#
# Cython/Python language types
#
from __future__ import absolute_import
import copy
import hashlib
import re
try:
reduce
except NameError:
from functools import reduce
from functools import partial
from itertools import product
from Cython.Utils import cached_function
from .Code import UtilityCode, LazyUtilityCode, TempitaUtilityCode
from . import StringEncoding
from . import Naming
from .Errors import error, CannotSpecialize, performance_hint
class BaseType(object):
#
# Base class for all Cython types including pseudo-types.
# List of attribute names of any subtypes
subtypes = []
_empty_declaration = None
_specialization_name = None
default_format_spec = None
def can_coerce_to_pyobject(self, env):
return False
def can_coerce_from_pyobject(self, env):
return False
def can_coerce_to_pystring(self, env, format_spec=None):
return False
def convert_to_pystring(self, cvalue, code, format_spec=None):
raise NotImplementedError("C types that support string formatting must override this method")
def cast_code(self, expr_code):
return "((%s)%s)" % (self.empty_declaration_code(), expr_code)
def empty_declaration_code(self, pyrex=False):
if pyrex:
return self.declaration_code('', pyrex=True)
if self._empty_declaration is None:
self._empty_declaration = self.declaration_code('')
return self._empty_declaration
def specialization_name(self):
if self._specialization_name is None:
# This is not entirely robust.
common_subs = (self.empty_declaration_code()
# covers both "unsigned " and "signed "
.replace("signed ", "signed_")
.replace("long long", "long_long")
.replace(" ", "__"))
self._specialization_name = re.sub(
'[^a-zA-Z0-9_]', lambda x: '_%x_' % ord(x.group(0)), common_subs)
return self._specialization_name
def base_declaration_code(self, base_code, entity_code):
if entity_code:
return "%s %s" % (base_code, entity_code)
else:
return base_code
def __deepcopy__(self, memo):
"""
Types never need to be copied, if we do copy, Unfortunate Things
Will Happen!
"""
return self
def get_fused_types(self, result=None, seen=None, subtypes=None, include_function_return_type=False):
subtypes = subtypes or self.subtypes
if not subtypes:
return None
if result is None:
result = []
seen = set()
for attr in subtypes:
list_or_subtype = getattr(self, attr)
if list_or_subtype:
if isinstance(list_or_subtype, BaseType):
list_or_subtype.get_fused_types(result, seen, include_function_return_type=include_function_return_type)
else:
for subtype in list_or_subtype:
subtype.get_fused_types(result, seen, include_function_return_type=include_function_return_type)
return result
def specialize_fused(self, env):
if env.fused_to_specific:
return self.specialize(env.fused_to_specific)
return self
@property
def is_fused(self):
"""
Whether this type or any of its subtypes is a fused type
"""
# Add this indirection for the is_fused property to allow overriding
# get_fused_types in subclasses.
return self.get_fused_types()
def deduce_template_params(self, actual):
"""
Deduce any template params in this (argument) type given the actual
argument type.
https://en.cppreference.com/w/cpp/language/function_template#Template_argument_deduction
"""
return {}
def __lt__(self, other):
"""
For sorting. The sorting order should correspond to the preference of
conversion from Python types.
Override to provide something sensible. This is only implemented so that
python 3 doesn't trip
"""
return id(type(self)) < id(type(other))
def py_type_name(self):
"""
Return the name of the Python type that can coerce to this type.
"""
def typeof_name(self):
"""
Return the string with which fused python functions can be indexed.
"""
if self.is_builtin_type or self.py_type_name() == 'object':
index_name = self.py_type_name()
else:
index_name = str(self)
return index_name
def check_for_null_code(self, cname):
"""
Return the code for a NULL-check in case an UnboundLocalError should
be raised if an entry of this type is referenced before assignment.
Returns None if no check should be performed.
"""
return None
def invalid_value(self):
"""
Returns the most invalid value an object of this type can assume as a
C expression string. Returns None if no such value exists.
