metainfo.py

# Copyright 2018 Markus Scheidgen
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an"AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from typing import Type, TypeVar, Union, Tuple, Iterable, List, Any, Dict, Set, \
    Callable as TypingCallable, cast
from collections.abc import Iterable as IterableABC, Sequence
import sys
import inspect
import re
import json
import itertools

import numpy as np
import pint
import pint.unit
import pint.quantity
import aniso8601
from datetime import datetime
import pytz
import docstring_parser
import jmespath

from nomad.units import ureg as units


m_package: 'Package' = None

is_bootstrapping = True
MSectionBound = TypeVar('MSectionBound', bound='MSection')
T = TypeVar('T')

# Make pylint believe all bootstrap quantities are actual properties even though
# we have to initialize them to None due to bootstrapping
_placeholder_quantity: 'Quantity' = property()  # type: ignore
if True:
    _placeholder_quantity: 'Quantity' = None  # type: ignore

_primitive_types = {str: str, int: int, float: float, bool: bool}


# Metainfo errors

class MetainfoError(Exception):
    ''' Metainfo related errors. '''
    pass


class DeriveError(MetainfoError):
    ''' An error occurred while computing a derived value. '''
    pass


class MetainfoReferenceError(MetainfoError):
    ''' An error indicating that a reference could not be resolved. '''
    pass


# Metainfo quantity data types

class MEnum(Sequence):
    '''Allows to define str types with values limited to a pre-set list of possible values.'''
    def __init__(self, *args, **kwargs):
        # Supports one big list in place of args
        if len(args) == 1 and isinstance(args[0], list):
            args = args[0]

        # If non-named arguments are given, the default is to have them placed
        # into a dictionary with their string value as both the enum name and
        # the value.
        for arg in args:
            if arg in kwargs:
                raise ValueError("Duplicate value '{}' provided for enum".format(arg))
            kwargs[arg] = arg

        self._list = list(kwargs.values())
        self._values = set(kwargs.values())  # For allowing constant time member check
        self._map = kwargs

    def __getattr__(self, attr):
        return self._map[attr]

    def __getitem__(self, index):
        return self._list[index]

    def __len__(self):
        return len(self._list)


class MProxy():
    ''' A placeholder object that acts as reference to a value that is not yet resolved.

    Attributes:
        url: The reference represented as an URL string.
    '''

    def __init__(
            self, m_proxy_value: Union[str, int, dict], m_proxy_section: 'MSection' = None,
            m_proxy_quantity: 'Quantity' = None):
        self.m_proxy_value = m_proxy_value
        self.m_proxy_section = m_proxy_section
        self.m_proxy_resolved = None
        self.m_proxy_quantity = m_proxy_quantity

    def m_proxy_resolve(self):
        if self.m_proxy_section and self.m_proxy_quantity and not self.m_proxy_resolved:
            self.m_proxy_resolved = self.m_proxy_quantity.type.resolve(self)

        if self.m_proxy_resolved is not None and isinstance(self, MProxy):
            setattr(self, '__class__', self.m_proxy_resolved.__class__)
            self.__dict__.update(**self.m_proxy_resolved.__dict__)

        return self.m_proxy_resolved

    def __getattr__(self, key):
        if self.m_proxy_resolve() is not None:
            return getattr(self.m_proxy_resolved, key)

        raise ReferenceError('could not resolve %s' % self.m_proxy_value)


class SectionProxy(MProxy):
    def m_proxy_resolve(self):
        if self.m_proxy_section and not self.m_proxy_resolved:
            root = self.m_proxy_section
            while root.m_parent is not None and not isinstance(root, Package):
                root = root.m_parent

            if isinstance(root, Package):
                self.m_proxy_resolved = root.all_definitions.get(self.m_proxy_value)

            if self.m_proxy_resolved is None:
                raise ReferenceError('could not resolve %s' % self.m_proxy_value)

        return self.m_proxy_resolved


class DataType:
    '''
    Allows to define custom data types that can be used in the meta-info.

