types¶

Module with general purpose types.

nutils.types.aspreprocessor(apply)¶

Convert apply into a preprocessor decorator. When applied to a function, wrapped, the returned decorator preprocesses the arguments with apply before calling wrapped. The apply function should return a tuple of args (tuple or list) and kwargs (dict). The decorated function wrapped will be called with wrapped(*args, **kwargs). The apply function is allowed to change the signature of the decorated function.

Examples

The following preprocessor swaps two arguments.

>>> @aspreprocessor
... def swapargs(a, b):
...   return (b, a), {}

Decorating a function with swapargs will cause the arguments to be swapped before the wrapped function is called.

>>> @swapargs
... def func(a, b):
...   return a, b
>>> func(1, 2)
(2, 1)

nutils.types.apply_annotations(wrapped)¶

Decorator that applies annotations to arguments. All annotations should be callables taking one argument and returning a transformed argument. All annotations are strongly recommended to be idempotent.

If a parameter of the decorated function has a default value None and the value of this parameter is None as well, the annotation is omitted.

Examples

Consider the following function.

>>> @apply_annotations
... def f(a:tuple, b:int):
...   return a + (b,)

When calling f with a list and str as arguments, the apply_annotations() decorator first applies tuple and int to the arguments before passing them to the decorated function.

>>> f([1, 2], '3')
(1, 2, 3)

The following example illustrates the behavior of parameters with default value None.

>>> addone = lambda x: x+1
>>> @apply_annotations
... def g(a:addone=None):
...   return a

When calling g without arguments or with argument None, the annotation addone is not applied. Note that None + 1 would raise an exception.

>>> g() is None
True
>>> g(None) is None
True

When passing a different value, the annotation is applied:

>>> g(1)
2

nutils.types.argument_canonicalizer(signature)¶

Returns a function that converts arguments matching signature to canonical positional and keyword arguments. If possible, an argument is added to the list of positional arguments, otherwise to the keyword arguments dictionary. The returned arguments include default values.

Parameters:: signature (inspect.Signature) – The signature of a function to generate canonical arguments for.
Returns:: A function that returns a tuple of a tuple of positional arguments and a dict of keyword arguments.
Return type:: callable

Examples

Consider the following function.

>>> def f(a, b=4, *, c): pass

The argument_canonicalizer for f is generated as follows:

>>> canon = argument_canonicalizer(inspect.signature(f))

Calling canon with parameter b passed as keyword returns arguments with parameter b as positional argument:

>>> canon(1, c=3, b=2)
((1, 2), {'c': 3})

When calling canon without parameter b the default value is added to the positional arguments:

>>> canon(1, c=3)
((1, 4), {'c': 3})

nutils.types.nutils_hash(data)¶

Compute a stable hash of immutable object data. The hash is not affected by Python’s hash randomization (see object.__hash__()).

Parameters:: data – An immutable object of type bool, int, float, complex, str, bytes, tuple, frozenset, or Ellipsis or None, or the type itself, or an object with a __nutils_hash__ attribute.
Returns:: The hash of data.
Return type:: 40 bytes

class nutils.types.CacheMeta(name, bases, namespace, **kwargs)¶

Bases: ABCMeta

Metaclass that adds caching functionality to properties and methods listed in the special attribute __cache__. If an attribute is of type property, the value of the property will be computed at the first attribute access and served from cache subsequently. If an attribute is a method, the arguments and return value are cached and the cached value will be used if a subsequent call is made with the same arguments; if not, the cache will be overwritten. The cache lives in private attributes in the instance. The metaclass supports the use of __slots__. If a subclass redefines a cached property or method (in the sense of this metaclass) of a base class, the property or method of the subclass is not automatically cached; __cache__ should be used in the subclass explicitly.