"""
class PyrexType(BaseType):
#
# Base class for all Cython types
#
# is_pyobject boolean Is a Python object type
# is_extension_type boolean Is a Python extension type
# is_final_type boolean Is a final extension type
# is_numeric boolean Is a C numeric type
# is_int boolean Is a C integer type
# is_float boolean Is a C floating point type
# is_complex boolean Is a C complex type
# is_void boolean Is the C void type
# is_array boolean Is a C array type
# is_ptr boolean Is a C pointer type
# is_null_ptr boolean Is the type of NULL
# is_reference boolean Is a C reference type
# is_rvalue_reference boolean Is a C++ rvalue reference type
# is_const boolean Is a C const type
# is_volatile boolean Is a C volatile type
# is_cv_qualified boolean Is a C const or volatile type
# is_cfunction boolean Is a C function type
# is_struct_or_union boolean Is a C struct or union type
# is_struct boolean Is a C struct type
# is_cpp_class boolean Is a C++ class
# is_optional_cpp_class boolean Is a C++ class with variable lifetime handled with std::optional
# is_enum boolean Is a C enum type
# is_cpp_enum boolean Is a C++ scoped enum type
# is_typedef boolean Is a typedef type
# is_string boolean Is a C char * type
# is_pyunicode_ptr boolean Is a C PyUNICODE * type
# is_cpp_string boolean Is a C++ std::string type
# python_type_constructor_name string or None non-None if it is a Python type constructor that can be indexed/"templated"
# is_unicode_char boolean Is either Py_UCS4 or Py_UNICODE
# is_returncode boolean Is used only to signal exceptions
# is_error boolean Is the dummy error type
# is_buffer boolean Is buffer access type
# is_pythran_expr boolean Is Pythran expr
# is_numpy_buffer boolean Is Numpy array buffer
# has_attributes boolean Has C dot-selectable attributes
# needs_cpp_construction boolean Needs C++ constructor and destructor when used in a cdef class
# needs_refcounting boolean Needs code to be generated similar to incref/gotref/decref.
# Largely used internally.
# refcounting_needs_gil boolean Reference counting needs GIL to be acquired.
# equivalent_type type A C or Python type that is equivalent to this Python or C type.
# default_value string Initial value that can be assigned before first user assignment.
# declaration_value string The value statically assigned on declaration (if any).
# entry Entry The Entry for this type
#
# declaration_code(entity_code,
# for_display = 0, dll_linkage = None, pyrex = 0)
# Returns a code fragment for the declaration of an entity
# of this type, given a code fragment for the entity.
# * If for_display, this is for reading by a human in an error
# message; otherwise it must be valid C code.
# * If dll_linkage is not None, it must be 'DL_EXPORT' or
# 'DL_IMPORT', and will be added to the base type part of
# the declaration.
# * If pyrex = 1, this is for use in a 'cdef extern'
# statement of a Cython include file.
#
# assignable_from(src_type)
# Tests whether a variable of this type can be
# assigned a value of type src_type.
#
# same_as(other_type)
# Tests whether this type represents the same type
# as other_type.
#
# as_argument_type():
# Coerces array and C function types into pointer type for use as
# a formal argument type.
#
is_pyobject = 0
is_unspecified = 0
is_extension_type = 0
is_final_type = 0
is_builtin_type = 0
is_cython_builtin_type = 0
is_numeric = 0
is_int = 0
is_float = 0
is_complex = 0
is_void = 0
is_array = 0
is_ptr = 0
is_null_ptr = 0
is_reference = 0
is_fake_reference = 0
is_rvalue_reference = 0
is_const = 0
is_volatile = 0
is_cv_qualified = 0
is_cfunction = 0
is_struct_or_union = 0
is_cpp_class = 0
is_optional_cpp_class = 0
python_type_constructor_name = None
is_cpp_string = 0
is_struct = 0
is_enum = 0
is_cpp_enum = False
is_typedef = 0
is_string = 0
is_pyunicode_ptr = 0
is_unicode_char = 0
is_returncode = 0
is_error = 0
is_buffer = 0
is_ctuple = 0
is_memoryviewslice = 0
is_pythran_expr = 0
is_numpy_buffer = 0
has_attributes = 0
needs_cpp_construction = 0
needs_refcounting = 0
refcounting_needs_gil = True
equivalent_type = None
default_value = ""
declaration_value = ""
def resolve(self):
# If a typedef, returns the base type.
return self
def specialize(self, values):
# Returns the concrete type if this is a fused type, or otherwise the type itself.
# May raise Errors.CannotSpecialize on failure
return self
def literal_code(self, value):
# Returns a C code fragment representing a literal
# value of this type.
return str(value)
def __str__(self):
return self.declaration_code("", for_display = 1).strip()
def same_as(self, other_type, **kwds):
return self.same_as_resolved_type(other_type.resolve(), **kwds)
def same_as_resolved_type(self, other_type):
return self == other_type or other_type is error_type
def subtype_of(self, other_type):
return self.subtype_of_resolved_type(other_type.resolve())
def subtype_of_resolved_type(self, other_type):
return self.same_as(other_type)
def assignable_from(self, src_type):
return self.assignable_from_resolved_type(src_type.resolve())
def assignable_from_resolved_type(self, src_type):
return self.same_as(src_type)
def assignment_failure_extra_info(self, src_type, src_name):
"""Override if you can provide useful extra information about why an assignment didn't work.
src_name may be None if unavailable"""
return ""
def as_argument_type(self):
return self
def is_complete(self):
# A type is incomplete if it is an unsized array,
# a struct whose attributes are not defined, etc.
return 1
def is_simple_buffer_dtype(self):
return False
def can_be_optional(self):
"""Returns True if type can be used with typing.Optional[]."""
