Author: Armin Rigo <[email protected]>
Branch: rpython-hash
Changeset: r89807:52687af2d35a
Date: 2017-01-27 22:00 +0100
http://bitbucket.org/pypy/pypy/changeset/52687af2d35a/
Log: (untested) remove some code that implements keeping the hashes
across translation
diff --git a/rpython/memory/gc/incminimark.py b/rpython/memory/gc/incminimark.py
--- a/rpython/memory/gc/incminimark.py
+++ b/rpython/memory/gc/incminimark.py
@@ -117,9 +117,7 @@
# The following flag is set on nursery objects of which we asked the id
# or the identityhash. It means that a space of the size of the object
-# has already been allocated in the nonmovable part. The same flag is
-# abused to mark prebuilt objects whose hash has been taken during
-# translation and is statically recorded.
+# has already been allocated in the nonmovable part.
GCFLAG_HAS_SHADOW = first_gcflag << 3
# The following flag is set temporarily on some objects during a major
@@ -208,10 +206,6 @@
# by GCFLAG_xxx above.
HDR = lltype.Struct('header', ('tid', lltype.Signed))
typeid_is_in_field = 'tid'
- withhash_flag_is_in_field = 'tid', GCFLAG_HAS_SHADOW
- # ^^^ prebuilt objects may have the flag GCFLAG_HAS_SHADOW;
- # then they are one word longer, the extra word storing the hash.
-
# During a minor collection, the objects in the nursery that are
# moved outside are changed in-place: their header is replaced with
@@ -2640,40 +2634,22 @@
return shadow
_find_shadow._dont_inline_ = True
- @specialize.arg(2)
- def id_or_identityhash(self, gcobj, is_hash):
+ def id_or_identityhash(self, gcobj):
"""Implement the common logic of id() and identityhash()
of an object, given as a GCREF.
"""
obj = llmemory.cast_ptr_to_adr(gcobj)
- #
if self.is_valid_gc_object(obj):
if self.is_in_nursery(obj):
obj = self._find_shadow(obj)
- elif is_hash:
- if self.header(obj).tid & GCFLAG_HAS_SHADOW:
- #
- # For identityhash(), we need a special case for some
- # prebuilt objects: their hash must be the same before
- # and after translation. It is stored as an extra word
- # after the object. But we cannot use it for id()
- # because the stored value might clash with a real one.
- size = self.get_size(obj)
- i = (obj + size).signed[0]
- # Important: the returned value is not mangle_hash()ed!
- return i
- #
- i = llmemory.cast_adr_to_int(obj)
- if is_hash:
- i = mangle_hash(i)
- return i
+ return llmemory.cast_adr_to_int(obj)
id_or_identityhash._always_inline_ = True
def id(self, gcobj):
- return self.id_or_identityhash(gcobj, False)
+ return self.id_or_identityhash(gcobj)
def identityhash(self, gcobj):
- return self.id_or_identityhash(gcobj, True)
+ return mangle_hash(self.id_or_identityhash(gcobj))
# ----------
# Finalizers
diff --git a/rpython/memory/gc/minimark.py b/rpython/memory/gc/minimark.py
--- a/rpython/memory/gc/minimark.py
+++ b/rpython/memory/gc/minimark.py
@@ -104,9 +104,7 @@
# The following flag is set on nursery objects of which we asked the id
# or the identityhash. It means that a space of the size of the object
-# has already been allocated in the nonmovable part. The same flag is
-# abused to mark prebuilt objects whose hash has been taken during
-# translation and is statically recorded.
+# has already been allocated in the nonmovable part.
GCFLAG_HAS_SHADOW = first_gcflag << 3
# The following flag is set temporarily on some objects during a major
@@ -149,9 +147,6 @@
# by GCFLAG_xxx above.
HDR = lltype.Struct('header', ('tid', lltype.Signed))
typeid_is_in_field = 'tid'
- withhash_flag_is_in_field = 'tid', GCFLAG_HAS_SHADOW
- # ^^^ prebuilt objects may have the flag GCFLAG_HAS_SHADOW;
- # then they are one word longer, the extra word storing the hash.
_ADDRARRAY = lltype.Array(llmemory.Address, hints={'nolength': True})
@@ -1868,40 +1863,22 @@
return shadow
_find_shadow._dont_inline_ = True
- @specialize.arg(2)
- def id_or_identityhash(self, gcobj, is_hash):
+ def id_or_identityhash(self, gcobj):
"""Implement the common logic of id() and identityhash()
of an object, given as a GCREF.
