#! python
# -*- coding: utf-8 -*-
#
# Copyright 2013-2019 European Commission (JRC);
# Licensed under the EUPL (the 'Licence');
# You may not use this work except in compliance with the Licence.
# You may obtain a copy of the Licence at: http://ec.europa.eu/idabc/eupl
"""
A :dfn:`pandas-model` is a tree of strings, numbers, sequences, dicts, pandas instances and resolvable
URI-references, implemented by :class:`Pandel`.
"""
import abc
import binascii
import collections.abc as cabc
import functools as fnt
import numbers
import pickle
import re
from collections import namedtuple
from json.decoder import JSONDecoder
from json.encoder import JSONEncoder
from typing import Union
from unittest.mock import MagicMock
from urllib.parse import urljoin
import jsonschema
import numpy as np
import pandas as pd
from jsonschema import ValidationError
from jsonschema.exceptions import RefResolutionError, SchemaError
from pandas.core.generic import NDFrame
__commit__ = ""
_value_with_units_regex = re.compile(
r"""^\s*
(?P<name>[^[<]*?) # column-name
\s*
(?P<units> # start parenthesized-units optional-group
\[ # units enclosed in []
[^\]]*
\]
|
< # units enclosed in <>
[^)]*
>
)? # end parenthesized-units
\s*$""",
re.X,
)
_units_cleaner_regex = re.compile(r"^[<[]|[\]>]$")
"""An item-descriptor with units, i.e. used as a table-column header."""
_U = namedtuple("United", ("name", "units"))
[docs]def parse_value_with_units(arg):
"""
Parses *name-units* pairs (i.e. used as a table-column header).
:return: a United(name, units) named-tuple, or `None` if bad syntax;
note that ``name=''`` but ``units=None`` when missing.
Examples::
>>> parse_value_with_units('value [units]')
United(name='value', units='units')
>>> parse_value_with_units('foo bar <bar/krow>')
United(name='foo bar', units='bar/krow')
>>> parse_value_with_units('no units')
United(name='no units', units=None)
>>> parse_value_with_units('')
United(name='', units=None)
But notice::
>>> assert parse_value_with_units('ok but [bad units') is None
>>> parse_value_with_units('<only units>')
United(name='', units='only units')
>>> parse_value_with_units(None) # doctest:+ELLIPSIS
Traceback (most recent call last):
TypeError: expected string or ...
"""
m = _value_with_units_regex.match(arg)
if m:
res = m.groupdict()
units = res["units"]
if units:
res["units"] = _units_cleaner_regex.sub("", units)
return _U(**res)
[docs]class ModelOperations(namedtuple("ModelOperations", "inp out conv")):
"""
Customization functions for traversing, I/O, and converting self-or-descendant branch (sub)model values.
"""
[docs] def __new__(cls, inp=None, out=None, conv=None):
"""
:param list inp: the `args-list` to :meth:`Pandel._read_branch()`
:param out: The args to :meth:`Pandel._write_branch()`, that may be specified either as:
* an `args-list`, that will apply for all model data-types (lists, dicts & pandas),
* a map of ``type`` --> ``args-list``, where the ``None`` key is the *catch-all* case,
* a function returning the `args-list` for some branch-value,
with signature: ``def get_write_branch_args(branch)``.
:param conv: The conversion-functions (:dfn:`convertors`) for the various model's data-types.
The convertors have signature ``def convert(branch)``, and they may be
specified either as:
* a map of ``(from_type, to_type)`` --> ``conversion_func()``, where the ``None`` key
is the *catch-all* case,
* a "master-switch" function returning the appropriate convertor
depending on the requested conversion.
The master-function's signature is ``def get_convertor(from_branch, to_branch)``.
The minimum convertors demanded by :class:`Pandel` are (at least, check the code for more):
* DataFrame <--> dict
* Series <--> dict
* ndarray <--> list
"""
return super(ModelOperations, cls).__new__(cls, inp, out, conv)
def choose_out_args(self, branch):
pass
def choose_convertor(self, from_type, to_type):
pass
cabc.Sequence.register(pd.Series)
cabc.Mapping.register(pd.DataFrame)
cabc.Mapping.register(pd.Series)
def _is_array(checker, instance):
return not isinstance(instance, str) and isinstance(
instance, (cabc.Sequence, np.ndarray)
)
def _is_object(checker, instance):
return isinstance(instance, cabc.Mapping)
def _is_bool(checker, instance):
return isinstance(instance, (bool, np.bool_))
def _is_integer(checker, instance):
# bool inherits from int, so ensure bools aren't reported as ints
if isinstance(instance, bool):
return False
return isinstance(instance, (int, np.integer))
def _is_null(checker, instance):
if instance is None or instance is float("nan") or instance is np.nan:
return True
## Handle np.arrays's future "bool" behavior.
# see https://github.com/numpy/numpy/issues/9583
#
try:
return instance.size == 1 and np.isnan(instance).all()
except Exception:
pass
return False
[docs]def _find_additional_properties(instance, schema):
"""
Return the set of additional properties for the given ``instance``.
Weeds out properties that should have been validated by ``properties`` and
/ or ``patternProperties``.
Assumes ``instance`` is dict-like already.
