remove CacheMeta-based caching (#784)

This PR removes all uses of the `__cache__` mechanism, facilitated by
the `CacheMeta` meta class, and replaces it by the cached_property and
`lru_cache` decorators which are considered to give better visibility to
caching structures. The `CacheMeta` class is removed as no longer
relevant.

nutils/_backports.py

@@ -0,0 +1,53 @@
+"""
+The backports module provides minimal fallback implementations for functions
+that are introduced in versions of Python that are newer than the minimum
+required version. Function behavour is equal to its Python counterpart only to
+the extent that the Nutils code base requires it to be. As such,
+implementations found in this module should not be relied upon as general
+drop-on replacements.
+"""
+
+# Introduced in Python 3.8
+
+try:
+    from functools import cached_property
+except ImportError:
+
+    # Fallback implementation. Notable difference: no lock is used to prevent
+    # race conditions in multi-threaded contexts.
+
+    class cached_property:
+
+        def __init__(self, func):
+            self.func = func
+
+        def __set_name__(self, owner, name):
+            self.attrname = name
+
+        def __get__(self, instance, owner=None):
+            if instance is None:
+                return self
+            try:
+                val = instance.__dict__[self.attrname]
+            except KeyError:
+                val = instance.__dict__[self.attrname] = self.func(instance)
+            return val
+
+
+try:
+    from math import comb
+except ImportError:
+
+    # Fallback implementation. Notable difference: if k > n, this
+    # implementation raises a ValueError rather than returning 0.
+
+    import math, functools, operator
+
+    def comb(n, k):
+        a, b = sorted([k, n-k])
+        numer = functools.reduce(operator.mul, range(1+b, 1+n), 1)
+        denom = math.factorial(a)
+        return numer // denom
+
+
+# vim:sw=4:sts=4:et

nutils/_util.py

@@ -384,20 +384,6 @@ def unique(items, key=None):
     return unique, indices
 
 
-try:
-    cached_property = functools.cached_property
-except AttributeError:  # python < 3.8
-    def cached_property(func):  # minimal backport
-        @functools.wraps(func)
-        def wrapped(self):
-            try:
-                val = self.__dict__[func.__name__]
-            except KeyError:
-                self.__dict__[func.__name__] = val = func(self)
-            return val
-        return property(wrapped)
-
-
 def defaults_from_env(f):
     '''Decorator for changing function defaults based on environment.