pairs.pop(hash_value, notpresent)
if other is notpresent:
other = notpresent_indexed
else:
# The pairs are symmetrical.
# removing the other direction of pair
# so it does not get used.
del pairs[other]
the_other_hashes.remove(other)
other = hashtable[other]
return other
# Phase 3: paired-subtree diffs may be deferred so they can run in a
# worker pool. ``pending_subtree_jobs is None`` keeps the inline
# serial path (and the existing ordering of inline ``_diff`` calls
# vs. ``_report_result`` calls) — so any feature that disables
# subtree parallelism degrades cleanly to today's behavior.
mp_config = getattr(self, '_mp_config', None)
use_mp = (
mp_config is not None and mp_config.enabled
and self._subtree_parallel_safe()
)
pending_subtree_jobs = [] if use_mp else None
if self.report_repetition:
for hash_value in hashes_added:
if self._count_diff() is StopIteration:
return # pragma: no cover. This is already covered for addition (when report_repetition=False).
other = get_other_pair(hash_value)
item_id = id(other.item)
indexes = t2_hashtable[hash_value].indexes if other.item is notpresent else other.indexes
# When we report repetitions, we want the child_relationship_param2 only if there is no repetition.
# Because when there is a repetition, we report it in a different way (iterable_items_added_at_indexes for example).
# When there is no repetition, we want child_relationship_param2 so that we report the "new_path" correctly.
if len(t2_hashtable[hash_value].indexes) == 1:
index2 = t2_hashtable[hash_value].indexes[0]
else:
index2 = None
for i in indexes:
change_level = level.branch_deeper(
other.item,
t2_hashtable[hash_value].item,
child_relationship_class=SubscriptableIterableRelationship,
child_relationship_param=i,
child_relationship_param2=index2,
)
if other.item is notpresent:
self._report_result('iterable_item_added', change_level, local_tree=local_tree)
else:
parents_ids_added = add_to_frozen_set(parents_ids, item_id)
if pending_subtree_jobs is None:
self._diff(change_level, parents_ids_added, local_tree=local_tree)
else:
pending_subtree_jobs.append(
(change_level, other.item, t2_hashtable[hash_value].item, parents_ids_added))
for hash_value in hashes_removed:
if self._count_diff() is StopIteration:
return # pragma: no cover. This is already covered for addition.
other = get_other_pair(hash_value, in_t1=False)
item_id = id(other.item)
# When we report repetitions, we want the child_relationship_param2 only if there is no repetition.
# Because when there is a repetition, we report it in a different way (iterable_items_added_at_indexes for example).
# When there is no repetition, we want child_relationship_param2 so that we report the "new_path" correctly.
if other.item is notpresent or len(other.indexes) > 1:
index2 = None
else:
index2 = other.indexes[0]
for i in t1_hashtable[hash_value].indexes:
change_level = level.branch_deeper(
t1_hashtable[hash_value].item,
other.item,
child_relationship_class=SubscriptableIterableRelationship,
child_relationship_param=i,
child_relationship_param2=index2,
)
if other.item is notpresent:
self._report_result('iterable_item_removed', change_level, local_tree=local_tree)
else:
# I was not able to make a test case for the following 2 lines since the cases end up
# getting resolved above in the hashes_added calcs. However I am leaving these 2 lines
# in case things change in future.
parents_ids_added = add_to_frozen_set(parents_ids, item_id) # pragma: no cover.
if pending_subtree_jobs is None: # pragma: no cover.
self._diff(change_level, parents_ids_added, local_tree=local_tree)
else: # pragma: no cover.
pending_subtree_jobs.append(
(change_level, t1_hashtable[hash_value].item, other.item, parents_ids_added))
items_intersect = t2_hashes.intersection(t1_hashes)
for hash_value in items_intersect:
t1_indexes = t1_hashtable[hash_value].indexes
t2_indexes = t2_hashtable[hash_value].indexes
t1_indexes_len = len(t1_indexes)
t2_indexes_len = len(t2_indexes)
if t1_indexes_len != t2_indexes_len: # this is a repetition change!