    The metainfo supports the most types out of the box. These includes the python build-in
    primitive types (int, bool, str, float, ...), references to sections, and enums.
    However, in some occasions you need to add custom data types.

    This base class lets you customize various aspects of value treatment. This includes
    type checks and various value transformations. This allows to store values in the
    section differently from how the usermight set/get them, and it allows to have non
    serializeable values that are transformed on de-/serialization.
    '''
    def set_normalize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        ''' Transforms the given value before it is set and checks its type. '''
        return value

    def get_normalize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        ''' Transforms the given value when it is get. '''
        return value

    def serialize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        ''' Transforms the given value when making the section serializeable. '''
        return value

    def deserialize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        ''' Transforms the given value from its serializeable form. '''
        return value


range_re = re.compile(r'(\d)\.\.(\d|\*)')


class _Dimension(DataType):

    def set_normalize(self, section, quantity_def: 'Quantity', value):
        if isinstance(value, int):
            return value

        if isinstance(value, str):
            if value.isidentifier():
                return value
            if re.match(range_re, value):
                return value

        if isinstance(value, Section):
            return value

        if isinstance(value, type) and hasattr(value, 'm_def'):
            return value

        if isinstance(value, str):
            # TODO raise a warning or allow this?
            # In the old metainfo there are cases where an expression is used
            # that is later evaluated in the parser
            return value

        raise TypeError('%s is not a valid dimension' % str(value))

    @staticmethod
    def check_dimension(section, dimension, length):
        if isinstance(dimension, int):
            return dimension == length
        if isinstance(dimension, str):
            if dimension.isidentifier():
                return dimension == getattr(section, dimension)

            m = re.match(range_re, dimension)
            start = int(m.group(1))
            end = -1 if m.group(2) == '*' else int(m.group(2))
            return start <= length and (end == -1 or length <= end)


class _Unit(DataType):
    def set_normalize(self, section, quantity_def: 'Quantity', value):
        if isinstance(value, str):
            value = units.parse_units(value)

        elif not isinstance(value, pint.unit._Unit):
            raise TypeError('Units must be given as str or pint Unit instances.')

        return value

    def serialize(self, section, quantity_def: 'Quantity', value):
        return value.__str__()

    def deserialize(self, section, quantity_def: 'Quantity', value):
        return units.parse_units(value)


class _Callable(DataType):
    def serialize(self, section, quantity_def: 'Quantity', value):
        raise MetainfoError('Callables cannot be serialized')

    def deserialize(self, section, quantity_def: 'Quantity', value):
        raise MetainfoError('Callables cannot be serialized')


class _QuantityType(DataType):
    ''' Data type for defining the type of a metainfo quantity.

    A metainfo quantity type can be one of

    - python build-in primitives: int, float, bool, str
    - numpy dtypes, e.g. f, int32
    - a section definition to define references
    - an MEnum instance to use it's values as possible str values
    - a custom datatype, i.e. instance of :class:`DataType`
    - Any
    '''

    def set_normalize(self, section, quantity_def, value):
        if value in [str, int, float, bool]:
            return value

        if isinstance(value, MEnum):
            for enum_value in value._values:
                if not isinstance(enum_value, str):
                    raise TypeError('MEnum value %s is not a string.' % enum_value)
            return value

        if type(value) == np.dtype:
            return value

        if isinstance(value, Section):
            return value

        if isinstance(value, Reference) and isinstance(value.target_section_def, MProxy):
            proxy = value.target_section_def
            proxy.m_proxy_section = section
            proxy.m_proxy_quantity = quantity_def
            return value

        if isinstance(value, DataType):
            return value

        if value == Any:
            return value

        if isinstance(value, type):
            section = getattr(value, 'm_def', None)
            if section is not None:
                return Reference(section)