"""
obj = llmemory.cast_ptr_to_adr(gcobj)
- #
if self.is_valid_gc_object(obj):
if self.is_in_nursery(obj):
obj = self._find_shadow(obj)
- elif is_hash:
- if self.header(obj).tid & GCFLAG_HAS_SHADOW:
- #
- # For identityhash(), we need a special case for some
- # prebuilt objects: their hash must be the same before
- # and after translation. It is stored as an extra word
- # after the object. But we cannot use it for id()
- # because the stored value might clash with a real one.
- size = self.get_size(obj)
- i = (obj + size).signed[0]
- # Important: the returned value is not mangle_hash()ed!
- return i
- #
- i = llmemory.cast_adr_to_int(obj)
- if is_hash:
- i = mangle_hash(i)
- return i
+ return llmemory.cast_adr_to_int(obj)
id_or_identityhash._always_inline_ = True
def id(self, gcobj):
- return self.id_or_identityhash(gcobj, False)
+ return self.id_or_identityhash(gcobj)
def identityhash(self, gcobj):
- return self.id_or_identityhash(gcobj, True)
+ return mangle_hash(self.id_or_identityhash(gcobj))
# ----------
# Finalizers
diff --git a/rpython/memory/gc/semispace.py b/rpython/memory/gc/semispace.py
--- a/rpython/memory/gc/semispace.py
+++ b/rpython/memory/gc/semispace.py
@@ -48,9 +48,6 @@
HDR = lltype.Struct('header', ('tid', lltype.Signed)) # XXX or rffi.INT?
typeid_is_in_field = 'tid'
- withhash_flag_is_in_field = 'tid', _GCFLAG_HASH_BASE * 0x2
- # ^^^ prebuilt objects either have GC_HASH_TAKEN_ADDR or they
- # have GC_HASH_HASFIELD (and then they are one word longer).
FORWARDSTUB = lltype.GcStruct('forwarding_stub',
('forw', llmemory.Address))
FORWARDSTUBPTR = lltype.Ptr(FORWARDSTUB)
diff --git a/rpython/memory/gctransform/boehm.py
b/rpython/memory/gctransform/boehm.py
--- a/rpython/memory/gctransform/boehm.py
+++ b/rpython/memory/gctransform/boehm.py
@@ -194,7 +194,6 @@
def gcheader_initdata(self, obj):
hdr = lltype.malloc(self.HDR, immortal=True)
- #hdr.hash = lltype.identityhash_nocache(obj._as_ptr())
return hdr._obj
diff --git a/rpython/memory/gctransform/framework.py
b/rpython/memory/gctransform/framework.py
--- a/rpython/memory/gctransform/framework.py
+++ b/rpython/memory/gctransform/framework.py
@@ -610,18 +610,8 @@
def special_funcptr_for_type(self, TYPE):
return self.layoutbuilder.special_funcptr_for_type(TYPE)
- def gc_header_for(self, obj, needs_hash=False):
+ def gc_header_for(self, obj):
hdr = self.gcdata.gc.gcheaderbuilder.header_of_object(obj)
- withhash, flag = self.gcdata.gc.withhash_flag_is_in_field
- x = getattr(hdr, withhash)
- TYPE = lltype.typeOf(x)
- x = lltype.cast_primitive(lltype.Signed, x)
- if needs_hash:
- x |= flag # set the flag in the header
- else:
- x &= ~flag # clear the flag in the header
- x = lltype.cast_primitive(TYPE, x)
- setattr(hdr, withhash, x)
return hdr
def get_hash_offset(self, T):
@@ -1514,22 +1504,9 @@
def gcheader_initdata(self, obj):
o = lltype.top_container(obj)
- needs_hash = self.get_prebuilt_hash(o) is not None
- hdr = self.gc_header_for(o, needs_hash)
+ hdr = self.gc_header_for(o)
return hdr._obj
- def DISABLED_get_prebuilt_hash(self, obj):
- # for prebuilt objects that need to have their hash stored and
- # restored. Note that only structures that are StructNodes all
- # the way have their hash stored (and not e.g. structs with var-
- # sized arrays at the end). 'obj' must be the top_container.