Examples

An example of a class with a cached property:

>>> class T(metaclass=CacheMeta):
...   __cache__ = 'x',
...   @property
...   def x(self):
...     print('uncached')
...     return 1

The print statement is added to illustrate when method x (as defined above) is called:

>>> t = T()
>>> t.x
uncached
1
>>> t.x
1

An example of a class with a cached method:

>>> class U(metaclass=CacheMeta):
...   __cache__ = 'y',
...   def y(self, a):
...     print('uncached')
...     return a

Again, the print statement is added to illustrate when the method y (as defined above) is called:

>>> u = U()
>>> u.y(1)
uncached
1
>>> u.y(1)
1
>>> u.y(2)
uncached
2
>>> u.y(2)
2

class nutils.types.Immutable(*args, **kwargs)¶

Bases: object

Base class for immutable types. This class adds equality tests, traditional hashing (hash()), nutils hashing (nutils_hash()) and pickling, all based solely on the (positional) intialization arguments, args for future reference. Keyword-only arguments are not supported. All arguments should be hashable by nutils_hash().

Positional and keyword initialization arguments are canonicalized automatically (by argument_canonicalizer()). If the __init__ method of a subclass is decorated with preprocessors (see aspreprocessor()), the preprocessors are applied to the initialization arguments and args becomes the preprocessed positional part.

Examples

Consider the following class.

>>> class Plain(Immutable):
...   def __init__(self, a, b):
...     pass

Calling Plain with equivalent positional or keyword arguments produces equal instances:

>>> Plain(1, 2) == Plain(a=1, b=2)
True

Passing unhashable values to Plain will fail:

>>> Plain([1, 2], [3, 4]) 
Traceback (most recent call last):
    ...
TypeError: unhashable type: 'list'

This can be solved by adding and applying annotations to __init__. The following class converts its initialization arguments to tuple automaticaly:

>>> class Annotated(Immutable):
...   @apply_annotations
...   def __init__(self, a:tuple, b:tuple):
...     pass

Calling Annotated with either lists of 1, 2 and 3, 4 or tuples gives equal instances:

>>> Annotated([1, 2], [3, 4]) == Annotated((1, 2), (3, 4))
True

class nutils.types.Singleton(*args, **kwargs)¶

Bases: Immutable

Subclass of Immutable that creates a single instance per unique set of initialization arguments.

Examples

Consider the following class.

>>> class Plain(Singleton):
...   def __init__(self, a, b):
...     pass

Calling Plain with equivalent positional or keyword arguments produces one instance:

>>> Plain(1, 2) is Plain(a=1, b=2)
True

Consider the folling class with annotations.

>>> class Annotated(Singleton):
...   @apply_annotations
...   def __init__(self, a:tuple, b:tuple):
...     pass

Calling Annotated with either lists of 1, 2 and 3, 4 or tuples gives a single instance:

>>> Annotated([1, 2], [3, 4]) is Annotated((1, 2), (3, 4))
True

__eq__(self, value, /)¶: Return self==value.

class nutils.types.arraydata(dtype, shape, bytes)¶

Bases: Singleton

hashable array container.

The container can be used for fast equality checks and for dictionary keys. Data is copied at construction and canonicalized by casting it to the platform’s primary data representation (e.g. int64 i/o int32). It can be retrieved via numpy.asarray(). Additionally the arraydata object provides direct access to the array’s shape, dtype and bytes.

Example

>>> a = numpy.array([1,2,3])
>>> w = arraydata(a)
>>> w == arraydata([1,2,4]) # NOTE: equality only if entire array matches
False
>>> numpy.asarray(w)
array([1, 2, 3])

nutils.types.strictint(value)¶

Converts any type that is a subclass of numbers.Integral (e.g. int and numpy.int64) to int, and fails otherwise. Notable differences with the behavior of int:

strictint() does not convert a str to an int.
strictint() does not truncate float to an int.

Examples

>>> strictint(1), type(strictint(1))
(1, <class 'int'>)
>>> strictint(numpy.int64(1)), type(strictint(numpy.int64(1)))
(1, <class 'int'>)
>>> strictint(1.0)
Traceback (most recent call last):
    ...
ValueError: not an integer: 1.0
>>> strictint('1')
Traceback (most recent call last):
    ...
ValueError: not an integer: '1'

nutils.types.strictfloat(value)¶

Converts any type that is a subclass of numbers.Real (e.g. float and numpy.float64) to float, and fails otherwise. Notable difference with the behavior of float:

strictfloat() does not convert a str to an float.