# create "change" entry, keep current level untouched to handle further changes
repetition_change_level = level.branch_deeper(
t1_hashtable[hash_value].item,
t2_hashtable[hash_value].item, # nb: those are equal!
child_relationship_class=SubscriptableIterableRelationship,
child_relationship_param=t1_hashtable[hash_value]
.indexes[0])
repetition_change_level.additional['repetition'] = RemapDict(
old_repeat=t1_indexes_len,
new_repeat=t2_indexes_len,
old_indexes=t1_indexes,
new_indexes=t2_indexes)
self._report_result('repetition_change',
repetition_change_level, local_tree=local_tree)
else:
for hash_value in hashes_added:
if self._count_diff() is StopIteration:
return
other = get_other_pair(hash_value)
item_id = id(other.item)
index = t2_hashtable[hash_value].indexes[0] if other.item is notpresent else other.indexes[0]
index2 = t2_hashtable[hash_value].indexes[0]
change_level = level.branch_deeper(
other.item,
t2_hashtable[hash_value].item,
child_relationship_class=SubscriptableIterableRelationship,
child_relationship_param=index,
child_relationship_param2=index2,
)
if other.item is notpresent:
self._report_result('iterable_item_added', change_level, local_tree=local_tree)
else:
parents_ids_added = add_to_frozen_set(parents_ids, item_id)
if pending_subtree_jobs is None:
self._diff(change_level, parents_ids_added, local_tree=local_tree)
else:
pending_subtree_jobs.append(
(change_level, other.item, t2_hashtable[hash_value].item, parents_ids_added))
for hash_value in hashes_removed:
if self._count_diff() is StopIteration:
return # pragma: no cover. This is already covered for addition.
other = get_other_pair(hash_value, in_t1=False)
item_id = id(other.item)
index = t1_hashtable[hash_value].indexes[0]
index2 = t1_hashtable[hash_value].indexes[0] if other.item is notpresent else other.indexes[0]
change_level = level.branch_deeper(
t1_hashtable[hash_value].item,
other.item,
child_relationship_class=SubscriptableIterableRelationship,
child_relationship_param=index,
child_relationship_param2=index2,
)
if other.item is notpresent:
self._report_result('iterable_item_removed', change_level, local_tree=local_tree)
else:
# Just like the case when report_repetition = True, these lines never run currently.
# However they will stay here in case things change in future.
parents_ids_added = add_to_frozen_set(parents_ids, item_id) # pragma: no cover.
if pending_subtree_jobs is None: # pragma: no cover.
self._diff(change_level, parents_ids_added, local_tree=local_tree)
else: # pragma: no cover.
pending_subtree_jobs.append(
(change_level, t1_hashtable[hash_value].item, other.item, parents_ids_added))
if pending_subtree_jobs:
self._dispatch_subtree_jobs(pending_subtree_jobs, _original_type, local_tree)
def _diff_booleans(self, level, local_tree=None):
if level.t1 != level.t2:
self._report_result('values_changed', level, local_tree=local_tree)
def _diff_numbers(self, level, local_tree=None, report_type_change=True):
"""Diff Numbers"""
if report_type_change:
t1_type = "number" if self.ignore_numeric_type_changes else level.t1.__class__.__name__
t2_type = "number" if self.ignore_numeric_type_changes else level.t2.__class__.__name__
else:
t1_type = t2_type = ''
if self.use_log_scale:
if not logarithmic_similarity(level.t1, level.t2, threshold=self.log_scale_similarity_threshold):
self._report_result('values_changed', level, local_tree=local_tree)
elif self.math_epsilon is not None:
if not is_close(level.t1, level.t2, abs_tol=self.math_epsilon):
self._report_result('values_changed', level, local_tree=local_tree)
elif self.significant_digits is None:
if level.t1 != level.t2:
self._report_result('values_changed', level, local_tree=local_tree)
else:
# Bernhard10: I use string formatting for comparison, to be consistent with usecases where
# data is read from files that were previously written from python and
# to be consistent with on-screen representation of numbers.
# Other options would be abs(t1-t2)<10**-self.significant_digits
# or math.is_close (python3.5+)
# Note that abs(3.25-3.251) = 0.0009999999999998899 < 0.001
# Note also that "{:.3f}".format(1.1135) = 1.113, but "{:.3f}".format(1.11351) = 1.114
# For Decimals, format seems to round 2.5 to 2 and 3.5 to 4 (to closest even number)
t1_s = self.number_to_string(level.t1,
significant_digits=self.significant_digits,
number_format_notation=self.number_format_notation) # type: ignore
t2_s = self.number_to_string(level.t2,
significant_digits=self.significant_digits,
number_format_notation=self.number_format_notation) # type: ignore
t1_s = KEY_TO_VAL_STR.format(t1_type, t1_s)
t2_s = KEY_TO_VAL_STR.format(t2_type, t2_s)
if t1_s != t2_s:
self._report_result('values_changed', level, local_tree=local_tree)
def _diff_ipranges(self, level, local_tree=None):
"""Diff IP ranges"""
if str(level.t1) != str(level.t2):
self._report_result('values_changed', level, local_tree=local_tree)
def _diff_datetime(self, level, local_tree=None):
"""Diff DateTimes"""
level.t1 = datetime_normalize(self.truncate_datetime, level.t1, default_timezone=self.default_timezone)
level.t2 = datetime_normalize(self.truncate_datetime, level.t2, default_timezone=self.default_timezone)
if level.t1 != level.t2:
self._report_result('values_changed', level, local_tree=local_tree)
def _diff_time(self, level, local_tree=None):
"""Diff DateTimes"""
if self.truncate_datetime:
level.t1 = datetime_normalize(self.truncate_datetime, level.t1, default_timezone=self.default_timezone)
level.t2 = datetime_normalize(self.truncate_datetime, level.t2, default_timezone=self.default_timezone)
if level.t1 != level.t2:
self._report_result('values_changed', level, local_tree=local_tree)
def _diff_uuids(self, level, local_tree=None):
"""Diff UUIDs"""
if level.t1.int != level.t2.int:
self._report_result('values_changed', level, local_tree=local_tree)
def _diff_numpy_array(self, level, parents_ids=frozenset(), local_tree=None):
"""Diff numpy arrays"""
if level.path() not in self._numpy_paths:
self._numpy_paths[level.path()] = get_type(level.t2).__name__
if np is None:
# This line should never be run. If it is ever called means the type check detected a numpy array
# which means numpy module needs to be available. So np can't be None.
raise ImportError(CANT_FIND_NUMPY_MSG) # pragma: no cover
if (self.ignore_order_func and not self.ignore_order_func(level)) or not self.ignore_order:
# fast checks
if self.significant_digits is None:
if np.array_equal(level.t1, level.t2, equal_nan=self.ignore_nan_inequality):
return # all good
else:
try:
np.testing.assert_almost_equal(level.t1, level.t2, decimal=self.significant_digits)
except TypeError:
np.array_equal(level.t1, level.t2, equal_nan=self.ignore_nan_inequality)
except AssertionError:
pass # do detailed checking below
else:
return # all good
# compare array meta-data
_original_type = level.t1.dtype
if level.t1.shape != level.t2.shape:
# arrays are converted to python lists so that certain features of DeepDiff can apply on them easier.
# They will be converted back to Numpy at their final dimension.
level.t1 = level.t1.tolist()
level.t2 = level.t2.tolist()
self._diff_iterable(level, parents_ids, _original_type=_original_type, local_tree=local_tree)
else:
# metadata same -- the difference is in the content
shape = level.t1.shape
dimensions = len(shape)
if dimensions == 1:
self._diff_iterable(level, parents_ids, _original_type=_original_type, local_tree=local_tree)
elif (self.ignore_order_func and self.ignore_order_func(level)) or self.ignore_order:
# arrays are converted to python lists so that certain features of DeepDiff can apply on them easier.
# They will be converted back to Numpy at their final dimension.
level.t1 = level.t1.tolist()
level.t2 = level.t2.tolist()
self._diff_iterable_with_deephash(level, parents_ids, _original_type=_original_type, local_tree=local_tree)
else:
for (t1_path, t1_row), (t2_path, t2_row) in zip(
get_numpy_ndarray_rows(level.t1, shape),
get_numpy_ndarray_rows(level.t2, shape)):
new_level = level.branch_deeper(
t1_row,
t2_row,
child_relationship_class=NumpyArrayRelationship,
child_relationship_param=t1_path,
child_relationship_param2=t2_path,
)
self._diff_iterable_in_order(new_level, parents_ids, _original_type=_original_type, local_tree=local_tree)
def _diff_types(self, level, local_tree=None):
"""Diff types"""
level.report_type = 'type_changes'
self._report_result('type_changes', level, local_tree=local_tree)
def _count_diff(self):
if (self.max_diffs is not None and self._stats[DIFF_COUNT] > self.max_diffs):
if not self._stats[MAX_DIFF_LIMIT_REACHED]:
self._stats[MAX_DIFF_LIMIT_REACHED] = True
logger.warning(MAX_DIFFS_REACHED_MSG.format(self.max_diffs))
return StopIteration
self._stats[DIFF_COUNT] += 1
if self.cache_size and self.cache_tuning_sample_size:
self._auto_tune_cache()
def _auto_tune_cache(self):
take_sample = (self._stats[DIFF_COUNT] % self.cache_tuning_sample_size == 0)
if self.cache_tuning_sample_size:
if self._stats[DISTANCE_CACHE_ENABLED]:
if take_sample:
self._auto_off_cache()
# Turn on the cache once in a while
elif self._stats[DIFF_COUNT] % self._shared_parameters[_ENABLE_CACHE_EVERY_X_DIFF] == 0:
self.progress_logger('Re-enabling the distance and level caches.')