"""
properties = schema.get("properties", {})
patterns = "|".join(schema.get("patternProperties", {}))
for property in instance.keys(): # keys() wer missin, PATCHING JUST FOR THIS!!!!
if property not in properties:
if patterns and re.search(patterns, property):
continue
yield property
def _rule_additionalProperties(validator, aP, instance, schema):
if not validator.is_type(instance, "object"):
return
extras = set(_find_additional_properties(instance, schema))
if validator.is_type(aP, "object"):
for extra in extras:
for error in validator.descend(instance[extra], aP, path=extra):
yield error
elif not aP and extras:
if "patternProperties" in schema:
patterns = sorted(schema["patternProperties"])
if len(extras) == 1:
verb = "does"
else:
verb = "do"
error = "%s %s not match any of the regexes: %s" % (
", ".join(map(repr, sorted(extras))),
verb,
", ".join(map(repr, patterns)),
)
yield ValidationError(error)
else:
from jsonschema import _utils
error = "Additional properties are not allowed (%s %s unexpected)"
yield ValidationError(error % _utils.extras_msg(extras))
def _rule_propertyNames(validator, propertyNames, instance, schema):
if not validator.is_type(instance, "object"):
return
for property in instance.keys():
for error in validator.descend(instance=property, schema=propertyNames):
yield error
def _is_null_in_type(typ):
if not typ:
return False
if isinstance(typ, str):
return typ == "null"
return "null" in typ
[docs]def first_defined(*var, default=None):
"""Return the 1st non-none `var`, or `default`."""
for v in var:
if v is not None:
return v
return default
[docs]def _rule_auto_defaults_properties(
validator,
properties,
instance,
schema,
original_props_rule,
auto_default,
auto_default_nulls,
auto_remove_nulls,
):
"""
Adapted from: https://python-jsonschema.readthedocs.io/en/stable/faq/#frequently-asked-questions
"""
if not _is_object(None, instance):
return
for property, subschema in properties.items():
prop_given = property in instance
if "default" in subschema and (
(
first_defined(subschema.get("autoDefault"), auto_default)
and not prop_given
)
or (
first_defined(subschema.get("autoDefaultNull"), auto_default_nulls)
and not _is_null_in_type(subschema.get("type"))
and prop_given
and _is_null(None, instance[property])
)
):
instance[property] = subschema["default"]
elif (
first_defined(subschema.get("autoRemoveNull"), auto_remove_nulls)
and not _is_null_in_type(subschema.get("type"))
and prop_given
and _is_null(None, instance[property])
):
del instance[property]
for error in original_props_rule(validator, properties, instance, schema):
yield error
[docs]def rule_enum(validator, enums, instance, schema):
"""Overridden to evade pandas-equals after Julian/jsonschema#575 fixed bool != 0,1 (v3.0.2)."""
unbool = jsonschema._utils.unbool
if instance is 0 or instance is 1:
unbooled = unbool(instance)
if all(unbooled != unbool(each) for each in enums):
yield ValidationError("%r is not one of %r" % (instance, enums))
elif instance not in enums:
yield ValidationError("%r is not one of %r" % (instance, enums))
[docs]def PandelVisitor(
schema,
resolver=None,
format_checker=None,
auto_default: Union[bool, None] = True,
auto_default_nulls: Union[bool, None] = False,
auto_remove_nulls: Union[bool, None] = False,
):
"""
A customized jsonschema-validator suporting instance-trees with pandas and numpy objects, natively.
:param auto_default:
When the tri-state bool ``autoDefault`` in schema or this param are enabled,
it applies any schema's ``default`` value if a property is missing and
schema's ``type`` does not support `nulls`.
- Independent of `auto_default_nulls` (you may enable both).
- See meth:`_rule_auto_defaults_properties`.
:param auto_default_nulls:
When the tri-state bool ``autoDefaultNull`` in schema or this param are
it applies any schema's ``default`` value if the property is `null` and
schema's ``type`` does not support `nulls`.
- Independent of `auto_default` (you may enable both).
- Take precedence over `auto_remove_nulls`.
- See meth:`_rule_auto_defaults_properties`.
:param auto_remove_nulls:
When the tri-state bool ``autoRemoveNull`` in schema or this param are
it removes a `null` property value if the schema's ``type`` does not accept `nulls`.
- See meth:`_rule_auto_defaults_properties`.
.. ATTENTION::
If this is enabled, any `required` properties rule must FOLLOW
the `properties` rule.
Any pandas or numpy instance (for example ``obj``) is treated like that:
+----------------------------+------------------------------------------+
| Python Type | JSON Equivalence |
+----------------------------+------------------------------------------+
| :class:`pandas.DataFrame` | as ``object`` *json-type*, with: |
| | keys: ``obj.columns`` (MUST be strings) |
| | values: ``obj[col].values`` |
| | |
| | NOTE: len(df) on rows(!), not columns. |
+----------------------------+------------------------------------------+
| :class:`pandas.Series` | - as ``object`` *json-type*, with: |
| | keys: ``obj.index`` (MUST be strings) |
| | values: ``obj.values`` |
| | - as ``array`` *json-type* |
+----------------------------+------------------------------------------+
| :class:`np.ndarray` | as ``array`` *json-type* IF ndim == 1 |
+----------------------------+------------------------------------------+
| :class:`cabc.Sequence` | as ``array`` IF not string |
| | (like lists, tuples) |
+----------------------------+------------------------------------------+
Note that the value of each dataFrame column is a :``ndarray`` instances.