Examples

>>> strictfloat(1), type(strictfloat(1))
(1.0, <class 'float'>)
>>> strictfloat(numpy.float64(1.2)), type(strictfloat(numpy.float64(1.2)))
(1.2, <class 'float'>)
>>> strictfloat(1.2+3.4j)
Traceback (most recent call last):
    ...
ValueError: not a real number: (1.2+3.4j)
>>> strictfloat('1.2')
Traceback (most recent call last):
    ...
ValueError: not a real number: '1.2'

nutils.types.strictstr(value)¶

Returns value unmodified if it is a str, and fails otherwise. Notable difference with the behavior of str:

strictstr() does not call __str__ methods of objects to automatically convert objects to strs.

Examples

Passing a str to strictstr() works:

>>> strictstr('spam')
'spam'

Passing an int will fail:

>>> strictstr(1)
Traceback (most recent call last):
    ...
ValueError: not a 'str': 1

class nutils.types.strict(*args, **kwargs)¶

Bases: object

Type checker. The function strict[cls](value) returns value unmodified if value is an instance of cls, otherwise a ValueError is raised.

Examples

The strict[int] function passes integers unmodified:

>>> strict[int](1)
1

Other types fail:

>>> strict[int]('1')
Traceback (most recent call last):
    ...
ValueError: expected an object of type 'int' but got '1' with type 'str'

__weakref__¶: list of weak references to the object (if defined)

class nutils.types.tuple(*args, **kwargs)¶

Bases: tuple

Wrapper of tuple that supports a user-defined item constructor via the notation tuple[I], with I the item constructor. This is shorthand for lambda items: tuple(map(I, items)). The item constructor should be any callable that takes one argument.

Examples

A tuple with items processed with strictint():

>>> tuple[strictint]((False, 1, 2, numpy.int64(3)))
(0, 1, 2, 3)

If the item constructor raises an exception, the construction of the tuple failes accordingly:

>>> tuple[strictint]((1, 2, 3.4))
Traceback (most recent call last):
    ...
ValueError: not an integer: 3.4

class nutils.types.frozendict(base)¶

Bases: Mapping

An immutable version of dict. The frozendict is hashable and both the keys and values should be hashable as well.

Custom key and value constructors can be supplied via the frozendict[K,V] notation, with K the key constructor and V the value constructor, which is roughly equivalent to lambda *args, **kwargs: {K(k): V(v) for k, v in dict(*args, **kwargs).items()}.

Examples

A frozendict with strictstr() as key constructor and strictfloat() as value constructor:

>>> frozendict[strictstr,strictfloat]({'spam': False})
frozendict({'spam': 0.0})

Similar but with non-strict constructors:

>>> frozendict[str,float]({None: '1.2'})
frozendict({'None': 1.2})

Applying the strict constructors to invalid data raises an exception:

>>> frozendict[strictstr,strictfloat]({None: '1.2'})
Traceback (most recent call last):
    ...
ValueError: not a 'str': None

class nutils.types.frozenmultiset(items)¶

Bases: Container

An immutable multiset. A multiset is a generalization of a set: items may occur more than once. Two mutlisets are equal if they have the same set of items and the same item multiplicities.

A custom item constructor can be supplied via the notation frozenmultiset[I], with I the item constructor. This is shorthand for lambda items: frozenmultiset(map(I, items)). The item constructor should be any callable that takes one argument.

Examples

>>> a = frozenmultiset(['spam', 'bacon', 'spam'])
>>> b = frozenmultiset(['sausage', 'spam'])

The frozenmultiset objects support +, - and & operators:

>>> a + b
frozenmultiset(['spam', 'bacon', 'spam', 'sausage', 'spam'])
>>> a - b
frozenmultiset(['bacon', 'spam'])
>>> a & b
frozenmultiset(['spam'])

The order of the items is irrelevant:

>>> frozenmultiset(['spam', 'spam', 'eggs']) == frozenmultiset(['spam', 'eggs', 'spam'])
True

The multiplicities, however, are not:

>>> frozenmultiset(['spam', 'spam', 'eggs']) == frozenmultiset(['spam', 'eggs'])
False

__and__(self, other)¶: Return a frozenmultiset with elements from the left and right hand sides with strict positive multiplicity, where the multiplicity is the minimum of multiplicitie of the left and right hand side.

