def seen_prebuilt_key(self, x):
# In case we are an r_dict, we don't ask for the hash ourselves.
# Note that if the custom hashing function ends up asking for
# the hash of x, then it must use compute_hash() itself, so it
# works out.
if not self.dictkey.custom_eq_hash:
compute_hash(x)
def test_hash_preservation(self):
from pypy.rlib.objectmodel import compute_hash
from pypy.rlib.objectmodel import current_object_addr_as_int
class C:
pass
class D(C):
pass
c = C()
d = D()
compute_hash(d) # force to be cached on 'd', but not on 'c'
#
def fn():
d2 = D()
return (compute_hash(d2), current_object_addr_as_int(d2),
compute_hash(c), compute_hash(d),
compute_hash(("Hi", None, (7.5, 2, d))))
f = self.getcompiled(fn)
res = f()
# xxx the next line is too precise, checking the exact implementation
assert res[0] == ~res[1]
assert res[2] != compute_hash(c) # likely
assert res[3] == compute_hash(d)
assert res[4] == compute_hash(("Hi", None, (7.5, 2, d)))
def fn():
d2 = D()
return (compute_hash(d2),
current_object_addr_as_int(d2),
compute_hash(c),
compute_hash(d),
compute_hash(("Hi", None, (7.5, 2, d))))
def define_hash_preservation(cls):
from pypy.rlib.objectmodel import compute_hash
from pypy.rlib.objectmodel import compute_identity_hash
from pypy.rlib.objectmodel import current_object_addr_as_int
class C:
pass
class D(C):
pass
c = C()
d = D()
h_d = compute_hash(d) # force to be cached on 'd', but not on 'c'
h_t = compute_hash(("Hi", None, (7.5, 2, d)))
S = lltype.GcStruct('S', ('x', lltype.Signed),
('a', lltype.Array(lltype.Signed)))
s = lltype.malloc(S, 15, zero=True)
h_s = compute_identity_hash(s) # varsized: hash not saved/restored
#
def f():
if compute_hash(c) != compute_identity_hash(c): return 12
if compute_hash(d) != h_d: return 13
if compute_hash(("Hi", None, (7.5, 2, d))) != h_t: return 14
c2 = C()
h_c2 = compute_hash(c2)
if compute_hash(c2) != h_c2: return 15
if compute_identity_hash(s) == h_s: return 16 # unlikely
i = 0
while i < 6:
rgc.collect()
if compute_hash(c2) != h_c2: return i
i += 1
return 42
return f
def test_hash_preservation(self):
from pypy.rlib.objectmodel import compute_hash
from pypy.rlib.objectmodel import current_object_addr_as_int
class C:
pass
class D(C):
pass
c = C()
d = D()
compute_hash(d) # force to be cached on 'd', but not on 'c'
#
def fn():
d2 = D()
return (compute_hash(d2),
current_object_addr_as_int(d2),
compute_hash(c),
compute_hash(d),
compute_hash(("Hi", None, (7.5, 2, d))))
f = self.getcompiled(fn)
res = f()
# xxx the next line is too precise, checking the exact implementation
assert res[0] == res[1]
assert res[2] != compute_hash(c) # likely
assert res[3] == compute_hash(d)
assert res[4] == compute_hash(("Hi", None, (7.5, 2, d)))
def fn():
d2 = D()
return (compute_hash(d2),
current_object_addr_as_int(d2),
compute_hash(c),
compute_hash(d),
compute_hash(("Hi", None, (7.5, 2, d))))
def seen_prebuilt_key(self, x):
# In case we are an r_dict, we don't ask for the hash ourselves.
# Note that if the custom hashing function ends up asking for
# the hash of x, then it must use compute_hash() itself, so it
# works out.
if not self.dictkey.custom_eq_hash:
compute_hash(x)
def _index_cache(self, selector):
space = self.space
cache = space.fromcache(IndexCache)
SHIFT2 = r_uint.BITS - space.config.objspace.std.methodcachesizeexp
SHIFT1 = SHIFT2 - 5
attrs_as_int = objectmodel.current_object_addr_as_int(self)
# ^^^Note: see comment in typeobject.py for
# _pure_lookup_where_with_method_cache()
hash_selector = objectmodel.compute_hash(selector)
product = intmask(attrs_as_int * hash_selector)
index_hash = (r_uint(product) ^ (r_uint(product) << SHIFT1)) >> SHIFT2
# ^^^Note2: same comment too
cached_attr = cache.attrs[index_hash]
if cached_attr is self:
cached_selector = cache.selectors[index_hash]
if cached_selector == selector:
index = cache.indices[index_hash]
if space.config.objspace.std.withmethodcachecounter:
name = selector[0]
cache.hits[name] = cache.hits.get(name, 0) + 1
return index
index = self._index(selector)
cache.attrs[index_hash] = self
cache.selectors[index_hash] = selector
cache.indices[index_hash] = index
if space.config.objspace.std.withmethodcachecounter:
name = selector[0]
cache.misses[name] = cache.misses.get(name, 0) + 1
return index
def _pure_lookup_where_with_method_cache(w_self, name, version_tag):
space = w_self.space
SHIFT = r_uint.BITS - space.config.objspace.std.methodcachesizeexp
version_tag_as_int = current_object_addr_as_int(version_tag)