        if isinstance(value, MProxy):
            value.m_proxy_section = section
            value.m_proxy_quantity = quantity_def
            return value

        raise MetainfoError(
            'Type %s of %s is not a valid metainfo quantity type' %
            (value, quantity_def))

    def serialize(self, section, quantity_def, value):
        if value is str or value is int or value is float or value is bool:
            return dict(type_kind='python', type_data=value.__name__)

        if isinstance(value, MEnum):
            return dict(type_kind='Enum', type_data=list(value))

        if type(value) == np.dtype:
            return dict(type_kind='numpy', type_data=str(value))

        if isinstance(value, Reference):
            return dict(type_kind='reference', type_data=value.target_section_def.m_path())

        if isinstance(value, DataType):
            module = value.__class__.__module__
            if module is None or module == str.__class__.__module__:
                type_data = value.__class__.__name__
            else:
                type_data = '%s.%s' % (module, value.__class__.__name__)

            return dict(type_kind='custom', type_data=type_data)

        if value == Any:
            return dict(type_kind='Any')

        raise MetainfoError(
            'Type %s of %s is not a valid metainfo quantity type' %
            (value, quantity_def))


class Reference(DataType):
    ''' Datatype used for reference quantities. '''

    def __init__(self, section_def: Union['Section', 'SectionProxy']):
        self.target_section_def = section_def

    def resolve(self, proxy) -> 'MSection':
        '''
        Resolve the given proxy. The proxy is guaranteed to have a context and
        will to be not yet resolved.
        '''
        return proxy.m_proxy_section.m_resolve(proxy.m_proxy_value)

    def set_normalize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        if isinstance(self.target_section_def, MProxy):
            proxy = self.target_section_def
            proxy.m_proxy_section = section.m_def
            proxy.m_proxy_quantity = quantity_def
            self.target_section_def = proxy.m_proxy_resolve()

        if self.target_section_def.m_follows(Definition.m_def):
            # special case used in metainfo definitions, where we reference metainfo definitions
            # using their Python class. E.g. referencing a section definition using its
            # class instead of the object: Run vs. Run.m_def
            if isinstance(value, type):
                definition = getattr(value, 'm_def', None)
                if definition is not None and definition.m_follows(self.target_section_def):
                    return definition

        if isinstance(value, (str, int, dict)):
            return MProxy(value, m_proxy_section=section, m_proxy_quantity=quantity_def)

        if isinstance(value, MProxy):
            value.m_proxy_section = section
            value.m_proxy_quantity = quantity_def
            return value

        if not isinstance(value, MSection):
            raise TypeError(
                'The value %s is not a section and can not be used as a reference.' % value)

        if not value.m_follows(self.target_section_def):  # type: ignore
            raise TypeError(
                '%s is not a %s and therefore an invalid value of %s.' %
                (value, self.target_section_def, quantity_def))

        return value

    def serialize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        return value.m_path()

    def deserialize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        return MProxy(value, m_proxy_section=section, m_proxy_quantity=quantity_def)


class _Datetime(DataType):

    def _parse(self, datetime_str: str) -> datetime:
        try:
            return aniso8601.parse_datetime(datetime_str)
        except ValueError:
            pass

        try:
            return aniso8601.parse_date(datetime_str)
        except ValueError:
            pass

        try:
            # TODO necessary?
            import flask_restplus.inputs
            return flask_restplus.inputs.datetime_from_rfc822(datetime_str)
        except ValueError:
            pass

        raise TypeError('Invalid date literal "{0}"'.format(datetime_str))

    def _convert(self, value):
        if value is None:
            return None

        if isinstance(value, str):
            value = self._parse(value)

        elif isinstance(value, (int, float)):
            value = datetime.fromtimestamp(value)

        elif isinstance(value, pint.Quantity):
            value = datetime.fromtimestamp(value.magnitude)

        if not isinstance(value, datetime):
            raise TypeError('%s is not a datetime.' % value)