The simplest validations of an object or a pandas-instance is like this:
>>> import pandas as pd
>>> schema = {
... 'type': 'object',
... }
>>> pv = PandelVisitor(schema)
>>> pv.validate({'foo': 'bar'})
>>> pv.validate(pd.Series({'foo': 1}))
>>> pv.validate([1,2]) ## A sequence is invalid here.
Traceback (most recent call last):
...
jsonschema.exceptions.ValidationError: [1, 2] is not of type 'object'
<BLANKLINE>
Failed validating 'type' in schema:
{'type': 'object'}
<BLANKLINE>
On instance:
[1, 2]
Or demanding specific properties with ``required`` and no ``additionalProperties``:
>>> schema = {
... 'type': 'object',
... 'properties': {
... 'foo': {}
... },
... 'required': ['foo'],
... 'additionalProperties': False,
... }
>>> pv = PandelVisitor(schema)
>>> pv.validate(pd.Series({'foo': 1}))
>>> pv.validate(pd.Series({'foo': 1, 'bar': 2})) ## Additional 'bar' is present!
Traceback (most recent call last):
...
jsonschema.exceptions.ValidationError: Additional properties are not allowed ('bar' was unexpected)
<BLANKLINE>
Failed validating 'additionalProperties' in schema:
{'additionalProperties': False,
'properties': {'foo': {}},
'required': ['foo'],
'type': 'object'}
<BLANKLINE>
On instance:
foo 1
bar 2
dtype: int64
>>> pv.validate(pd.Series({})) ## Required 'foo' missing!
Traceback (most recent call last):
...
jsonschema.exceptions.ValidationError: 'foo' is a required property
<BLANKLINE>
Failed validating 'required' in schema:
{'additionalProperties': False,
'properties': {'foo': {}},
'required': ['foo'],
'type': 'object'}
<BLANKLINE>
On instance:
Series([], dtype: float64)
"""
validator = jsonschema.validators.validator_for(schema)
props_rule = validator.VALIDATORS["properties"]
rules = {"additionalProperties": _rule_additionalProperties}
if any((auto_default, auto_default_nulls, auto_remove_nulls)):
rules["properties"] = fnt.partial(
_rule_auto_defaults_properties,
original_props_rule=props_rule,
auto_default=auto_default,
auto_default_nulls=auto_default_nulls,
auto_remove_nulls=auto_remove_nulls,
)
if "propertyNames" in validator.VALIDATORS:
rules["propertyNames"] = _rule_propertyNames
if hasattr(jsonschema._utils, "unbool"):
rules["enum"] = rule_enum # fix pandas after jsonschema-3.0.2
ValidatorClass = jsonschema.validators.extend(
validator,
type_checker=validator.TYPE_CHECKER.redefine_many(
{
"array": _is_array,
"object": _is_object,
"integer": _is_integer,
"boolean": _is_bool,
"null": _is_null,
}
),
validators=rules,
)
return ValidatorClass(schema, resolver=resolver, format_checker=format_checker)
[docs]class Pandel(object):
"""
Builds, validates and stores a *pandas-model*, a mergeable stack of JSON-schema abiding trees of
strings and numbers, assembled with
* sequences,
* dictionaries,
* :class:`pandas.DataFrame`,
* :class:`pandas.Series`, and
* URI-references to other model-trees.
.. _pandel-overview:
**Overview**
The **making of a model** involves, among others, schema-validating, reading :dfn:`subtree-branches`
from URIs, cloning, converting and merging multiple :dfn:`sub-models` in a single :dfn:`unified-model` tree,
without side-effecting given input.
All these happen in 4+1 steps::
....................... Model Construction .................
------------ : _______ ___________ :
/ top_model /==>|Resolve|->|PreValidate|-+ :
-----------' : |___0___| |_____1_____| | :
------------ : _______ ___________ | _____ ________ ______ : --------
/ base-model/==>|Resolve|->|PreValidate|-+->|Merge|->|Validate|->|Curate|==>/ model /
-----------' : |___0___| |_____1_____| |_ 2__| |___3____| |__4+__|: -------'
............................................................
All steps are executed "lazily" using generators (with :keyword:`yield`).
Before proceeding to the next step, the previous one must have completed successfully.
That way, any ad-hoc code in building-step-5(*curation*), for instance, will not suffer a horrible death
due to badly-formed data.
[TODO] The **storing of a model** simply involves distributing model parts into different files and/or formats,
again without side-effecting the unified-model.
.. _pandel-building-model:
**Building model**
Here is a detailed description of each building-step:
1. :meth:`_resolve` and substitute any `json-references <http://tools.ietf.org/html/draft-pbryan-zyp-json-ref-03>`_
present in the submodels with content-fragments fetched from the referred URIs.
The submodels are **cloned** first, to avoid side-effecting them.
Although by default a combination of *JSON* and *CSV* files is expected, this can be customized,
either by the content in the json-ref, within the model (see below), or
as :ref:`explained <pandel-customization>` below.
The **extended json-refs syntax** supported provides for passing arguments into :meth:`_read_branch()`
and :meth:`_write_branch()` methods. The syntax is easier to explain by showing what
the default :attr:`_global_cntxt` corresponds to, for a ``DataFrame``::
{
"$ref": "http://example.com/example.json#/foo/bar",
"$inp": ["AUTO"],
"$out": ["CSV", "encoding=UTF-8"]
}
And here what is required to read and (later) store into a HDF5 local file with a predefined name::
{
"$ref": "file://./filename.hdf5",
"$inp": ["AUTO"],
"$out": ["HDF5"]
}
.. Warning:: Step NOT IMPLEMENTED YET!