# ^^^Note: if the version_tag object is moved by a moving GC, the
# existing method cache entries won't be found any more; new
# entries will be created based on the new address. The
# assumption is that the version_tag object won't keep moving all
# the time - so using the fast current_object_addr_as_int() instead
# of a slower solution like hash() is still a good trade-off.
hash_name = compute_hash(name)
method_hash = r_uint(intmask(version_tag_as_int * hash_name)) >> SHIFT
cached_version_tag = space.method_cache_versions[method_hash]
if cached_version_tag is version_tag:
cached_name = space.method_cache_names[method_hash]
if cached_name is name:
tup = space.method_cache_lookup_where[method_hash]
if space.config.objspace.std.withmethodcachecounter:
space.method_cache_hits[name] = \
space.method_cache_hits.get(name, 0) + 1
# print "hit", w_self, name
return tup
tup = w_self._lookup_where_all_typeobjects(name)
space.method_cache_versions[method_hash] = version_tag
space.method_cache_names[method_hash] = name
space.method_cache_lookup_where[method_hash] = tup
if space.config.objspace.std.withmethodcachecounter:
space.method_cache_misses[name] = \
space.method_cache_misses.get(name, 0) + 1
# print "miss", w_self, name
return tup
def _index_cache(self, selector):
space = self.space
cache = space.fromcache(IndexCache)
SHIFT2 = r_uint.BITS - space.config.objspace.std.methodcachesizeexp
SHIFT1 = SHIFT2 - 5
attrs_as_int = objectmodel.current_object_addr_as_int(self)
# ^^^Note: see comment in typeobject.py for
# _pure_lookup_where_with_method_cache()
hash_selector = objectmodel.compute_hash(selector)
product = intmask(attrs_as_int * hash_selector)
index_hash = (r_uint(product) ^ (r_uint(product) << SHIFT1)) >> SHIFT2
# ^^^Note2: same comment too
cached_attr = cache.attrs[index_hash]
if cached_attr is self:
cached_selector = cache.selectors[index_hash]
if cached_selector == selector:
index = cache.indices[index_hash]
if space.config.objspace.std.withmethodcachecounter:
name = selector[0]
cache.hits[name] = cache.hits.get(name, 0) + 1
return index
index = self._index(selector)
cache.attrs[index_hash] = self
cache.selectors[index_hash] = selector
cache.indices[index_hash] = index
if space.config.objspace.std.withmethodcachecounter:
name = selector[0]
cache.misses[name] = cache.misses.get(name, 0) + 1
return index
def _pure_lookup_where_with_method_cache(w_self, name, version_tag):
space = w_self.space
SHIFT = r_uint.BITS - space.config.objspace.std.methodcachesizeexp
version_tag_as_int = current_object_addr_as_int(version_tag)