        return value

    def set_normalize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        return self._convert(value)

    def serialize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        if value is None:
            return None

        value.replace(tzinfo=pytz.utc)
        return value.isoformat()

    def deserialize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        return self._convert(value)


class _JSON(DataType):
    pass


class _Capitalized(DataType):
    def set_normalize(self, section: 'MSection', quantity_def: 'Quantity', value: Any) -> Any:
        if value is not None and len(value) >= 1:
            return value[0].capitalize() + value[1:]

        return value


Dimension = _Dimension()
Unit = _Unit()
QuantityType = _QuantityType()
Callable = _Callable()
Datetime = _Datetime()
JSON = _JSON()
Capitalized = _Capitalized()


# Metainfo data storage and reflection interface

class MObjectMeta(type):

    def __new__(self, cls_name, bases, dct):
        do_init = dct.get('do_init', None)
        if do_init is not None:
            del(dct['do_init'])
        else:
            do_init = True

        cls = super().__new__(self, cls_name, bases, dct)

        init = getattr(cls, '__init_cls__')
        if init is not None and do_init and not is_bootstrapping:
            init()
        return cls


SectionDef = Union[str, 'Section', 'SubSection', Type[MSectionBound]]
''' Type for section definition references.

This can either be :

- the name of the section
- the section definition itself
- the definition of a sub section
- or the section definition Python class
'''


def constraint(warning):
    ''' A decorator for methods implementing constraints. '''
    f = None
    if not isinstance(warning, bool):
        f = warning
        warning = False

    def decorator(f):
        setattr(f, 'm_constraint', True)
        setattr(f, 'm_warning', warning)
        return f

    if f is None:
        return decorator
    else:
        return decorator(f)


class MResource():
    '''
    Represents a collection of related metainfo data, i.e. a set of :class:`MSection` instances.
    '''

    def __init__(self, logger=None):
        self.__data: Dict['Section', List['MSection']] = dict()
        self.contents: List['MSection'] = []
        self.logger = logger

    def create(self, section_cls: Type[MSectionBound], *args, **kwargs) -> MSectionBound:
        '''
        Create an instance of the given section class and adds it to this resource as
        a root section. The m_parent_index will be set sequentially among root sections of
        the same section definition starting with 0.
        '''
        index = 0
        for content in self.contents:
            if content.m_follows(section_cls.m_def):
                index = max(index, content.m_parent_index + 1)

        result = section_cls(*args, **kwargs)
        result.m_parent_index = index

        self.add(result)
        return cast(MSectionBound, result)

    def add(self, section):
        '''
        Add the given section to this resource. Will also add all its contents to the
        resource and make all contest available for :func:`all`. Will also remove
        all contents from possible other resources. A section can only be contained in
        one resource at a time.

        This is potentially expensive. Do not add a section that already has a deep tree
        of sub-sections. Ideally, add the root section first. If you create sub sections
        afterwards, they will be automatically added to this resource.
        '''
        if section.m_resource is not None:
            section.m_resource.remove(section)

        for content in section.m_all_contents(include_self=True):
            content.m_resource = self
            self.__data.setdefault(content.m_def, []).append(content)

        if section.m_parent is None:
            self.contents.append(section)

    def remove(self, section):
        assert section.m_resource == self, 'Can only remove section from the resource that contains it.'
        section.m_resource = None
        self.__data.get(section.m_def).remove(section)
        if section.m_parent is None:
            self.contents.remove(section)

    def all(self, section_cls: Type[MSectionBound]) -> List[MSectionBound]:
        ''' Returns all instances of the given section class in this resource. '''
        return cast(List[MSectionBound], self.__data.get(section_cls.m_def, []))

    def unload(self):
        ''' Breaks all references among the contain metainfo sections to allow GC. '''
        for collections in self.__data.values():
            for section in collections:
                section.m_parent = None
                section.__dict__.clear()
            collections.clear()