# decreasing the sampling frequency
self._shared_parameters[_ENABLE_CACHE_EVERY_X_DIFF] *= 10
self._stats[DISTANCE_CACHE_ENABLED] = True
if take_sample:
for key in (PREVIOUS_DIFF_COUNT, PREVIOUS_DISTANCE_CACHE_HIT_COUNT):
self._stats[key] = self._stats[key[9:]]
def _auto_off_cache(self):
"""
Auto adjust the cache based on the usage
"""
if self._stats[DISTANCE_CACHE_ENABLED]:
angle = (self._stats[DISTANCE_CACHE_HIT_COUNT] - self._stats['PREVIOUS {}'.format(DISTANCE_CACHE_HIT_COUNT)]) / (self._stats[DIFF_COUNT] - self._stats[PREVIOUS_DIFF_COUNT])
if angle < self.CACHE_AUTO_ADJUST_THRESHOLD:
self._stats[DISTANCE_CACHE_ENABLED] = False
self.progress_logger('Due to minimal cache hits, {} is disabled.'.format('distance cache'))
def _use_custom_operator(self, level):
"""
For each level we check all custom operators.
If any one of them was a match for the level, we run the diff of the operator.
If the operator returned True, the operator must have decided these objects should not
be compared anymore. It might have already reported their results.
In that case the report will appear in the final results of this diff.
Otherwise basically the 2 objects in the level are being omitted from the results.
"""
for operator in self.custom_operators:
if operator.match(level):
prevent_default = operator.give_up_diffing(level=level, diff_instance=self)
if prevent_default:
return True
return False
def _diff(self, level, parents_ids=frozenset(), _original_type=None, local_tree=None):
"""
The main diff method
**parameters**
level: the tree level or tree node
parents_ids: the ids of all the parent objects in the tree from the current node.
_original_type: If the objects had an original type that was different than what currently exists in the level.t1 and t2
"""
if self._count_diff() is StopIteration:
return
if self._use_custom_operator(level):
return
if level.t1 is level.t2:
return
if self._skip_this(level):
return
report_type_change = True
if get_type(level.t1) != get_type(level.t2):
for type_group in self.ignore_type_in_groups:
if self.type_check_func(level.t1, type_group) and self.type_check_func(level.t2, type_group):
report_type_change = False
break
if self.use_enum_value and isinstance(level.t1, Enum):
level.t1 = level.t1.value
report_type_change = False
if self.use_enum_value and isinstance(level.t2, Enum):
level.t2 = level.t2.value
report_type_change = False
if report_type_change:
self._diff_types(level, local_tree=local_tree)