- TYPE = lltype.typeOf(obj)
- if not isinstance(TYPE, lltype.GcStruct):
- return None
- if TYPE._is_varsize():
- return None
- return getattr(obj, '_hash_cache_', None)
-
def get_finalizer_queue_index(self, hop):
fq_tag = hop.spaceop.args[0].value
assert 'FinalizerQueue TAG' in fq_tag.expr
diff --git a/rpython/memory/gctransform/refcounting.py
b/rpython/memory/gctransform/refcounting.py
--- a/rpython/memory/gctransform/refcounting.py
+++ b/rpython/memory/gctransform/refcounting.py
@@ -18,8 +18,7 @@
class RefcountingGCTransformer(GCTransformer):
malloc_zero_filled = True
- HDR = lltype.Struct("header", ("refcount", lltype.Signed),
- ("hash", lltype.Signed))
+ HDR = lltype.Struct("header", ("refcount", lltype.Signed))
def __init__(self, translator):
super(RefcountingGCTransformer, self).__init__(translator, inline=True)
@@ -78,9 +77,7 @@
def ll_identityhash(addr):
obj = llmemory.cast_adr_to_ptr(addr, HDRPTR)
- h = obj.hash
- if h == 0:
- obj.hash = h = llmemory.cast_adr_to_int(addr)
+ h = llmemory.cast_adr_to_int(addr)
return h
if self.translator:
@@ -178,7 +175,6 @@
if not self.gcheaderbuilder.get_header(p):
hdr = self.gcheaderbuilder.new_header(p)
hdr.refcount = sys.maxint // 2
- hdr.hash = lltype.identityhash_nocache(p)
def static_deallocation_funcptr_for_type(self, TYPE):
if TYPE in self.static_deallocator_funcptrs:
diff --git a/rpython/memory/gctransform/transform.py
b/rpython/memory/gctransform/transform.py
--- a/rpython/memory/gctransform/transform.py
+++ b/rpython/memory/gctransform/transform.py
@@ -374,9 +374,6 @@
return hop.cast_result(rmodel.inputconst(lltype.Ptr(ARRAY_TYPEID_MAP),
lltype.nullptr(ARRAY_TYPEID_MAP)))
- def get_prebuilt_hash(self, obj):
- return None
-
class MinimalGCTransformer(BaseGCTransformer):
def __init__(self, parenttransformer):
diff --git a/rpython/rlib/objectmodel.py b/rpython/rlib/objectmodel.py
--- a/rpython/rlib/objectmodel.py
+++ b/rpython/rlib/objectmodel.py
@@ -484,17 +484,11 @@
"""RPython equivalent of object.__hash__(x). This returns the
so-called 'identity hash', which is the non-overridable default hash
of Python. Can be called for any RPython-level object that turns
- into a GC object, but not NULL. The value is not guaranteed to be the
- same before and after translation, except for RPython instances on the
- lltypesystem.
+ into a GC object, but not NULL. The value will be different before
+ and after translation.
"""
assert x is not None
- result = object.__hash__(x)
- try:
- x.__dict__['__precomputed_identity_hash'] = result
- except (TypeError, AttributeError):
- pass
- return result
+ return object.__hash__(x)
def compute_unique_id(x):
"""RPython equivalent of id(x). The 'x' must be an RPython-level
diff --git a/rpython/rtyper/lltypesystem/lltype.py
b/rpython/rtyper/lltypesystem/lltype.py
--- a/rpython/rtyper/lltypesystem/lltype.py
+++ b/rpython/rtyper/lltypesystem/lltype.py
@@ -1380,20 +1380,11 @@
return callb(*args)
raise TypeError("%r instance is not a function" % (self._T,))
- def _identityhash(self, cache=True):
+ def _identityhash(self):
p = normalizeptr(self)
- try:
- return p._obj._hash_cache_
- except AttributeError:
- assert self._T._gckind == 'gc'
- assert self # not for NULL
- result = hash(p._obj)
- if cache:
- try:
- p._obj._hash_cache_ = result
- except AttributeError:
- pass
- return result
+ assert self._T._gckind == 'gc'
+ assert self # not for NULL
+ return hash(p._obj)
class _ptr(_abstract_ptr):
__slots__ = ('_TYPE',
@@ -1759,7 +1750,7 @@
class _struct(_parentable):
_kind = "structure"
- __slots__ = ('_hash_cache_', '_compilation_info')
+ __slots__ = ('_compilation_info',)
def __new__(self, TYPE, n=None, initialization=None, parent=None,
parentindex=None):
@@ -2442,24 +2433,6 @@
return SomeInteger()
-def identityhash_nocache(p):
- """Version of identityhash() to use from backends that don't care about
- caching."""