return False
def struct_nesting_depth(self):
# Returns the number levels of nested structs. This is
# used for constructing a stack for walking the run-time
# type information of the struct.
return 1
def global_init_code(self, entry, code):
# abstract
pass
def needs_nonecheck(self):
return 0
def _assign_from_py_code(self, source_code, result_code, error_pos, code,
from_py_function=None, error_condition=None, extra_args=None,
special_none_cvalue=None):
args = ', ' + ', '.join('%s' % arg for arg in extra_args) if extra_args else ''
convert_call = "%s(%s%s)" % (
from_py_function or self.from_py_function,
source_code,
args,
)
if self.is_enum:
convert_call = typecast(self, c_long_type, convert_call)
if special_none_cvalue:
# NOTE: requires 'source_code' to be simple!
convert_call = "(__Pyx_Py_IsNone(%s) ? (%s) : (%s))" % (
source_code, special_none_cvalue, convert_call)
return '%s = %s; %s' % (
result_code,
convert_call,
code.error_goto_if(error_condition or self.error_condition(result_code), error_pos))
def _generate_dummy_refcounting(self, code, *ignored_args, **ignored_kwds):
if self.needs_refcounting:
raise NotImplementedError("Ref-counting operation not yet implemented for type %s" %
self)
def _generate_dummy_refcounting_assignment(self, code, cname, rhs_cname, *ignored_args, **ignored_kwds):
if self.needs_refcounting:
raise NotImplementedError("Ref-counting operation not yet implemented for type %s" %
self)
code.putln("%s = %s" % (cname, rhs_cname))
generate_incref = generate_xincref = generate_decref = generate_xdecref \
= generate_decref_clear = generate_xdecref_clear \
= generate_gotref = generate_xgotref = generate_giveref = generate_xgiveref \
= _generate_dummy_refcounting
generate_decref_set = generate_xdecref_set = _generate_dummy_refcounting_assignment
def nullcheck_string(self, code, cname):
if self.needs_refcounting:
raise NotImplementedError("Ref-counting operation not yet implemented for type %s" %
self)
code.putln("1")
def cpp_optional_declaration_code(self, entity_code, dll_linkage=None):
# declares an std::optional c++ variable
raise NotImplementedError(
"cpp_optional_declaration_code only implemented for c++ classes and not type %s" % self)
def public_decl(base_code, dll_linkage):
if dll_linkage:
return "%s(%s)" % (dll_linkage, base_code.replace(',', ' __PYX_COMMA '))
else:
return base_code
def create_typedef_type(name, base_type, cname, is_external=0, namespace=None):
if is_external:
if base_type.is_complex or base_type.is_fused:
raise ValueError("%s external typedefs not supported" % (
"Fused" if base_type.is_fused else "Complex"))
if base_type.is_complex or base_type.is_fused:
return base_type
return CTypedefType(name, base_type, cname, is_external, namespace)
class CTypedefType(BaseType):
#
# Pseudo-type defined with a ctypedef statement in a
# 'cdef extern from' block.
# Delegates most attribute lookups to the base type.
# (Anything not defined here or in the BaseType is delegated.)
#
# qualified_name string
# typedef_name string
# typedef_cname string
# typedef_base_type PyrexType
# typedef_is_external bool
is_typedef = 1
typedef_is_external = 0
to_py_utility_code = None
from_py_utility_code = None
subtypes = ['typedef_base_type']
def __init__(self, name, base_type, cname, is_external=0, namespace=None):
assert not base_type.is_complex
self.typedef_name = name
self.typedef_cname = cname
self.typedef_base_type = base_type
self.typedef_is_external = is_external
self.typedef_namespace = namespace
def invalid_value(self):
return self.typedef_base_type.invalid_value()
def resolve(self):
return self.typedef_base_type.resolve()
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if pyrex or for_display:
base_code = self.typedef_name
else:
base_code = public_decl(self.typedef_cname, dll_linkage)
if self.typedef_namespace is not None and not pyrex:
base_code = "%s::%s" % (self.typedef_namespace.empty_declaration_code(), base_code)
return self.base_declaration_code(base_code, entity_code)
def as_argument_type(self):
return self
def cast_code(self, expr_code):
# If self is really an array (rather than pointer), we can't cast.
# For example, the gmp mpz_t.
if self.typedef_base_type.is_array:
base_type = self.typedef_base_type.base_type
return CPtrType(base_type).cast_code(expr_code)
else:
return BaseType.cast_code(self, expr_code)
def specialize(self, values):
base_type = self.typedef_base_type.specialize(values)
namespace = self.typedef_namespace.specialize(values) if self.typedef_namespace else None
if base_type is self.typedef_base_type and namespace is self.typedef_namespace:
return self
else:
return create_typedef_type(self.typedef_name, base_type, self.typedef_cname,
0, namespace)
def __repr__(self):
return "<CTypedefType %s>" % self.typedef_cname
def __str__(self):
return self.typedef_name
def _create_utility_code(self, template_utility_code,
template_function_name):
type_name = type_identifier(self.typedef_cname)
utility_code = template_utility_code.specialize(
type = self.typedef_cname,
TypeName = type_name)
function_name = template_function_name % type_name
return utility_code, function_name
def create_to_py_utility_code(self, env):
if self.typedef_is_external:
if not self.to_py_utility_code:
base_type = self.typedef_base_type
if type(base_type) is CIntType:
self.to_py_function = "__Pyx_PyInt_From_" + self.specialization_name()
env.use_utility_code(TempitaUtilityCode.load_cached(
"CIntToPy", "TypeConversion.c",
context={"TYPE": self.empty_declaration_code(),
"TO_PY_FUNCTION": self.to_py_function}))
return True
elif base_type.is_float:
pass # XXX implement!