# ^^^Note: if the version_tag object is moved by a moving GC, the
# existing method cache entries won't be found any more; new
# entries will be created based on the new address. The
# assumption is that the version_tag object won't keep moving all
# the time - so using the fast current_object_addr_as_int() instead
# of a slower solution like hash() is still a good trade-off.
hash_name = compute_hash(name)
method_hash = r_uint(intmask(version_tag_as_int * hash_name)) >> SHIFT
cached_version_tag = space.method_cache_versions[method_hash]
if cached_version_tag is version_tag:
cached_name = space.method_cache_names[method_hash]
if cached_name is name:
tup = space.method_cache_lookup_where[method_hash]
if space.config.objspace.std.withmethodcachecounter:
space.method_cache_hits[name] = \
space.method_cache_hits.get(name, 0) + 1
# print "hit", w_self, name
return tup
tup = w_self._lookup_where(name)
space.method_cache_versions[method_hash] = version_tag
space.method_cache_names[method_hash] = name
space.method_cache_lookup_where[method_hash] = tup
if space.config.objspace.std.withmethodcachecounter:
space.method_cache_misses[name] = \
space.method_cache_misses.get(name, 0) + 1
# print "miss", w_self, name
return tup
def f(n):
x = ''
while n > 13:
myjitdriver.can_enter_jit(n=n, x=x)
myjitdriver.jit_merge_point(n=n, x=x)
x += chr(n)
n -= 1
return compute_hash(x)
def descr_code__hash__(self):
space = self.space
result = compute_hash(self.co_name)
result ^= self.co_argcount
result ^= self.co_nlocals
result ^= self.co_flags
result ^= self.co_firstlineno
result ^= compute_hash(self.co_code)
for name in self.co_varnames: result ^= compute_hash(name)
for name in self.co_freevars: result ^= compute_hash(name)
for name in self.co_cellvars: result ^= compute_hash(name)
w_result = space.wrap(intmask(result))
for w_name in self.co_names_w:
w_result = space.xor(w_result, space.hash(w_name))
for w_const in self.co_consts_w:
w_result = space.xor(w_result, space.hash(w_const))
return w_result
def descr_code__hash__(self):
space = self.space
result = compute_hash(self.co_name)
result ^= self.co_argcount
result ^= self.co_nlocals
result ^= self.co_flags
result ^= self.co_firstlineno
result ^= compute_hash(self.co_code)
for name in self.co_varnames: result ^= compute_hash(name)
for name in self.co_freevars: result ^= compute_hash(name)
for name in self.co_cellvars: result ^= compute_hash(name)
w_result = space.wrap(intmask(result))
for w_name in self.co_names_w:
w_result = space.xor(w_result, space.hash(w_name))
for w_const in self.co_consts_w:
w_result = space.xor(w_result, space.hash(w_const))
return w_result
def hash__Unicode(space, w_uni):
s = w_uni._value
if space.config.objspace.std.withrope:
# be compatible with the special ropes hash
# XXX no caching
if len(s) == 0:
return space.wrap(0)
x = 0
for c in s:
x = intmask((1000003 * x) + ord(c))
x <<= 1
x ^= len(s)
x ^= ord(s[0])
h = intmask(x)
return space.wrap(h)
x = compute_hash(s)
return space.wrap(x)
def f():
if compute_hash(c) != compute_identity_hash(c): return 12
if compute_hash(d) != h_d: return 13
if compute_hash(("Hi", None, (7.5, 2, d))) != h_t: return 14
c2 = C()
h_c2 = compute_hash(c2)
if compute_hash(c2) != h_c2: return 15
if compute_identity_hash(s) == h_s: return 16 # unlikely
i = 0
while i < 6:
rgc.collect()
if compute_hash(c2) != h_c2: return i
i += 1
return 42
def hash(self, space):
# XXX duplicate logic from tupleobject.py
mult = 1000003
x = 0x345678
z = nValues
for i in iter_n:
value = getattr(self, 'value%s' % i)
if typetuple[i] == object:
y = space.int_w(space.hash(value))
elif typetuple[i] == float:
# get the correct hash for float which is an
# integer & other less frequent cases
from pypy.objspace.std.floatobject import _hash_float
y = _hash_float(space, value)
else:
y = compute_hash(value)
x = (x ^ y) * mult
z -= 1
mult += 82520 + z + z
x += 97531
return space.wrap(intmask(x))
def _pure_lookup_where_with_method_cache(w_self, name, version_tag):
space = w_self.space
cache = space.fromcache(MethodCache)
SHIFT2 = r_uint.BITS - space.config.objspace.std.methodcachesizeexp
SHIFT1 = SHIFT2 - 5
version_tag_as_int = current_object_addr_as_int(version_tag)
# ^^^Note: if the version_tag object is moved by a moving GC, the
# existing method cache entries won't be found any more; new
# entries will be created based on the new address. The
# assumption is that the version_tag object won't keep moving all
# the time - so using the fast current_object_addr_as_int() instead
# of a slower solution like hash() is still a good trade-off.
hash_name = compute_hash(name)
product = intmask(version_tag_as_int * hash_name)
method_hash = (r_uint(product) ^ (r_uint(product) << SHIFT1)) >> SHIFT2
# ^^^Note2: we used to just take product>>SHIFT2, but on 64-bit
# platforms SHIFT2 is really large, and we loose too much information
# that way (as shown by failures of the tests that typically have
# method names like 'f' who hash to a number that has only ~33 bits).
cached_version_tag = cache.versions[method_hash]
if cached_version_tag is version_tag:
cached_name = cache.names[method_hash]
if cached_name is name:
tup = cache.lookup_where[method_hash]
if space.config.objspace.std.withmethodcachecounter:
cache.hits[name] = cache.hits.get(name, 0) + 1
# print "hit", w_self, name
return tup
tup = w_self._lookup_where_all_typeobjects(name)
cache.versions[method_hash] = version_tag
cache.names[method_hash] = name
cache.lookup_where[method_hash] = tup
if space.config.objspace.std.withmethodcachecounter:
cache.misses[name] = cache.misses.get(name, 0) + 1
# print "miss", w_self, name
return tup
def _pure_lookup_where_with_method_cache(w_self, name, version_tag):
space = w_self.space
cache = space.fromcache(MethodCache)
SHIFT2 = r_uint.BITS - space.config.objspace.std.methodcachesizeexp
SHIFT1 = SHIFT2 - 5
version_tag_as_int = current_object_addr_as_int(version_tag)
