Move Interface hashing and comparison to C; 2.5 to 15x speedup in micro benchmarks

CHANGES.rst

@@ -210,6 +210,17 @@
   hierarchy, ``Interface`` could be assigned too high a priority.
   See `issue 8 <https://github.com/zopefoundation/zope.interface/issues/8>`_.
 
+- Implement sorting, equality, and hashing in C for ``Interface``
+  objects. In micro benchmarks, this makes those operations 40% to 80%
+  faster. This translates to a 20% speed up in querying adapters.
+
+  Note that this changes certain implementation details. In
+  particular, ``InterfaceClass`` now has a non-default metaclass, and
+  it is enforced that ``__module__`` in instances of
+  ``InterfaceClass`` is read-only.
+
+  See `PR 183 <https://github.com/zopefoundation/zope.interface/pull/183>`_.
+
 
 4.7.2 (2020-03-10)
 ==================

MANIFEST.in

@@ -18,3 +18,4 @@ global-exclude coverage.xml
 global-exclude appveyor.yml
 
 prune docs/_build
+prune benchmarks

benchmarks/micro.py

@@ -0,0 +1,234 @@
+import pyperf
+
+from zope.interface import Interface
+from zope.interface import classImplements
+from zope.interface import implementedBy
+from zope.interface.interface import InterfaceClass
+from zope.interface.registry import Components
+
+# Long, mostly similar names are a worst case for equality
+# comparisons.
+ifaces = [
+    InterfaceClass('I' + ('0' * 20) + str(i), (Interface,), {})
+    for i in range(100)
+]
+
+class IWideInheritance(*ifaces):
+    """
+    Inherits from 100 unrelated interfaces.
+    """
+
+class WideInheritance(object):
+    pass
+classImplements(WideInheritance, IWideInheritance)
+
+def make_deep_inheritance():
+    children = []
+    base = Interface
+    for iface in ifaces:
+        child = InterfaceClass('IDerived' + base.__name__, (iface, base,), {})
+        base = child
+        children.append(child)
+    return children
+
+deep_ifaces = make_deep_inheritance()
+
+class DeepestInheritance(object):
+    pass
+classImplements(DeepestInheritance, deep_ifaces[-1])
+
+def make_implementer(iface):
+    c = type('Implementer' + iface.__name__, (object,), {})
+    classImplements(c, iface)
+    return c
+
+implementers = [
+    make_implementer(iface)
+    for iface in ifaces
+]
+
+providers = [
+    implementer()
+    for implementer in implementers
+]
+
+INNER = 100
+
+def bench_in(loops, o):
+    t0 = pyperf.perf_counter()
+    for _ in range(loops):
+        for _ in range(INNER):
+            o.__contains__(Interface)
+
+    return pyperf.perf_counter() - t0
+
+def bench_sort(loops, objs):
+    import random
+    rand = random.Random(8675309)
+
+    shuffled = list(objs)
+    rand.shuffle(shuffled)
+
+    t0 = pyperf.perf_counter()
+    for _ in range(loops):
+        for _ in range(INNER):
+            sorted(shuffled)
+
+    return pyperf.perf_counter() - t0
+
+def bench_query_adapter(loops, components, objs=providers):
+    # One time through to prime the caches
+    for iface in ifaces:
+        for provider in providers:
+            components.queryAdapter(provider, iface)
+
+    t0 = pyperf.perf_counter()
+    for _ in range(loops):
+        for iface in ifaces:
+            for provider in objs:
+                components.queryAdapter(provider, iface)
+    return pyperf.perf_counter() - t0
+
+
+def bench_getattr(loops, name, get=getattr):
+    t0 = pyperf.perf_counter()
+    for _ in range(loops):
+        for _ in range(INNER):
+            get(Interface, name) # 1
+            get(Interface, name) # 2
+            get(Interface, name) # 3
+            get(Interface, name) # 4
+            get(Interface, name) # 5
+            get(Interface, name) # 6
+            get(Interface, name) # 7
+            get(Interface, name) # 8
+            get(Interface, name) # 9
+            get(Interface, name) # 10
+    return pyperf.perf_counter() - t0
+
+runner = pyperf.Runner()
+
+runner.bench_time_func(
+    'read __module__', # stored in C, accessed through __getattribute__
+    bench_getattr,
+    '__module__',
+    inner_loops=INNER * 10
+)
+
+runner.bench_time_func(
+    'read __name__', # stored in C, accessed through PyMemberDef
+    bench_getattr,
+    '__name__',
+    inner_loops=INNER * 10
+)
+
+runner.bench_time_func(
+    'read __doc__', # stored in Python instance dictionary directly
+    bench_getattr,
+    '__doc__',
+    inner_loops=INNER * 10
+)
+
+runner.bench_time_func(
+    'read providedBy', # from the class, wrapped into a method object.
+    bench_getattr,
+    'providedBy',
+    inner_loops=INNER * 10
+)
+
+runner.bench_time_func(
+    'query adapter (no registrations)',
+    bench_query_adapter,
+    Components(),
+    inner_loops=1
+)
+
+def populate_components():
+    def factory(o):
+        return 42
+
+    pop_components = Components()
+    for iface in ifaces:
+        for other_iface in ifaces:
+            pop_components.registerAdapter(factory, (iface,), other_iface, event=False)
+
+    return pop_components
+
+runner.bench_time_func(
+    'query adapter (all trivial registrations)',
+    bench_query_adapter,
+    populate_components(),
+    inner_loops=1
+)
+
+runner.bench_time_func(
+    'query adapter (all trivial registrations, wide inheritance)',
+    bench_query_adapter,
+    populate_components(),
+    [WideInheritance()],
+    inner_loops=1
+)
+
+runner.bench_time_func(
+    'query adapter (all trivial registrations, deep inheritance)',
+    bench_query_adapter,
+    populate_components(),
+    [DeepestInheritance()],
+    inner_loops=1
+)
+
+runner.bench_time_func(
+    'sort interfaces',
+    bench_sort,
+    ifaces,
+    inner_loops=INNER,
+)
+
+runner.bench_time_func(
+    'sort implementedBy',
+    bench_sort,
+    [implementedBy(p) for p in implementers],
+    inner_loops=INNER,
+)
+
+runner.bench_time_func(
+    'sort mixed',
+    bench_sort,
+    [implementedBy(p) for p in implementers] + ifaces,
+    inner_loops=INNER,
+)
+
+runner.bench_time_func(
+    'contains (empty dict)',
+    bench_in,
+    {},
+    inner_loops=INNER
+)
+
+runner.bench_time_func(
+    'contains (populated dict: interfaces)',
+    bench_in,
+    {k: k for k in ifaces},
+    inner_loops=INNER
+)
+
+runner.bench_time_func(
+    'contains (populated list: interfaces)',
+    bench_in,
+    ifaces,
+    inner_loops=INNER
+)
+
+runner.bench_time_func(
+    'contains (populated dict: implementedBy)',
+    bench_in,
+    {implementedBy(p): 1 for p in implementers},
+    inner_loops=INNER
+)
+
+runner.bench_time_func(
+    'contains (populated list: implementedBy)',
+    bench_in,
+    [implementedBy(p) for p in implementers],
+    inner_loops=INNER
+)

src/zope/interface/_compat.py

@@ -166,4 +166,5 @@ def find_impl():
     # name (for testing and documentation)
     globs[name + 'Py'] = py_impl
     globs[name + 'Fallback'] = py_impl
+
     return c_impl