elif base_type.is_complex:
pass # XXX implement!
pass
elif base_type.is_cpp_string:
cname = "__pyx_convert_PyObject_string_to_py_%s" % type_identifier(self)
context = {
'cname': cname,
'type': self.typedef_cname,
}
from .UtilityCode import CythonUtilityCode
env.use_utility_code(CythonUtilityCode.load(
"string.to_py", "CppConvert.pyx", context=context))
self.to_py_function = cname
return True
if self.to_py_utility_code:
env.use_utility_code(self.to_py_utility_code)
return True
# delegation
return self.typedef_base_type.create_to_py_utility_code(env)
def create_from_py_utility_code(self, env):
if self.typedef_is_external:
if not self.from_py_utility_code:
base_type = self.typedef_base_type
if type(base_type) is CIntType:
self.from_py_function = "__Pyx_PyInt_As_" + self.specialization_name()
env.use_utility_code(TempitaUtilityCode.load_cached(
"CIntFromPy", "TypeConversion.c",
context={
"TYPE": self.empty_declaration_code(),
"FROM_PY_FUNCTION": self.from_py_function,
"IS_ENUM": base_type.is_enum,
}))
return True
elif base_type.is_float:
pass # XXX implement!
elif base_type.is_complex:
pass # XXX implement!
elif base_type.is_cpp_string:
cname = '__pyx_convert_string_from_py_%s' % type_identifier(self)
context = {
'cname': cname,
'type': self.typedef_cname,
}
from .UtilityCode import CythonUtilityCode
env.use_utility_code(CythonUtilityCode.load(
"string.from_py", "CppConvert.pyx", context=context))
self.from_py_function = cname
return True
if self.from_py_utility_code:
env.use_utility_code(self.from_py_utility_code)
return True
# delegation
return self.typedef_base_type.create_from_py_utility_code(env)
def to_py_call_code(self, source_code, result_code, result_type, to_py_function=None):
if to_py_function is None:
to_py_function = self.to_py_function
return self.typedef_base_type.to_py_call_code(
source_code, result_code, result_type, to_py_function)
def from_py_call_code(self, source_code, result_code, error_pos, code,
from_py_function=None, error_condition=None,
special_none_cvalue=None):
return self.typedef_base_type.from_py_call_code(
source_code, result_code, error_pos, code,
from_py_function or self.from_py_function,
error_condition or self.error_condition(result_code),
special_none_cvalue=special_none_cvalue,
)
def overflow_check_binop(self, binop, env, const_rhs=False):
env.use_utility_code(UtilityCode.load("Common", "Overflow.c"))
type = self.empty_declaration_code()
name = self.specialization_name()
if binop == "lshift":
env.use_utility_code(TempitaUtilityCode.load_cached(
"LeftShift", "Overflow.c",
context={'TYPE': type, 'NAME': name, 'SIGNED': self.signed}))
else:
if const_rhs:
binop += "_const"
_load_overflow_base(env)
env.use_utility_code(TempitaUtilityCode.load_cached(
"SizeCheck", "Overflow.c",
context={'TYPE': type, 'NAME': name}))
env.use_utility_code(TempitaUtilityCode.load_cached(
"Binop", "Overflow.c",
context={'TYPE': type, 'NAME': name, 'BINOP': binop}))
return "__Pyx_%s_%s_checking_overflow" % (binop, name)
def error_condition(self, result_code):
if self.typedef_is_external:
if self.exception_value:
condition = "(%s == %s)" % (
result_code, self.cast_code(self.exception_value))
if self.exception_check:
condition += " && PyErr_Occurred()"
return condition
# delegation
return self.typedef_base_type.error_condition(result_code)
def __getattr__(self, name):
return getattr(self.typedef_base_type, name)
def py_type_name(self):
return self.typedef_base_type.py_type_name()
def can_coerce_to_pyobject(self, env):
return self.typedef_base_type.can_coerce_to_pyobject(env)
def can_coerce_from_pyobject(self, env):
return self.typedef_base_type.can_coerce_from_pyobject(env)
class MemoryViewSliceType(PyrexType):
is_memoryviewslice = 1
default_value = "{ 0, 0, { 0 }, { 0 }, { 0 } }"
has_attributes = 1
needs_refcounting = 1 # Ideally this would be true and reference counting for
# memoryview and pyobject code could be generated in the same way.
# However, memoryviews are sufficiently specialized that this doesn't
# seem practical. Implement a limited version of it for now
refcounting_needs_gil = False # __PYX_XCLEAR_MEMVIEW acquires GIL internally.