2. Loosely :meth:`_prevalidate` each sub-model separately with :term:`json-schema`,
where any pandas-instances (DataFrames and Series) are left as is.
It is the duty of the developer to ensure that the prevalidation-schema is *loose enough* that
it allows for various submodel-forms, prior to merging, to pass.
3. Recursively **clone** and :meth:`_merge` sub-models in a single unified-model tree.
Branches from sub-models higher in the stack override the respective ones from the sub-models below,
recursively. Different object types need to be **converted** appropriately (ie. merging
a ``dict`` with a ``DataFrame`` results into a ``DataFrame``, so the dictionary has to convert
to dataframe).
The required **conversions** into pandas classes can be customized as :ref:`explained <pandel-customization>`
below. Series and DataFrames cannot merge together, and Sequences do not merge
with any other object-type (themselfs included), they just "overwrite".
The default convertor-functions defined both for submodels and models are listed in the following table:
============ ========== =========================================
From: To: Method:
============ ========== =========================================
dict DataFrame ``pd.DataFrame`` (the constructor)
DataFrame dict ``lambda df: df.to_dict('list')``
dict Series ``pd.Series`` (the constructor)
Series dict :meth:`lambda sr: sr.to_dict()`
============ ========== =========================================
4. Strictly json-:meth:`_validate` the unified-model (ie enforcing ``required`` schema-rules).
The required **conversions** from pandas classes can be customized as :ref:`explained <pandel-customization>`
below.
The default convertor-functions are the same as above.
5. (Optionally) Apply the :meth:`_curate` functions on the the model to enforce dependencies and/or any
ad-hoc generation-rules among the data. You can think of bash-like expansion patterns,
like ``${/some/path:=$HOME}`` or expressions like ``%len(../other/path)``.
.. _pandel-storing:
**Storing model**
When storing model-parts, if unspecified, the filenames to write into will be deduced from the jsonpointer-path
of the ``$out``'s parent, by substituting "strange" chars with undescores(``_``).
.. Warning:: Functionality NOT IMPLEMENTED YET!
.. _pandel-customization:
**Customization**
Some operations within steps (namely *conversion* and *IO*) can be customized by the following means
(from lower to higher precedance):
a. The global-default :class:`ModelOperations` instance on the :attr:`_global_cntxt`,
applied on both submodels and unified-model.
For example to channel the whole reading/writing of models through
`HDF5 <http://pandas.pydata.org/pandas-docs/stable/io.html#io-hdf5>`_ data-format, it would suffice
to modify the :attr:`_global_cntxt` like that::
pm = FooPandelModel() ## some concrete model-maker
io_args = ["HDF5"]
pm.mod_global_operations(inp=io_args, out=io_args)
b. [TODO] Extra-properties on the json-schema applied on both submodels and unified-model for the specific path defined.
The supported properties are the non-functional properties of :class:`ModelOperations`.
d. Specific-properties regarding *IO* operations within each submodel - see the *resolve* building-step,
above.
c. Context-maps of ``json_paths`` --> :class:`ModelOperations` instances, installed by :meth:`add_submodel()` and
:attr:`unified_contexts` on the model-maker. They apply to self-or-descedant subtree of each model.
The `json_path` is a strings obeying a simplified :term:`json-pointer` syntax (no char-normalizations yet),
ie ``/some/foo/1/pointer``. An empty-string(``''``) matches all model.
When multiple convertors match for a model-value, the selected convertor to be used is the most specific one
(the one with longest prefix). For instance, on the model::
[ { "foo": { "bar": 0 } } ]
all of the following would match the ``0`` value:
- the global-default :attr:`_global_cntxt`,
- ``/``, and
- ``/0/foo``
but only the last's context-props will be applied.
.. _Attributes:
**Atributes**
.. Attribute:: model
The model-tree that will receive the merged submodels after :meth:`build()` has been invoked.
Depending on the submodels, the top-value can be any of the supported model data-types.
.. Attribute:: _submodel_tuples
The stack of (``submodel``, ``path_ops``) tuples. The list's 1st element is the :dfn:`base-model`,
the last one, the :dfn:`top-model`. Use the :meth:`add_submodel()` to build this list.
.. Attribute:: _global_cntxt
A :class:`ModelOperations` instance acting as the global-default context for the unified-model and all submodels.
Use :meth:`mod_global_operations()` to modify it.
.. Attribute:: _curate_funcs
The sequence of *curate* functions to be executed as the final step by :meth:`_curate()`.
They are "normal" functions (not generators) with signature::
def curate_func(model_maker):
pass ## ie: modify ``model_maker.model``.
Better specify this list of functions on construction time.
.. Attribute:: _errored
An internal boolean flag that becomes ``True`` if any build-step has failed,
to halt proceeding to the next one. It is ``None`` if build has not started yet.
.. _pandel-examples:
**Examples**
The basic usage requires to subclass your own model-maker, just so that a *json-schema*
is provided for both validation-steps, 2 & 4:
>>> class MyModel(Pandel):
... def _get_json_schema(self, is_prevalidation):
... return { ## Define the json-schema.