return
# This is an edge case where t1=None or t2=None and None is in the ignore type group.
if level.t1 is None or level.t2 is None:
self._report_result('values_changed', level, local_tree=local_tree)
return
if self.ignore_nan_inequality and isinstance(level.t1, (float, np_floating)) and str(level.t1) == str(level.t2) == 'nan':
return
if isinstance(level.t1, booleans):
self._diff_booleans(level, local_tree=local_tree)
elif isinstance(level.t1, strings):
# Special handling when comparing string with UUID and ignore_uuid_types is True
if self.ignore_uuid_types and isinstance(level.t2, uuids):
try:
# Convert string to UUID for comparison
t1_uuid = uuid.UUID(str(level.t1))
if t1_uuid.int != level.t2.int:
self._report_result('values_changed', level, local_tree=local_tree)
except (ValueError, AttributeError):
# If string is not a valid UUID, report as changed
self._report_result('values_changed', level, local_tree=local_tree)
else:
self._diff_str(level, local_tree=local_tree)
elif isinstance(level.t1, datetime.datetime):
self._diff_datetime(level, local_tree=local_tree)
elif isinstance(level.t1, ipranges):
self._diff_ipranges(level, local_tree=local_tree)
elif isinstance(level.t1, (datetime.date, datetime.timedelta, datetime.time)):
self._diff_time(level, local_tree=local_tree)
elif isinstance(level.t1, uuids):
# Special handling when comparing UUID with string and ignore_uuid_types is True
if self.ignore_uuid_types and isinstance(level.t2, str):
try:
# Convert string to UUID for comparison
t2_uuid = uuid.UUID(level.t2)
if level.t1.int != t2_uuid.int:
self._report_result('values_changed', level, local_tree=local_tree)
except (ValueError, AttributeError):
# If string is not a valid UUID, report as changed
self._report_result('values_changed', level, local_tree=local_tree)
else:
self._diff_uuids(level, local_tree=local_tree)
elif isinstance(level.t1, numbers):
self._diff_numbers(level, local_tree=local_tree, report_type_change=report_type_change)
elif isinstance(level.t1, Mapping):
self._diff_dict(level, parents_ids, local_tree=local_tree)
elif isinstance(level.t1, tuple):
self._diff_tuple(level, parents_ids, local_tree=local_tree)
elif isinstance(level.t1, (set, frozenset, SetOrdered)):
self._diff_set(level, local_tree=local_tree)
elif isinstance(level.t1, np_ndarray):
self._diff_numpy_array(level, parents_ids, local_tree=local_tree)
elif isinstance(level.t1, PydanticBaseModel):
self._diff_obj(level, parents_ids, local_tree=local_tree, is_pydantic_object=True)
elif isinstance(level.t1, Iterable):
self._diff_iterable(level, parents_ids, _original_type=_original_type, local_tree=local_tree)
elif isinstance(level.t1, Enum):
self._diff_enum(level, parents_ids, local_tree=local_tree)
else:
self._diff_obj(level, parents_ids)
def _get_view_results(self, view, verbose_level=None):
"""
Get the results based on the view
"""
result = self.tree
if not self.report_repetition: # and self.is_root:
result.mutual_add_removes_to_become_value_changes()
if view == TREE_VIEW:
pass
elif view == TEXT_VIEW:
effective_verbose_level = verbose_level if verbose_level is not None else self.verbose_level
result = TextResult(tree_results=self.tree, verbose_level=effective_verbose_level)
result.remove_empty_keys()
elif view == DELTA_VIEW:
result = self._to_delta_dict(report_repetition_required=False)
elif view == COLORED_VIEW:
result = ColoredView(t2=self.t2, tree_result=self.tree, compact=False)
elif view == COLORED_COMPACT_VIEW:
result = ColoredView(t2=self.t2, tree_result=self.tree, compact=True)
else:
raise ValueError(INVALID_VIEW_MSG.format(view))
return result
@staticmethod
def _get_key_for_group_by(row, group_by, item_name):
"""
Get the key value to group a row by, using the specified group_by parameter.
Example
>>> row = {'first': 'John', 'middle': 'Joe', 'last': 'Smith'}
>>> DeepDiff._get_key_for_group_by(row, 'first', 't1')
'John'
>>> nested_row = {'id': 123, 'demographics': {'names': {'first': 'John', 'middle': 'Joe', 'last': 'Smith'}}}
>>> group_by = lambda x: x['demographics']['names']['first']
>>> DeepDiff._get_key_for_group_by(nested_row, group_by, 't1')
'John'
Args:
row (dict): The dictionary (row) to extract the group by key from.
group_by (str or callable): The key name or function to call to get to the key value to group by.
item_name (str): The name of the item, used for error messages.
Returns:
str: The key value to group by.
Raises:
KeyError: If the specified key is not found in the row.
"""
try:
if callable(group_by):
return group_by(row)
return row.pop(group_by)
except KeyError:
logger.error("Unable to group {} by {}. The key is missing in {}".format(item_name, group_by, row))
raise
def _group_iterable_to_dict(self, item, group_by, item_name):
"""
Convert a list of dictionaries into a dictionary of dictionaries
where the key is the value of the group_by key in each dictionary.