scope = None
# These are special cased in Defnode
from_py_function = None
to_py_function = None
exception_value = None
exception_check = True
subtypes = ['dtype']
def __init__(self, base_dtype, axes):
"""
MemoryViewSliceType(base, axes)
Base is the C base type; axes is a list of (access, packing) strings,
where access is one of 'full', 'direct' or 'ptr' and packing is one of
'contig', 'strided' or 'follow'. There is one (access, packing) tuple
for each dimension.
the access specifiers determine whether the array data contains
pointers that need to be dereferenced along that axis when
retrieving/setting:
'direct' -- No pointers stored in this dimension.
'ptr' -- Pointer stored in this dimension.
'full' -- Check along this dimension, don't assume either.
the packing specifiers specify how the array elements are laid-out
in memory.
'contig' -- The data is contiguous in memory along this dimension.
At most one dimension may be specified as 'contig'.
'strided' -- The data isn't contiguous along this dimension.
'follow' -- Used for C/Fortran contiguous arrays, a 'follow' dimension
has its stride automatically computed from extents of the other
dimensions to ensure C or Fortran memory layout.
C-contiguous memory has 'direct' as the access spec, 'contig' as the
*last* axis' packing spec and 'follow' for all other packing specs.
Fortran-contiguous memory has 'direct' as the access spec, 'contig' as
the *first* axis' packing spec and 'follow' for all other packing
specs.
"""
from . import Buffer, MemoryView
self.dtype = base_dtype
self.axes = axes
self.ndim = len(axes)
self.flags = MemoryView.get_buf_flags(self.axes)
self.is_c_contig, self.is_f_contig = MemoryView.is_cf_contig(self.axes)
assert not (self.is_c_contig and self.is_f_contig)
self.mode = MemoryView.get_mode(axes)
self.writable_needed = False
if not self.dtype.is_fused:
self.dtype_name = Buffer.mangle_dtype_name(self.dtype)
def __hash__(self):
return hash(self.__class__) ^ hash(self.dtype) ^ hash(tuple(self.axes))
def __eq__(self, other):
if isinstance(other, BaseType):
return self.same_as_resolved_type(other)
else:
return False
def __ne__(self, other):
# TODO drop when Python2 is dropped
return not (self == other)
def same_as_resolved_type(self, other_type):
return ((other_type.is_memoryviewslice and
#self.writable_needed == other_type.writable_needed and # FIXME: should be only uni-directional
self.dtype.same_as(other_type.dtype) and
self.axes == other_type.axes) or
other_type is error_type)
def needs_nonecheck(self):
return True
def is_complete(self):
# incomplete since the underlying struct doesn't have a cython.memoryview object.
return 0
def can_be_optional(self):
"""Returns True if type can be used with typing.Optional[]."""
return True
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
# XXX: we put these guards in for now...
assert not dll_linkage
from . import MemoryView
base_code = StringEncoding.EncodedString(
str(self) if pyrex or for_display else MemoryView.memviewslice_cname)
return self.base_declaration_code(
base_code,
entity_code)
def attributes_known(self):
if self.scope is None:
from . import Symtab
self.scope = scope = Symtab.CClassScope(
'mvs_class_'+self.specialization_suffix(),
None,
visibility='extern',
parent_type=self)
scope.directives = {}
scope.declare_var('_data', c_char_ptr_type, None,
cname='data', is_cdef=1)
return True
def declare_attribute(self, attribute, env, pos):
from . import MemoryView, Options
scope = self.scope
if attribute == 'shape':
scope.declare_var('shape',
c_array_type(c_py_ssize_t_type,
Options.buffer_max_dims),
pos,
cname='shape',
is_cdef=1)
elif attribute == 'strides':
scope.declare_var('strides',
c_array_type(c_py_ssize_t_type,
Options.buffer_max_dims),
pos,
cname='strides',
is_cdef=1)
elif attribute == 'suboffsets':
scope.declare_var('suboffsets',
c_array_type(c_py_ssize_t_type,
Options.buffer_max_dims),
pos,
cname='suboffsets',
is_cdef=1)
elif attribute in ("copy", "copy_fortran"):
ndim = len(self.axes)
follow_dim = [('direct', 'follow')]
contig_dim = [('direct', 'contig')]
to_axes_c = follow_dim * (ndim - 1) + contig_dim
to_axes_f = contig_dim + follow_dim * (ndim -1)
dtype = self.dtype
if dtype.is_cv_qualified:
dtype = dtype.cv_base_type
to_memview_c = MemoryViewSliceType(dtype, to_axes_c)
to_memview_f = MemoryViewSliceType(dtype, to_axes_f)
for to_memview, cython_name in [(to_memview_c, "copy"),