__add__(self, other)¶: Return a frozenmultiset with elements from the left and right hand sides with a multiplicity equal to the sum of the left and right hand sides.

__sub__(self, other)¶: Return a frozenmultiset with elements from the left hand sides with a multiplicity equal to the difference of the multiplicity of the left and right hand sides, truncated to zero. Elements with multiplicity zero are omitted.

nutils.types.frozenarray(arg, *, copy=True, dtype=None)¶

Create read-only Numpy array.

Parameters:

arg (numpy.ndarray or array_like) – Input data.
copy (bool) – If True (the default), do not modify the argument in place. No copy is ever forced if the argument is already immutable.
dtype (numpy.dtype or dtype_like, optional) – The desired data-type for the array.

Return type:

numpy.ndarray

class nutils.types.c_array(*args, **kwargs)¶

Bases: object

Converts an array-like object to a ctypes array with a specific dtype. The function c_array[dtype](array) returns array unmodified if array is already a ctypes array. If array is a numpy.ndarray, the array is converted if the dtype is correct and the array is contiguous; otherwise ValueError is raised. Otherwise, array is first converted to a contiguous numpy.ndarray and then converted to ctypes array. In the first two cases changes made to the ctypes array are reflected by the array argument: both are essentially views of the same data. In the third case, changes to either array or the returned ctypes array are not reflected by the other.

__weakref__¶: list of weak references to the object (if defined)

nutils.types.lru_cache(func)¶

Buffer-aware cache.

Returns values from a cache for previously seen arguments. Arguments must be hasheable objects or immutable Numpy arrays, the latter identified by the underlying buffer. Destruction of the buffer triggers a callback that removes the corresponding cache entry.

At present, any writeable array will silently disable caching. This bevaviour is transitional, with future versions requiring that all arrays be immutable.

Caution

When a decorated function returns an object that references its argument (for instance, by returning the argument itself), the cached value keeps an argument’s reference count from falling to zero, causing the object to remain in cache indefinitely. For this reason, care must be taken that the decorator is only applied to functions that return objects with no references to its arguments.

class nutils.types.attributes(**args)¶

Bases: object

Dictionary-like container with attributes instead of keys, instantiated using keyword arguments:

>>> A = attributes(foo=10, bar=True)
>>> A
attributes(bar=True, foo=10)
>>> A.foo
10

__weakref__¶: list of weak references to the object (if defined)

nutils.types.hashable_function(identifier)¶

Decorator that wraps the decorated function and adds a Nutils hash.

Return a decorator that wraps the decorated function and adds a Nutils hash based solely on the given identifier. The identifier can be anything that has a Nutils hash. The identifier should represent the behavior of the function and should be changed when the behavior of the function changes.

If used on methods, this decorator behaves like staticmethod().

Examples

Make some function func hashable:

>>> @hashable_function('func v1')
... def func(a, b):
...   return a + b
...

The Nutils hash can be obtained by calling nutils_hash on func:

>>> nutils_hash(func).hex()
'b7fed72647f6a88dd3ce3808b2710eede7d7b5a5'

Note that the hash is based solely on the identifier passed to the decorator. If we create another function other with the same identifier as func, then both have the same hash, despite returning different values.

>>> @hashable_function('func v1')
... def other(a, b):
...   return a * b
...
>>> nutils_hash(other) == nutils_hash(func)
True
>>> func(1, 2) == other(1, 2)
False

The decorator can also be applied on methods:

>>> class Spam:
...   @hashable_function('Spam.eggs v1')
...   def eggs(a, b):
...     # NOTE: `self` is absent because `hashable_function` behaves like `staticmethod`.
...     return a + b
...

The hash of eggs accessed via the class or an instance is the same:

>>> spam = Spam()
>>> nutils_hash(Spam.eggs).hex()
'dfdbb0ce20b617b17c3b854c23b2b9f7deb94cc6'
>>> nutils_hash(spam.eggs).hex()
'dfdbb0ce20b617b17c3b854c23b2b9f7deb94cc6'