"""
group_by_level2 = None
if isinstance(group_by, (list, tuple)):
group_by_level1 = group_by[0]
if len(group_by) > 1:
group_by_level2 = group_by[1]
else:
group_by_level1 = group_by
if isinstance(item, Iterable) and not isinstance(item, Mapping):
result = {}
item_copy = deepcopy(item)
for row in item_copy:
if isinstance(row, Mapping):
key1 = self._get_key_for_group_by(row, group_by_level1, item_name)
# Track keys created by group_by to avoid type prefixing later
if hasattr(self, 'group_by_keys'):
self.group_by_keys.add(key1)
if group_by_level2:
key2 = self._get_key_for_group_by(row, group_by_level2, item_name)
# Track level 2 keys as well
if hasattr(self, 'group_by_keys'):
self.group_by_keys.add(key2)
if key1 not in result:
result[key1] = {}
if self.group_by_sort_key:
if key2 not in result[key1]:
result[key1][key2] = []
result_key1_key2 = result[key1][key2]
if row not in result_key1_key2:
result_key1_key2.append(row)
else:
result[key1][key2] = row
else:
if self.group_by_sort_key:
if key1 not in result:
result[key1] = []
if row not in result[key1]:
result[key1].append(row)
else:
result[key1] = row
else:
msg = "Unable to group {} by {} since the item {} is not a dictionary.".format(item_name, group_by_level1, row)
logger.error(msg)
raise ValueError(msg)
if self.group_by_sort_key:
if group_by_level2:
for key1, row1 in result.items():
for key2, row in row1.items():
row.sort(key=self.group_by_sort_key)
else:
for key, row in result.items():
row.sort(key=self.group_by_sort_key)
return result
msg = "Unable to group {} by {}".format(item_name, group_by)
logger.error(msg)
raise ValueError(msg)
def get_stats(self):
"""
Get some stats on internals of the DeepDiff run.
"""
return self._stats
@property
def affected_paths(self):
"""
Get the list of paths that were affected.
Whether a value was changed or they were added or removed.
Example
>>> from pprint import pprint
>>> t1 = {1: 1, 2: 2, 3: [3], 4: 4}
>>> t2 = {1: 1, 2: 4, 3: [3, 4], 5: 5, 6: 6}
>>> ddiff = DeepDiff(t1, t2)
>>> pprint(ddiff, indent=4)
{ 'dictionary_item_added': ['root[5]', 'root[6]'],
'dictionary_item_removed': ['root[4]'],
'iterable_item_added': {'root[3][1]': 4},
'values_changed': {'root[2]': {'new_value': 4, 'old_value': 2}}}
>>> sorted(ddiff.affected_paths)
['root[2]', 'root[3][1]', 'root[4]', 'root[5]', 'root[6]']
>>> sorted(ddiff.affected_root_keys)
[2, 3, 4, 5, 6]
"""
result = SetOrdered()
for key in REPORT_KEYS:
value = self.get(key)
if value:
if isinstance(value, SetOrdered):
result |= value
else:
result |= SetOrdered(value.keys())
return result
@property
def affected_root_keys(self):
"""
Get the list of root keys that were affected.
Whether a value was changed or they were added or removed.
Example
>>> from pprint import pprint
>>> t1 = {1: 1, 2: 2, 3: [3], 4: 4}
>>> t2 = {1: 1, 2: 4, 3: [3, 4], 5: 5, 6: 6}
>>> ddiff = DeepDiff(t1, t2)
>>> pprint(ddiff, indent=4)
{ 'dictionary_item_added': ['root[5]', 'root[6]'],
'dictionary_item_removed': ['root[4]'],
'iterable_item_added': {'root[3][1]': 4},
'values_changed': {'root[2]': {'new_value': 4, 'old_value': 2}}}
>>> sorted(ddiff.affected_paths)
['root[2]', 'root[3][1]', 'root[4]', 'root[5]', 'root[6]']
>>> sorted(ddiff.affected_root_keys)
[2, 3, 4, 5, 6]
"""
result = SetOrdered()
for key in REPORT_KEYS:
value = self.tree.get(key)
if value:
if isinstance(value, SetOrdered):
values_list = value
else:
values_list = value.keys()
for item in values_list:
root_key = item.get_root_key()
if root_key is not notpresent:
result.add(root_key)
return result
def __str__(self):
if hasattr(self, '_colored_view') and self.view in {COLORED_VIEW, COLORED_COMPACT_VIEW}:
return str(self._colored_view)
return super().__str__()
if __name__ == "__main__": # pragma: no cover
import doctest
doctest.testmod()
Reference in new issue View Git Blame Copy Permalink