(to_memview_f, "copy_fortran")]:
copy_func_type = CFuncType(
to_memview,
[CFuncTypeArg("memviewslice", self, None)])
copy_cname = MemoryView.copy_c_or_fortran_cname(to_memview)
entry = scope.declare_cfunction(
cython_name,
copy_func_type, pos=pos, defining=1,
cname=copy_cname)
utility = MemoryView.get_copy_new_utility(pos, self, to_memview)
env.use_utility_code(utility)
MemoryView.use_cython_array_utility_code(env)
elif attribute in ("is_c_contig", "is_f_contig"):
# is_c_contig and is_f_contig functions
for (c_or_f, cython_name) in (('C', 'is_c_contig'), ('F', 'is_f_contig')):
is_contig_name = MemoryView.get_is_contig_func_name(c_or_f, self.ndim)
cfunctype = CFuncType(
return_type=c_bint_type,
args=[CFuncTypeArg("memviewslice", self, None)],
exception_value="-1",
)
entry = scope.declare_cfunction(cython_name,
cfunctype,
pos=pos,
defining=1,
cname=is_contig_name)
entry.utility_code_definition = MemoryView.get_is_contig_utility(c_or_f, self.ndim)
return True
def get_entry(self, node, cname=None, type=None):
from . import MemoryView, Symtab
if cname is None:
assert node.is_simple() or node.is_temp or node.is_elemental
cname = node.result()
if type is None:
type = node.type
entry = Symtab.Entry(cname, cname, type, node.pos)
return MemoryView.MemoryViewSliceBufferEntry(entry)
def conforms_to(self, dst, broadcast=False, copying=False):
"""
Returns True if src conforms to dst, False otherwise.
If conformable, the types are the same, the ndims are equal, and each axis spec is conformable.
Any packing/access spec is conformable to itself.
'direct' and 'ptr' are conformable to 'full'.
'contig' and 'follow' are conformable to 'strided'.
Any other combo is not conformable.
"""
from . import MemoryView
src = self
#if not copying and self.writable_needed and not dst.writable_needed:
# return False
src_dtype, dst_dtype = src.dtype, dst.dtype
# We can add but not remove const/volatile modifiers
# (except if we are copying by value, then anything is fine)
if not copying:
if src_dtype.is_const and not dst_dtype.is_const:
return False
if src_dtype.is_volatile and not dst_dtype.is_volatile:
return False
# const/volatile checks are done, remove those qualifiers
if src_dtype.is_cv_qualified:
src_dtype = src_dtype.cv_base_type
if dst_dtype.is_cv_qualified:
dst_dtype = dst_dtype.cv_base_type
if not src_dtype.same_as(dst_dtype):
return False
if src.ndim != dst.ndim:
if broadcast:
src, dst = MemoryView.broadcast_types(src, dst)
else:
return False
for src_spec, dst_spec in zip(src.axes, dst.axes):
src_access, src_packing = src_spec
dst_access, dst_packing = dst_spec
if src_access != dst_access and dst_access != 'full':
return False
if src_packing != dst_packing and dst_packing != 'strided' and not copying:
return False
return True
def valid_dtype(self, dtype, i=0):
"""
Return whether type dtype can be used as the base type of a
memoryview slice.
We support structs, numeric types and objects
"""
if dtype.is_complex and dtype.real_type.is_int:
return False
if dtype.is_struct and dtype.kind == 'struct':
for member in dtype.scope.var_entries:
if not self.valid_dtype(member.type):
return False
return True
return (
dtype.is_error or
# Pointers are not valid (yet)
# (dtype.is_ptr and valid_memslice_dtype(dtype.base_type)) or
(dtype.is_array and i < 8 and self.valid_dtype(dtype.base_type, i + 1)) or
dtype.is_numeric or
dtype.is_pyobject or
dtype.is_fused or # accept this as it will be replaced by specializations later
(dtype.is_typedef and self.valid_dtype(dtype.typedef_base_type))
)
def validate_memslice_dtype(self, pos):
if not self.valid_dtype(self.dtype):
error(pos, "Invalid base type for memoryview slice: %s" % self.dtype)
def assert_direct_dims(self, pos):
for access, packing in self.axes:
if access != 'direct':
error(pos, "All dimensions must be direct")
return False
return True
def transpose(self, pos):
if not self.assert_direct_dims(pos):
return error_type
return MemoryViewSliceType(self.dtype, self.axes[::-1])
def specialization_name(self):
return '%s_%s' % (
super(MemoryViewSliceType,self).specialization_name(),
self.specialization_suffix())
def specialization_suffix(self):
return "%s_%s" % (self.axes_to_name(), self.dtype_name)
def can_coerce_to_pyobject(self, env):
return True
def can_coerce_from_pyobject(self, env):
return True
def check_for_null_code(self, cname):
return cname + '.memview'
def create_from_py_utility_code(self, env):