    def m_to_dict(self, filter: TypingCallable[['MSection'], bool] = None):
        if filter is None:
            def filter(_):  # pylint: disable=function-redefined
                return True

        return {
            section.m_def.name: section.m_to_dict()
            for section in self.contents
            if filter(section)}

    def warning(self, *args, **kwargs):
        if self.logger is not None:
            self.logger.warn(*args, **kwargs)


class MSection(metaclass=MObjectMeta):  # TODO find a way to make this a subclass of collections.abs.Mapping
    '''
    The base-class for all *section defining classes* and respectively the base-class
    for all section objects.

    While we use *section classes* to *define sections*, it is important to note that
    the *section class* is a different Python object than the actual *section definition*.
    For each *section class* (a Python class), we automatically generate a *section definition*
    Python object that instantiates :class:`Section`. :class:`MSection` and :class:`Section`
    are completely different classes. :class:`MSection` is used as a base-class for all
    *section defining* classes and :class:`Section` is a *section class* that defines the
    section `Section`.

    Attributes:
        m_def: Each *section class* (and also *section instance*) has a build-in
            property ``m_def`` that refers to the actual *section definition*. While this defined
            automatically, you can do it manually to provide additional characteristics that cannot
            be covered in a Python class definition.

    All `section instances` indirectly instantiate the :class:`MSection` and therefore all
    members of :class:`MSection` are available on all `section instances`. :class:`MSection`
    provides many special attributes and functions (they all start with ``m_``) that allow
    to reflect on a `section's definition` and allow to manipulate the `section instance`
    without a priori knowledge of the `section defintion`.

    .. automethod:: m_set
    .. automethod:: m_get
    .. automethod:: m_add_values
    .. automethod:: m_get_sub_section
    .. automethod:: m_get_sub_sections
    .. automethod:: m_create
    .. automethod:: m_add_sub_section
    .. automethod:: m_remove_subsection

    There are some specific attributes for section instances that are sub-sections of
    another section. While sub-sections are directly accessible from the containing
    section by using the Python property that represents the sub-section (e.g.
    `run.section_system`), there is also a way to navigate from the sub-section to
    the containing section (`parent section`) using these Python properties:

    Attributes:
        m_parent:
            If this section is a sub-section, this references the parent section instance.

        m_parent_sub_section:
            If this section is a sub-section, this is the :class:`SubSection` that defines
            this relationship.

        m_parent_index:
            For repeatable sections, parent keep a list of sub-sections. This is the index
            of this section in the respective parent sub-section list.

        m_resource: The :class:`MResource` that contains and manages this section.

    Often some general tasks have to be performed on a whole tree of sections without
    knowing about the definitions in advance. The following methods allow to access
    sub-sections reflectively.

    .. automethod:: m_traverse
    .. automethod:: m_all_contents
    .. automethod:: m_contents
    .. automethod:: m_xpath

    Each section and all its quantities and contents can be transformed into a general
    JSON-serializable Python dictionary. Similarely, a section can be instantiated from
    such a Python dictionary. This allows to save and load sections to JSON-files or
    by other compatible means (e.g. document databases, binary JSON flavours).

    .. automethod:: m_to_dict
    .. automethod:: m_from_dict
    .. automethod:: m_update_from_dict
    .. automethod:: m_to_json
    '''

    m_def: 'Section' = None

    def __init__(
            self, m_def: 'Section' = None, m_resource: MResource = None, **kwargs):

        self.m_def: 'Section' = m_def
        self.m_parent: 'MSection' = None
        self.m_parent_sub_section: 'SubSection' = None
        self.m_parent_index = -1
        self.m_resource = m_resource
        self.m_mod_count = 0
        self.m_cache: dict = {}  # Dictionary for caching temporary values that are not persisted to the Archive

        # get missing m_def from class
        cls = self.__class__
        if self.m_def is None:
            self.m_def = cls.m_def

        # check m_def
        if cls.m_def is not None:
            if self.m_def != cls.m_def:
                MetainfoError('Section class and section definition must match.')

            if self.m_def.extends_base_section:
                MetainfoError('Section extends another section and cannot be instantiated.')

        else:
            if not is_bootstrapping:
                MetainfoError('Section has not m_def.')