... '$schema': 'http://json-schema.org/draft-04/schema#',
... 'required': [] if is_prevalidation else ['a', 'b'], ## Prevalidation is more loose.
... 'properties': {
... 'a': {'type': 'string'},
... 'b': {'type': 'number'},
... 'c': {'type': 'number'},
... }
... }
Then you can instanciate it and add your submodels:
>>> mm = MyModel()
>>> mm.add_submodel({"a": 'foo', "b": 1}) ## submodel-1 (base)
>>> mm.add_submodel(pd.Series({"a": "bar", "c": 2})) ## submodel-2 (top-model)
You then have to build the final unified-model (any validation errors would be reported at this point):
>>> mdl = mm.build()
Note that you can also access the unified-model in the :attr:`model` attribute.
You can now interrogate it:
>>> mdl['a'] == 'bar' ## Value overridden by top-model
True
>>> mdl['b'] == 1 ## Value left intact from base-model
True
>>> mdl['c'] == 2 ## New value from top-model
True
Lets try to build with invalid submodels:
>>> mm = MyModel()
>>> mm.add_submodel({'a': 1}) ## According to the schema, this should have been a string,
>>> mm.add_submodel({'b': 'string'}) ## and this one, a number.
>>> sorted(mm.build_iter(), key=lambda ex: ex.message) ## Fetch a list with all validation errors. # doctest: +NORMALIZE_WHITESPACE
[<ValidationError: "'string' is not of type 'number'">,
<ValidationError: "1 is not of type 'string'">,
<ValidationError: 'Gave-up building model after step 1.prevalidate (out of 4).'>]
>>> mdl = mm.model
>>> mdl is None ## No model constructed, failed before merging.
True
And lets try to build with valid submodels but invalid merged-one:
>>> mm = MyModel()
>>> mm.add_submodel({'a': 'a str'})
>>> mm.add_submodel({'c': 1})
>>> sorted(mm.build_iter(), key=lambda ex: ex.message) # doctest: +NORMALIZE_WHITESPACE
[<ValidationError: "'b' is a required property">,
<ValidationError: 'Gave-up building model after step 3.validate (out of 4).'>]
"""
__metaclass__ = abc.ABCMeta
[docs] def __init__(self, curate_funcs=()):
"""
:param sequence curate_funcs: See :attr:`_curate_funcs`.
"""
self.model = None
self._errored = None
self._submodel_tuples = []
self._curate_funcs = curate_funcs
self._global_cntxt = []
self._unified_contexts = None
[docs] def mod_global_operations(self, operations=None, **cntxt_kwargs):
"""
Since it is the fall-back operation for *conversions* and *IO* operation, it must exist and have
all its props well-defined for the class to work correctly.
:param ModelOperations operations: Replaces values of the installed context with
non-empty values from this one.
:param cntxt_kwargs: Replaces the keyworded-values on the existing `operations`.
See :class:`ModelOperations` for supported keywords.
"""
if operations:
assert isinstance(operations, ModelOperations), (
type(operations),
operations,
)
self._global_cntxt = operations
self._global_cntxt._replace(**cntxt_kwargs)
@property
def unified_contexts(self):
"""
A map of ``json_paths`` --> :class:`ModelOperations` instances acting on the unified-model.
"""
return self._unified_contexts
@unified_contexts.setter
def unified_contexts(self, path_ops):
assert isinstance(path_ops, cabc.Mapping), (type(path_ops), path_ops)
self._unified_contexts = path_ops
[docs] def _select_context(self, path, branch):
"""
Finds which context to use while visiting model-nodes, by enforcing the precedance-rules described
in the :ref:`Customizations <pandel-customization>`.
:param str path: the branch's jsonpointer-path
:param str branch: the actual branch's node
:return: the selected :class:`ModelOperations`
"""
pass
[docs] def _read_branch(self):
"""
Reads model-branches during *resolve* step.
"""
pass # TODO: impl read_branch()
[docs] def _write_branch(self):
"""
Writes model-branches during *distribute* step.
"""
pass # TODO: impl write_branch()
[docs] def _get_json_schema(self, is_prevalidation):
"""
:return: a json schema, more loose when `prevalidation` for each case
:rtype: dictionary
"""
# TODO: Make it a factory o
pass
def _rule_AdditionalProperties(self, validator, aP, required, instance, schema):
properties = schema.get("properties", {})
patterns = "|".join(schema.get("patternProperties", {}))
extras = set()
for prop in instance:
if prop not in properties:
if patterns and re.search(patterns, prop):
continue
extras.add(prop)
if validator.is_type(aP, "object"):
for extra in extras:
for error in validator.descend(instance[extra], aP, path=extra):
yield error
elif not aP and extras:
error = "Additional properties are not allowed (%s %s unexpected)"
yield ValidationError(error % jsonschema._utils.extras_msg(extras))
def _rule_Required(self, validator, required, instance, schema):
if (
validator.is_type(instance, "object")
or validator.is_type(instance, "DataFrame")
or validator.is_type(instance, "Series")
):
for prop in required:
if prop not in instance:
yield ValidationError("%r is a required property" % prop)
def _get_model_validator(self, schema):
return PandelVisitor(schema)
def _validate_json_model(self, schema, mdl):
validator = self._get_model_validator(schema)
for err in validator.iter_errors(mdl):
self._errored = True
yield err
[docs] def _clone_and_merge_submodels(self, a, b, path=""):
"""' Recursively merge b into a, cloning both. """
if isinstance(a, pd.DataFrame) or isinstance(b, pd.DataFrame):
a = pd.DataFrame() if a is None else pd.DataFrame(a)
b = pd.DataFrame() if b is None else pd.DataFrame(b)
a.update(b) # , 'outer') NOT IMPL YET
extra_b_items = list(set(b.columns) - set(a.columns))
a[extra_b_items] = b[extra_b_items]
elif isinstance(a, pd.Series) or isinstance(b, pd.Series):
a = pd.Series() if a is None else pd.Series(a)
b = pd.Series() if b is None else pd.Series(b)