from . import MemoryView, Buffer
# We don't have 'code', so use a LazyUtilityCode with a callback.
def lazy_utility_callback(code):
context['dtype_typeinfo'] = Buffer.get_type_information_cname(code, self.dtype)
return TempitaUtilityCode.load(
"ObjectToMemviewSlice", "MemoryView_C.c", context=context)
env.use_utility_code(MemoryView.memviewslice_init_code)
env.use_utility_code(LazyUtilityCode(lazy_utility_callback))
if self.is_c_contig:
c_or_f_flag = "__Pyx_IS_C_CONTIG"
elif self.is_f_contig:
c_or_f_flag = "__Pyx_IS_F_CONTIG"
else:
c_or_f_flag = "0"
suffix = self.specialization_suffix()
funcname = "__Pyx_PyObject_to_MemoryviewSlice_" + suffix
context = dict(
MemoryView.context,
buf_flag = self.flags,
ndim = self.ndim,
axes_specs = ', '.join(self.axes_to_code()),
dtype_typedecl = self.dtype.empty_declaration_code(),
struct_nesting_depth = self.dtype.struct_nesting_depth(),
c_or_f_flag = c_or_f_flag,
funcname = funcname,
)
self.from_py_function = funcname
return True
def from_py_call_code(self, source_code, result_code, error_pos, code,
from_py_function=None, error_condition=None,
special_none_cvalue=None):
# NOTE: auto-detection of readonly buffers is disabled:
# writable = self.writable_needed or not self.dtype.is_const
writable = not self.dtype.is_const
return self._assign_from_py_code(
source_code, result_code, error_pos, code, from_py_function, error_condition,
extra_args=['PyBUF_WRITABLE' if writable else '0'],
special_none_cvalue=special_none_cvalue,
)
def create_to_py_utility_code(self, env):
self._dtype_to_py_func, self._dtype_from_py_func = self.dtype_object_conversion_funcs(env)
return True
def to_py_call_code(self, source_code, result_code, result_type, to_py_function=None):
assert self._dtype_to_py_func
assert self._dtype_from_py_func
to_py_func = "(PyObject *(*)(char *)) " + self._dtype_to_py_func
from_py_func = "(int (*)(char *, PyObject *)) " + self._dtype_from_py_func
tup = (result_code, source_code, self.ndim, to_py_func, from_py_func, self.dtype.is_pyobject)
return "%s = __pyx_memoryview_fromslice(%s, %s, %s, %s, %d);" % tup
def dtype_object_conversion_funcs(self, env):
get_function = "__pyx_memview_get_%s" % self.dtype_name
set_function = "__pyx_memview_set_%s" % self.dtype_name
context = dict(
get_function = get_function,
set_function = set_function,
)
if self.dtype.is_pyobject:
utility_name = "MemviewObjectToObject"
else:
self.dtype.create_to_py_utility_code(env)
to_py_function = self.dtype.to_py_function
from_py_function = None
if not self.dtype.is_const:
self.dtype.create_from_py_utility_code(env)
from_py_function = self.dtype.from_py_function
if not (to_py_function or from_py_function):
return "NULL", "NULL"
if not to_py_function:
get_function = "NULL"
if not from_py_function:
set_function = "NULL"
utility_name = "MemviewDtypeToObject"
error_condition = (self.dtype.error_condition('value') or
'PyErr_Occurred()')
context.update(
to_py_function=to_py_function,
from_py_function=from_py_function,
dtype=self.dtype.empty_declaration_code(),
error_condition=error_condition,
)
utility = TempitaUtilityCode.load_cached(
utility_name, "MemoryView_C.c", context=context)
env.use_utility_code(utility)
return get_function, set_function
def axes_to_code(self):
"""Return a list of code constants for each axis"""
from . import MemoryView
d = MemoryView._spec_to_const
return ["(%s | %s)" % (d[a], d[p]) for a, p in self.axes]
def axes_to_name(self):
"""Return an abbreviated name for our axes"""
from . import MemoryView
d = MemoryView._spec_to_abbrev
return "".join(["%s%s" % (d[a], d[p]) for a, p in self.axes])
def error_condition(self, result_code):
return "!%s.memview" % result_code
def __str__(self):
from . import MemoryView
axes_code_list = []
for idx, (access, packing) in enumerate(self.axes):
flag = MemoryView.get_memoryview_flag(access, packing)
if flag == "strided":
axes_code_list.append(":")
else:
if flag == 'contiguous':
have_follow = [p for a, p in self.axes[idx - 1:idx + 2]
if p == 'follow']
if have_follow or self.ndim == 1:
flag = '1'
axes_code_list.append("::" + flag)
if self.dtype.is_pyobject:
dtype_name = self.dtype.name
else:
dtype_name = self.dtype
return "%s[%s]" % (dtype_name, ", ".join(axes_code_list))
def specialize(self, values):
"""This does not validate the base type!!"""