        # get annotations from kwargs
        self.m_annotations: Dict[str, Any] = {}
        other_kwargs = {}
        for key, value in kwargs.items():
            if key.startswith('a_'):
                self.m_annotations[key[2:]] = value
            else:
                other_kwargs[key] = value

        # get additional annotations from the section definition
        if not is_bootstrapping:
            for section_annotation in self.m_def.m_get_annotations(SectionAnnotation, as_list=True):
                for name, annotation in section_annotation.new(self).items():
                    self.m_annotations[name] = annotation

        # add annotation attributes for names annotations
        for annotation_name, annotation in self.m_annotations.items():
            setattr(self, 'a_%s' % annotation_name, annotation)

        # set remaining kwargs
        if is_bootstrapping:
            self.__dict__.update(**other_kwargs)  # type: ignore
        else:
            self.m_update(**other_kwargs)

    @classmethod
    def __init_cls__(cls):
        # ensure that the m_def is defined
        m_def = cls.__dict__.get('m_def')  # do not accedentally get the m_def from a potential base section
        if m_def is None:
            m_def = Section()
            setattr(cls, 'm_def', m_def)

        # Use class name if name is not explicitly defined
        if m_def.name is None:
            m_def.name = cls.__name__
        m_def._section_cls = cls

        # add base sections
        base_sections: List[Section] = []
        for base_cls in cls.__bases__:
            if base_cls != MSection:
                base_section = getattr(base_cls, 'm_def')
                if base_section is None:
                    raise TypeError(
                        'Section defining classes must have MSection or a decendant as '
                        'base classes.')
                base_sections.append(base_section)

        m_def.m_set(Section.base_sections, base_sections)

        # transfer names, descriptions, constraints, event_handlers
        constraints: Set[str] = set()
        event_handlers: Set[Callable] = set(m_def.event_handlers)
        for name, attr in cls.__dict__.items():
            # transfer names and descriptions for properties, init properties
            if isinstance(attr, Property):
                attr.name = name
                if attr.description is not None:
                    attr.description = inspect.cleandoc(attr.description).strip()
                    attr.__doc__ = attr.description

                if isinstance(attr, Quantity):
                    m_def.m_add_sub_section(Section.quantities, attr)
                elif isinstance(attr, SubSection):
                    m_def.m_add_sub_section(Section.sub_sections, attr)
                else:
                    raise NotImplementedError('Unknown property kind.')

            if inspect.isfunction(attr):
                method_name = attr.__name__

                # transfer constraints
                if getattr(attr, 'm_constraint', False):
                    constraint = method_name
                    constraints.add(constraint)

                # register event_handlers from event_handler methods
                if method_name.startswith('on_set') or method_name.startswith('on_add_sub_section'):
                    if attr not in event_handlers:
                        event_handlers.add(attr)

        # add handler and constraints from base sections
        for base_section in m_def.all_base_sections:
            for constraint in base_section.constraints:
                constraints.add(constraint)
            for event_handler in base_section.event_handlers:
                event_handlers.add(event_handler)

        m_def.constraints = list(constraints)
        m_def.event_handlers = list(event_handlers)

        # add section cls' section to the module's package
        module_name = cls.__module__
        pkg = Package.from_module(module_name)
        pkg.m_add_sub_section(Package.section_definitions, cls.m_def)

        # apply_google_docstrings
        # Parses the google doc string of the given class and properly updates the
        # definition descriptions.