# a.update(b) # DOES NOT append extra keys!
a = b.combine_first(a)
elif isinstance(a, cabc.Mapping) or isinstance(b, cabc.Mapping):
a = {} if a is None else {**a}
b = {} if b is None else {**b}
for key in b:
b_val = b[key]
if key in a:
val = self._clone_and_merge_submodels(
a[key], b_val, "%s/%s" % (path, key)
)
else:
val = b_val
a[key] = val
elif (isinstance(a, cabc.Sequence) and not isinstance(a, str)) or (
isinstance(b, cabc.Sequence) and not isinstance(b, str)
):
if b is not None:
val = b
else:
val = a
l = list()
for (i, item) in enumerate(val):
l.append(
self._clone_and_merge_submodels(item, None, "%s[%i]" % (path, i))
)
a = l
elif a is None and b is None:
raise ValidationError("Cannot merge Nones at path(%s)!" % path)
else:
if b is not None:
a = b
return a
[docs] def _resolve(self):
"Step-1"
if False:
yield
[docs] def _prevalidate(self):
"Step-1"
for (mdl, path_ops) in self._submodel_tuples:
schema = self._get_json_schema(is_prevalidation=True)
for err in self._validate_json_model(schema, mdl):
yield err
[docs] def _merge(self):
"Step-2"
for (mdl, path_ops) in self._submodel_tuples:
self.model = self._clone_and_merge_submodels(self.model, mdl)
if False:
yield # Just mark method as generator.
[docs] def _validate(self):
"Step-3"
schema = self._get_json_schema(is_prevalidation=False)
for err in self._validate_json_model(schema, self.model):
yield err
[docs] def _curate(self):
"Step-4: Invokes any curate-functions found in :attr:`_curate_funcs`."
if False:
yield # To be overriden, just mark method as generator.
for curfunc in self._curate_funcs:
curfunc(self)
[docs] def add_submodel(self, model, path_ops=None):
"""
Pushes on top a submodel, along with its context-map.
:param model: the model-tree (sequence, mapping, pandas-types)
:param dict path_ops: A map of ``json_paths`` --> :class:`ModelOperations` instances acting on the
unified-model. The `path_ops` may often be empty.
**Examples**
To change the default DataFrame --> dictionary convertor for a submodel, use the following:
>>> mdl = {'foo': 'bar'}
>>> submdl = ModelOperations(mdl, conv={(pd.DataFrame, dict): lambda df: df.to_dict('record')})
"""
if path_ops:
assert isinstance(path_ops, cabc.Mapping), (type(path_ops), path_ops)
return self._submodel_tuples.append((model, path_ops))
[docs] def build_iter(self):
"""
Iteratively build model, yielding any problems as :class:`ValidationError` instances.
For debugging, the unified model at :attr:`model` my contain intermediate results at any time,
even if construction has failed. Check the :attr:`_errored` flag if neccessary.
"""
steps = [
(self._prevalidate, "prevalidate"),
(self._merge, "merge"),
(self._validate, "validate"),
(self._curate, "curate"),
]
self._errored = False
self.model = None
for (i, (step, step_name)) in enumerate(steps, start=1):
try:
for err in step():
yield err
except ValidationError as ex:
self._errored = True
yield ex
except Exception as ex:
self._errored = True
nex = ValidationError(
"Model step-%i(%s) failed due to: %s" % (i, step_name, ex)
)
nex.cause = ex
yield nex
if self._errored:
yield ValidationError(
"Gave-up building model after step %i.%s (out of %i)."
% (i, step_name, len(steps))
)
break
[docs] def build(self):
"""
Attempts to build the model by exhausting :meth:`build_iter()`, or raises its 1st error.
Use this method when you do not want to waste time getting the full list of errors.
"""
err = next(self.build_iter(), None)
if err:
raise err
return self.model
def get(self, path, **kws):
resolve_jsonpointer(self.model, path, **kws)
def escape_jsonpointer_part(part):
return part.replace("~", "~0").replace("/", "~1")
def unescape_jsonpointer_part(part):
return part.replace("~1", "/").replace("~0", "~")
[docs]def iter_jsonpointer_parts(jsonpath):
"""
Generates the ``jsonpath`` parts according to jsonpointer spec.
:param str jsonpath: a jsonpath to resolve within document
:return: The parts of the path as generator), without
converting any step to int, and None if None.
:author: Julian Berman, ankostis
Examples::
>>> list(iter_jsonpointer_parts('/a/b'))
['a', 'b']
>>> list(iter_jsonpointer_parts('/a//b'))
['a', '', 'b']
>>> list(iter_jsonpointer_parts('/'))
['']
>>> list(iter_jsonpointer_parts(''))
[]
But paths are strings begining (NOT_MPL: but not ending) with slash('/')::
>>> list(iter_jsonpointer_parts(None))
Traceback (most recent call last):
AttributeError: 'NoneType' object has no attribute 'split'
>>> list(iter_jsonpointer_parts('a'))
Traceback (most recent call last):
jsonschema.exceptions.RefResolutionError: Jsonpointer-path(a) must start with '/'!