# ^^^Note: if the version_tag object is moved by a moving GC, the
# existing method cache entries won't be found any more; new
# entries will be created based on the new address. The
# assumption is that the version_tag object won't keep moving all
# the time - so using the fast current_object_addr_as_int() instead
# of a slower solution like hash() is still a good trade-off.
hash_name = compute_hash(name)
product = intmask(version_tag_as_int * hash_name)
method_hash = (r_uint(product) ^ (r_uint(product) << SHIFT1)) >> SHIFT2
# ^^^Note2: we used to just take product>>SHIFT2, but on 64-bit
# platforms SHIFT2 is really large, and we loose too much information
# that way (as shown by failures of the tests that typically have
# method names like 'f' who hash to a number that has only ~33 bits).
cached_version_tag = cache.versions[method_hash]
if cached_version_tag is version_tag:
cached_name = cache.names[method_hash]
if cached_name is name:
tup = cache.lookup_where[method_hash]
if space.config.objspace.std.withmethodcachecounter:
cache.hits[name] = cache.hits.get(name, 0) + 1
# print "hit", w_self, name
return tup
tup = w_self._lookup_where_all_typeobjects(name)
cache.versions[method_hash] = version_tag
cache.names[method_hash] = name
cache.lookup_where[method_hash] = tup
if space.config.objspace.std.withmethodcachecounter:
cache.misses[name] = cache.misses.get(name, 0) + 1
# print "miss", w_self, name
return tup
def hash__String(space, w_str):
s = w_str._value
x = compute_hash(s)
return wrapint(space, x)
def fn(f):
return compute_hash(f)
def f(x):
return objectmodel.compute_hash(x)
def hash(self):
return (compute_hash(self.name) ^ intmask(self.left.hash() << 1) ^
intmask(self.right.hash() << 2))
def f(n):
s = malloc(STR, n)
s.hash = 0
for i in range(n):
s.chars[i] = chr(i)
return s.gethash() - compute_hash('\x00\x01\x02\x03\x04')
def f(x):
return objectmodel.compute_hash(x)
def f(n):
return compute_hash((n, 6)) == compute_hash((3, n * 2))
def f(i):
if i:
t = (None, "abc")
else:
t = ("abc", None)
return compute_hash(t)
def f(i, j):
return compute_hash((i, j))
def f(n):
s = malloc(STR, n)
s.hash = 0
for i in range(n):
s.chars[i] = chr(i)
return s.gethash() - compute_hash("\x00\x01\x02\x03\x04")
def fn(i):
if i == 0:
s = const("")
else:
s = const("xxx")
return compute_hash(s)
def test_hash(self):
def fn(f):
return compute_hash(f)
res = self.interpret(fn, [1.5])
assert res == compute_hash(1.5)
def f(n):
s = malloc(UNICODE, n)
s.hash = 0
for i in range(n):
s.chars[i] = unichr(ord("A") + i)
return s.gethash() - compute_hash(u"ABCDE")
def f(i):
if i:
t = (None, "abc")
else:
t = ("abc", None)
return compute_hash(t)
def fn(x, y):
s1 = ''.join([x, 'e', 'l', 'l', 'o'])
s2 = ''.join([y, 'e', 'l', 'l', 'o'])
return (compute_hash(s1) == compute_hash(s2)) and (s1 is not s2)
def descr_hash(self, space):
return space.wrap(compute_hash(self.as_str()))
def hash_func(x):
return objectmodel.compute_hash(x.value)
def f(i, j):
return compute_hash((i, j))
def hash(self):
return compute_hash(self.name) ^ intmask(self.child.hash() << 1)
def hash(self, storage):
return compute_hash(self.unerase(storage))
def hash(self):
return (compute_hash(self.name) ^ intmask(self.left.hash() << 1)
^ intmask(self.right.hash() << 2))
def fn(x):
if x:
obj = A()
else:
obj = None
return compute_hash(obj)
def fn():
obj = None
return compute_hash(obj)
def hash(self):
return compute_hash(self.name) ^ intmask(self.child.hash() << 1)
def _get_hash_(self):
return compute_hash(self.value)
def fn(i):
if i == 0:
s = const('')
else:
s = const("xxx")
return compute_hash(s)
def f(n):
return compute_hash((n, 6)) == compute_hash((3, n*2))
def hash_func(x):
return objectmodel.compute_hash(x.value)
def f(n):
s = malloc(UNICODE, n)
s.hash = 0
for i in range(n):
s.chars[i] = unichr(ord('A') + i)
return s.gethash() - compute_hash(u'ABCDE')