        # This allows to document quantities and sub-sections with 'Args:' in the section
        # class. It will remove the 'Args' section from the section definition and will
        # set the respective pieces to the quantity and sub-section descriptions.
        docstring = cls.__doc__
        if docstring is not None:
            parsed_docstring = docstring_parser.parse(docstring)
            short = parsed_docstring.short_description
            dsc = parsed_docstring.long_description

            if short and dsc:
                description = '%s %s' % (short.strip(), dsc.strip())
            elif short:
                description = short.strip()
            elif dsc:
                description = dsc.strip()
            else:
                description = None

            if m_def.description is None:
                m_def.description = description

            for param in parsed_docstring.params:
                prop = m_def.all_properties.get(param.arg_name)
                if prop is not None:
                    if prop.description is None:
                        prop.description = param.description

        for content in m_def.m_all_contents(depth_first=True, include_self=True):
            cast(Definition, content).__init_metainfo__()

    def __setattr__(self, name, value):
        all_aliases = None
        if self.m_def is not None:
            all_aliases = self.m_def.all_aliases

        if all_aliases is not None and name in self.m_def.all_aliases:
            name = self.m_def.all_aliases[name].name

        return super().__setattr__(name, value)

    def __getattr__(self, name):
        # The existence of __getattr__ will make mypy and pylint ignore 'missing' dynamic
        # attributes and functions and wrong types of those.
        # Ideally we have a plugin for both that add the corrent type info

        if name in self.m_def.all_aliases:
            return getattr(self, self.m_def.all_aliases[name].name)

        raise AttributeError(name)

    def __check_np(self, quantity_def: 'Quantity', value: np.ndarray) -> np.ndarray:
        # TODO this feels expensive, first check, then possible convert very often?
        # if quantity_ref.type != value.dtype:
        #     raise MetainfoError(
        #         'Quantity dtype %s and value dtype %s do not match.' %
        #         (quantity_ref.type, value.dtype))

        return value

    def __set_normalize(self, quantity_def: 'Quantity', value: Any) -> Any:

        if isinstance(quantity_def.type, DataType):
            return quantity_def.type.set_normalize(self, quantity_def, value)

        elif isinstance(quantity_def.type, Section):
            if isinstance(value, MProxy):
                return value

            if not isinstance(value, MSection):
                raise TypeError(
                    'The value %s for reference quantity %s is not a section instance.' %
                    (value, quantity_def))

            if not value.m_follows(quantity_def.type):
                raise TypeError(
                    'The value %s for quantity %s does not follow %s' %
                    (value, quantity_def, quantity_def.type))

        elif isinstance(quantity_def.type, MEnum):
            if value not in quantity_def.type._values:
                raise TypeError(
                    'The value %s is not an enum value for quantity %s.' %
                    (value, quantity_def))

        elif quantity_def.type == Any:
            pass

        elif quantity_def.type == str and type(value) == np.str_:
            return str(value)

        elif quantity_def.type == bool and type(value) == np.bool_:
            return bool(value)

        else:
            if value is not None and type(value) != quantity_def.type:
                raise TypeError(
                    'The value %s with type %s for quantity %s is not of type %s' %
                    (value, type(value), quantity_def, quantity_def.type))

        return value

    def __to_np(self, quantity_def: 'Quantity', value):
        if isinstance(value, pint.quantity._Quantity):
            if quantity_def.unit is None:
                raise MetainfoError(
                    'The quantity %s has not a unit, but value %s has.' %
                    (quantity_def, value))
            value = value.to(quantity_def.unit).magnitude

        if type(value) != np.ndarray:
            if len(quantity_def.shape) > 0:
                try:
                    value = np.asarray(value)
                except TypeError:
                    raise TypeError(
                        'Could not convert value %s of %s to a numpy array' %
                        (value, quantity_def))
            elif type(value) != quantity_def.type.type:
                try:
                    value = quantity_def.type.type(value)
                except TypeError:
                    raise TypeError(
                        'Could not convert value %s of %s to a numpy scalar' %
                        (value, quantity_def))