#>>> list(iter_jsonpointer_parts('/a/'))
#Traceback (most recent call last):
#jsonschema.exceptions.RefResolutionError: Jsonpointer-path(a) must NOT ends with '/'!
"""
# if jsonpath.endswith('/'):
# msg = "Jsonpointer-path({}) must NOT finish with '/'!"
# raise RefResolutionError(msg.format(jsonpath))
parts = jsonpath.split("/")
if parts.pop(0) != "":
msg = "Jsonpointer-path({}) must start with '/'!"
raise RefResolutionError(msg.format(jsonpath))
for part in parts:
part = unescape_jsonpointer_part(part)
yield part
[docs]def iter_jsonpointer_parts_relaxed(jsonpointer):
"""
Like :func:`iter_jsonpointer_parts()` but accepting also non-absolute paths.
The 1st step of absolute-paths is always ''.
Examples::
>>> list(iter_jsonpointer_parts_relaxed('a'))
['a']
>>> list(iter_jsonpointer_parts_relaxed('a/'))
['a', '']
>>> list(iter_jsonpointer_parts_relaxed('a/b'))
['a', 'b']
>>> list(iter_jsonpointer_parts_relaxed('/a'))
['', 'a']
>>> list(iter_jsonpointer_parts_relaxed('/a/'))
['', 'a', '']
>>> list(iter_jsonpointer_parts_relaxed('/'))
['', '']
>>> list(iter_jsonpointer_parts_relaxed(''))
['']
"""
for part in jsonpointer.split("/"):
yield unescape_jsonpointer_part(part)
_scream = object()
[docs]def resolve_jsonpointer(doc, jsonpointer, default=_scream):
"""
Resolve a ``jsonpointer`` within the referenced ``doc``.
:param doc: the referrant document
:param str path: a jsonpointer to resolve within document
:param default: A value to return if path does not resolve.
:return: the resolved doc-item or raises :class:`RefResolutionError`
:raises: RefResolutionError (if cannot resolve path and no `default`)
Examples:
>>> dt = {
... 'pi':3.14,
... 'foo':'bar',
... 'df': pd.DataFrame(np.ones((3,2)), columns=list('VN')),
... 'sub': {
... 'sr': pd.Series({'abc':'def'}),
... }
... }
>>> resolve_jsonpointer(dt, '/pi', default=_scream)
3.14
>>> resolve_jsonpointer(dt, '/pi/BAD')
Traceback (most recent call last):
jsonschema.exceptions.RefResolutionError: Unresolvable JSON pointer('/pi/BAD')@(BAD)
>>> resolve_jsonpointer(dt, '/pi/BAD', 'Hi!')
'Hi!'
:author: Julian Berman, ankostis
"""
for part in iter_jsonpointer_parts(jsonpointer):
if isinstance(doc, cabc.Sequence):
# Array indexes should be turned into integers
try:
part = int(part)
except ValueError:
pass
try:
doc = doc[part]
except (TypeError, LookupError):
if default is _scream:
raise RefResolutionError(
"Unresolvable JSON pointer(%r)@(%s)" % (jsonpointer, part)
)
else:
return default
return doc
[docs]def resolve_path(doc, path, default=_scream, root=None):
"""
Like :func:`resolve_jsonpointer` also for relative-paths & attribute-branches.
:param doc: the referrant document
:param str path: An abdolute or relative path to resolve within document.
:param default: A value to return if path does not resolve.
:param root: Document for absolute paths, assumed `doc` if missing.
:return: the resolved doc-item or raises :class:`RefResolutionError`
:raises: RefResolutionError (if cannot resolve path and no `default`)
Examples:
>>> dt = {
... 'pi':3.14,
... 'foo':'bar',
... 'df': pd.DataFrame(np.ones((3,2)), columns=list('VN')),
... 'sub': {
... 'sr': pd.Series({'abc':'def'}),
... }
... }
>>> resolve_path(dt, '/pi', default=_scream)
3.14
>>> resolve_path(dt, 'df/V')
0 1.0
1 1.0
2 1.0
Name: V, dtype: float64
>>> resolve_path(dt, '/pi/BAD', 'Hi!')
'Hi!'
:author: Julian Berman, ankostis
"""
def resolve_root(d, p):
if not p:
return root or doc
raise ValueError()
part_resolvers = [
lambda d, p: d[int(p)],
lambda d, p: d[p],
lambda d, p: getattr(d, p),
resolve_root,
]
for part in iter_jsonpointer_parts_relaxed(path):
start_i = 0 if isinstance(doc, cabc.Sequence) else 1
for resolver in part_resolvers[start_i:]:
try:
doc = resolver(doc, part)
break
except (ValueError, TypeError, LookupError, AttributeError):
pass
else:
if default is _scream:
raise RefResolutionError("Unresolvable path(%r)@(%s)" % (path, part))
return default
return doc
[docs]def set_jsonpointer(doc, jsonpointer, value, object_factory=dict):
"""
Resolve a ``jsonpointer`` within the referenced ``doc``.