- assert p
- return p._identityhash(cache=False)
-
-def init_identity_hash(p, value):
- """For a prebuilt object p, initialize its hash value to 'value'."""
- assert isinstance(typeOf(p), Ptr)
- p = normalizeptr(p)
- if not p:
- raise ValueError("cannot change hash(NULL)!")
- if hasattr(p._obj, '_hash_cache_'):
- raise ValueError("the hash of %r was already computed" % (p,))
- if typeOf(p).TO._is_varsize():
- raise ValueError("init_identity_hash(): not for varsized types")
- p._obj._hash_cache_ = intmask(value)
-
def isCompatibleType(TYPE1, TYPE2):
return TYPE1._is_compatible(TYPE2)
diff --git a/rpython/rtyper/rclass.py b/rpython/rtyper/rclass.py
--- a/rpython/rtyper/rclass.py
+++ b/rpython/rtyper/rclass.py
@@ -785,7 +785,6 @@
def initialize_prebuilt_instance(self, value, classdef, result):
# must fill in the hash cache before the other ones
# (see test_circular_hash_initialization)
- self.initialize_prebuilt_hash(value, result)
self._initialize_data_flattenrec(self.initialize_prebuilt_data,
value, classdef, result)
@@ -943,11 +942,6 @@
rclass = getclassrepr(self.rtyper, classdef)
result.typeptr = rclass.getvtable()
- def initialize_prebuilt_hash(self, value, result):
- llattrvalue = getattr(value, '__precomputed_identity_hash', None)
- if llattrvalue is not None:
- lltype.init_identity_hash(result, llattrvalue)
-
def getfieldrepr(self, attr):
"""Return the repr used for the given attribute."""
if attr in self.fields:
diff --git a/rpython/translator/c/node.py b/rpython/translator/c/node.py
--- a/rpython/translator/c/node.py
+++ b/rpython/translator/c/node.py
@@ -623,52 +623,6 @@
assert not USESLOTS or '__dict__' not in dir(StructNode)
-class GcStructNodeWithHash(StructNode):
- # for the outermost level of nested structures, if it has a _hash_cache_.
- nodekind = 'struct'
- if USESLOTS:
- __slots__ = ()
-
- def get_hash_typename(self):
- return 'struct _hashT_%s @' % self.name
-
- def forward_declaration(self):
- T = self.getTYPE()
- assert self.typename == self.implementationtypename # no array part
- hash_typename = self.get_hash_typename()
- hash_offset = self.db.gctransformer.get_hash_offset(T)
- yield '%s {' % cdecl(hash_typename, '')
- yield '\tunion {'
- yield '\t\t%s;' % cdecl(self.implementationtypename, 'head')
- yield '\t\tchar pad[%s];' % name_signed(hash_offset, self.db)
- yield '\t} u;'
- yield '\tlong hash;'
- yield '};'
- yield '%s;' % (
- forward_cdecl(hash_typename, '_hash_' + self.name,
- self.db.standalone, self.is_thread_local()),)
- yield '#define %s _hash_%s.u.head' % (self.name, self.name)
-
- def implementation(self):
- hash_typename = self.get_hash_typename()
- hash = self.db.gctransformer.get_prebuilt_hash(self.obj)
- assert hash is not None
- lines = list(self.initializationexpr())
- lines.insert(0, '%s = { {' % (
- cdecl(hash_typename, '_hash_' + self.name,
- self.is_thread_local()),))
- lines.append('}, %s /* hash */ };' % name_signed(hash, self.db))
- return lines
-
-def gcstructnode_factory(db, T, obj):
- if (db.gctransformer and
- db.gctransformer.get_prebuilt_hash(obj) is not None):
- DISABLED
- cls = GcStructNodeWithHash
- else:
- cls = StructNode
- return cls(db, T, obj)
-
class ArrayNode(ContainerNode):
nodekind = 'array'
@@ -1058,7 +1012,7 @@
ContainerNodeFactory = {
Struct: StructNode,
- GcStruct: gcstructnode_factory,
+ GcStruct: StructNode,
Array: ArrayNode,
GcArray: ArrayNode,
FixedSizeArray: FixedSizeArrayNode,
_______________________________________________
pypy-commit mailing list
[email protected]
https://mail.python.org/mailman/listinfo/pypy-commit