        return self.__check_np(quantity_def, value)

    def m_set(self, quantity_def: 'Quantity', value: Any) -> None:
        ''' Set the given value for the given quantity. '''
        self.m_mod_count += 1

        if quantity_def.derived is not None:
            raise MetainfoError('The quantity %s is derived and cannot be set.' % quantity_def)

        if type(quantity_def.type) == np.dtype:
            value = self.__to_np(quantity_def, value)

        else:
            dimensions = len(quantity_def.shape)
            if dimensions == 0:
                value = self.__set_normalize(quantity_def, value)

            elif dimensions == 1:
                if type(value) == str or not isinstance(value, IterableABC):
                    raise TypeError(
                        'The shape of %s requires an iterable value, but %s is not iterable.' %
                        (quantity_def, value))

                value = list(self.__set_normalize(quantity_def, item) for item in value)

            else:
                raise MetainfoError(
                    'Only numpy arrays and dtypes can be used for higher dimensional '
                    'quantities: %s' % quantity_def)

        self.__dict__[quantity_def.name] = value

        for handler in self.m_def.event_handlers:
            if handler.__name__.startswith('on_set'):
                handler(self, quantity_def, value)

    def m_get(self, quantity_def: 'Quantity') -> Any:
        ''' Retrieve the given value for the given quantity. '''
        return quantity_def.__get__(self, Quantity)

    def m_is_set(self, quantity_def: 'Quantity') -> bool:
        ''' True if the given quantity is set. '''
        if quantity_def.derived is not None:
            return True

        return quantity_def.name in self.__dict__

    def m_add_values(self, quantity_def: 'Quantity', values: Any, offset: int) -> None:
        ''' Add (partial) values for the given quantity of higher dimensionality. '''
        # TODO
        raise NotImplementedError()

    def m_add_sub_section(self, sub_section_def: 'SubSection', sub_section: 'MSection') -> None:
        ''' Adds the given section instance as a sub section of the given sub section definition. '''

        self.m_mod_count += 1
        parent_index = -1
        if sub_section_def.repeats:
            parent_index = self.m_sub_section_count(sub_section_def)

        else:
            old_sub_section = self.__dict__.get(sub_section_def.name)
            if old_sub_section is not None:
                old_sub_section.m_parent = None
                old_sub_section.m_parent_sub_section = None
                old_sub_section.m_parent_index = -1
                if self.m_resource is not None:
                    self.m_resource.remove(sub_section)

        if sub_section is not None:
            sub_section.m_parent = self
            sub_section.m_parent_sub_section = sub_section_def
            sub_section.m_parent_index = parent_index
            if sub_section.m_resource is not None:
                sub_section.m_resource.remove(sub_section)
            if self.m_resource is not None:
                self.m_resource.add(sub_section)

        sub_section_name = sub_section_def.name
        if sub_section_def.repeats:
            sub_section_lst = self.__dict__.get(sub_section_name)
            if sub_section_lst is None:
                sub_section_lst = self.__dict__.setdefault(sub_section_name, [])

            sub_section_lst.append(sub_section)

        else:
            self.__dict__[sub_section_name] = sub_section

        for handler in self.m_def.event_handlers:
            if handler.__name__.startswith('on_add_sub_section'):
                handler(self, sub_section_def, sub_section)

    def m_remove_sub_section(self, sub_section_def: 'SubSection', index: int) -> None:
        ''' Removes the exiting section for a non repeatable sub section '''
        self.m_mod_count += 1

        if sub_section_def.repeats:
            sub_section = self.__dict__[sub_section_def.name][index]
            del(self.__dict__[sub_section_def.name][index])

        elif sub_section_def.name in self.__dict__:
            sub_section = self.__dict__[sub_section_def.name]
            del(self.__dict__[sub_section_def.name])