:param doc: the referrant document
:param str jsonpointer: a jsonpointer to the node to modify
:raises: RefResolutionError (if jsonpointer empty, missing, invalid-contet)
"""
parts = list(iter_jsonpointer_parts(jsonpointer))
# Will scream if used on 1st iteration.
#
pdoc = None
ppart = None
for i, part in enumerate(parts):
if isinstance(doc, cabc.Sequence) and not isinstance(doc, str):
# Array indexes should be turned into integers
#
doclen = len(doc)
if part == "-":
part = doclen
else:
try:
part = int(part)
except ValueError:
raise RefResolutionError(
"Expected numeric index(%s) for sequence at (%r)[%i]"
% (part, jsonpointer, i)
)
else:
if part > doclen:
raise RefResolutionError(
"Index(%s) out of bounds(%i) of (%r)[%i]"
% (part, doclen, jsonpointer, i)
)
try:
ndoc = doc[part]
except (LookupError):
break # Branch-extension needed.
except (TypeError): # Maybe indexing a string...
ndoc = object_factory()
pdoc[ppart] = ndoc
doc = ndoc
break # Branch-extension needed.
doc, pdoc, ppart = ndoc, doc, part
else:
doc = pdoc # If loop exhausted, cancel last assignment.
# Build branch with value-leaf.
#
nbranch = value
for part2 in reversed(parts[i + 1 :]):
ndoc = object_factory()
ndoc[part2] = nbranch
nbranch = ndoc
# Attach new-branch.
try:
doc[part] = nbranch
# Inserting last sequence-element raises IndexError("list assignment index
# out of range")
except IndexError:
doc.append(nbranch)
# except (IndexError, TypeError) as ex:
# if isinstance(ex, IndexError) or 'list indices must be integers' in str(ex):
# raise RefResolutionError("Incompatible content of JSON pointer(%r)@(%s)" % (jsonpointer, part))
# else:
# doc = {}
# parent_doc[parent_part] = doc
# doc[part] = value
def build_all_jsonpaths(schema):
# Totally quick an dirty, TODO: Use json-validator to build all json-paths.
forks = ["oneOf", "anyOf", "allOf"]
def _visit(schema, path, paths):
for f in forks:
objlist = schema.get(f)
if objlist:
for obj in objlist:
_visit(obj, path, paths)
props = schema.get("properties")
if props:
for p, obj in props.items():
_visit(obj, path + "/" + p, paths)
else:
paths.append(path)
paths = []
_visit(schema, "", paths)
return paths
_NONE = object()
"""Denotes non-existent json-schema attribute in :class:`JSchema`."""
[docs]class JSchema(object):
"""
Facilitates the construction of json-schema-v4 nodes on :class:`PStep` code.
It does just rudimentary args-name check. Further validations
should apply using a proper json-schema validator.
:param type: if omitted, derived as 'object' if it has children
:param kws: for all the rest see http://json-schema.org/latest/json-schema-validation.html
"""
type = (_NONE,) # @ReservedAssignment
items = (_NONE,) # @ReservedAssignment
required = (_NONE,)
title = (_NONE,)
description = (_NONE,)
minimum = (_NONE,)
exclusiveMinimum = (_NONE,)
maximum = (_NONE,)
exclusiveMaximum = (_NONE,)
patternProperties = (_NONE,)
pattern = (_NONE,)
enum = (_NONE,)
allOf = (_NONE,)
oneOf = (_NONE,)
anyOf = (_NONE,)
def todict(self):
return {k: v for k, v in vars(self).items() if v is not _NONE}
[docs]class JSONCodec:
"""
Json coders/decoders capable for (almost) all python objects, by pickling them.
Example::
>>> import json
>>> obj_list = [
... 3.14,
... {
... 'aa': pd.DataFrame([]),
... 2: np.array([]),
... 33: {'foo': 'bar'},
... },
... pd.DataFrame(np.random.randn(10, 2)),
... ('b', pd.Series({})),
... ]
>>> for o in obj_list + [obj_list]:
... s = json.dumps(o, cls=JSONCodec.Encoder)
... oo = json.loads(s, cls=JSONCodec.Decoder)
... #assert trees_equal(o, oo)
...
.. seealso::
For pickle-limitations: https://docs.python.org/3.7/library/pickle.html#pickle-picklable
"""
_ver_key = "_ver"
_ver = "0"
_obj = "$qpickle"
[docs] class Encoder(JSONEncoder):
[docs] def encode(self, o):
pickle_bytes = pickle.dumps(o)
pickle_str = binascii.b2a_qp(pickle_bytes).decode(encoding="utf8")
o = {JSONCodec._obj: pickle_str, JSONCodec._ver_key: JSONCodec._ver}
return JSONEncoder.encode(self, o)
[docs] class Decoder(JSONDecoder):
[docs] def decode(self, s):
o = JSONDecoder.decode(self, s)
pickle_str = o.get(JSONCodec._obj, None)
if pickle_str:
# file_ver = o[JSONCodec._ver_key]
# if file_ver != JSONCodec._ver:
# msg = 'Unsopported json-encoded version(%s != %s)!'
# raise ValueError(msg % (file_ver, JSONCodec._ver))
pickle_bytes = binascii.a2b_qp(pickle_str.encode(encoding="utf8"))
o = pickle.loads(pickle_bytes)
return o
if __name__ == "__main__": # pragma: no cover
raise NotImplementedError