dtype = self.dtype.specialize(values)
if dtype is not self.dtype:
return MemoryViewSliceType(dtype, self.axes)
return self
def cast_code(self, expr_code):
return expr_code
# When memoryviews are increfed currently seems heavily special-cased.
# Therefore, use our own function for now
def generate_incref(self, code, name, **kwds):
pass
def generate_incref_memoryviewslice(self, code, slice_cname, have_gil):
# TODO ideally would be done separately
code.putln("__PYX_INC_MEMVIEW(&%s, %d);" % (slice_cname, int(have_gil)))
# decref however did look to always apply for memoryview slices
# with "have_gil" set to True by default
def generate_xdecref(self, code, cname, nanny, have_gil):
code.putln("__PYX_XCLEAR_MEMVIEW(&%s, %d);" % (cname, int(have_gil)))
def generate_decref(self, code, cname, nanny, have_gil):
# Fall back to xdecref since we don't care to have a separate decref version for this.
self.generate_xdecref(code, cname, nanny, have_gil)
def generate_xdecref_clear(self, code, cname, clear_before_decref, **kwds):
self.generate_xdecref(code, cname, **kwds)
code.putln("%s.memview = NULL; %s.data = NULL;" % (cname, cname))
def generate_decref_clear(self, code, cname, **kwds):
# memoryviews don't currently distinguish between xdecref and decref
self.generate_xdecref_clear(code, cname, **kwds)
# memoryviews don't participate in giveref/gotref
generate_gotref = generate_xgotref = generate_xgiveref = generate_giveref = lambda *args: None
class BufferType(BaseType):
#
# Delegates most attribute lookups to the base type.
# (Anything not defined here or in the BaseType is delegated.)
#
# dtype PyrexType
# ndim int
# mode str
# negative_indices bool
# cast bool
# is_buffer bool
# writable bool
is_buffer = 1
writable = True
subtypes = ['dtype']
def __init__(self, base, dtype, ndim, mode, negative_indices, cast):
self.base = base
self.dtype = dtype
self.ndim = ndim
self.buffer_ptr_type = CPtrType(dtype)
self.mode = mode
self.negative_indices = negative_indices
self.cast = cast
self.is_numpy_buffer = self.base.name == "ndarray"
def can_coerce_to_pyobject(self,env):
return True
def can_coerce_from_pyobject(self,env):
return True
def as_argument_type(self):
return self
def specialize(self, values):
dtype = self.dtype.specialize(values)
if dtype is not self.dtype:
return BufferType(self.base, dtype, self.ndim, self.mode,
self.negative_indices, self.cast)
return self
def get_entry(self, node):
from . import Buffer
assert node.is_name
return Buffer.BufferEntry(node.entry)
def __getattr__(self, name):
return getattr(self.base, name)
def __repr__(self):
return "<BufferType %r>" % self.base
def __str__(self):
# avoid ', ', as fused functions split the signature string on ', '
cast_str = ''
if self.cast:
cast_str = ',cast=True'
return "%s[%s,ndim=%d%s]" % (self.base, self.dtype, self.ndim,
cast_str)
def assignable_from(self, other_type):
if other_type.is_buffer:
return (self.same_as(other_type, compare_base=False) and
self.base.assignable_from(other_type.base))
return self.base.assignable_from(other_type)
def same_as(self, other_type, compare_base=True):
if not other_type.is_buffer:
return other_type.same_as(self.base)
return (self.dtype.same_as(other_type.dtype) and
self.ndim == other_type.ndim and
self.mode == other_type.mode and
self.cast == other_type.cast and
(not compare_base or self.base.same_as(other_type.base)))
class PyObjectType(PyrexType):
#
# Base class for all Python object types (reference-counted).
#
# buffer_defaults dict or None Default options for buffer
name = "object"
is_pyobject = 1
default_value = "0"
declaration_value = "0"
buffer_defaults = None
is_external = False
is_subclassed = False
is_gc_simple = False
builtin_trashcan = False # builtin type using trashcan
needs_refcounting = True
def __str__(self):
return "Python object"
def __repr__(self):
return "<PyObjectType>"
def can_coerce_to_pyobject(self, env):
return True
def can_coerce_from_pyobject(self, env):
return True
def can_be_optional(self):
"""Returns True if type can be used with typing.Optional[]."""
return True
def default_coerced_ctype(self):
"""The default C type that this Python type coerces to, or None."""
return None
def assignable_from(self, src_type):
# except for pointers, conversion will be attempted
return not src_type.is_ptr or src_type.is_string or src_type.is_pyunicode_ptr
def is_simple_buffer_dtype(self):
return True
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if pyrex or for_display:
base_code = "object"
else:
base_code = public_decl("PyObject", dll_linkage)
entity_code = "*%s" % entity_code
return self.base_declaration_code(base_code, entity_code)
def as_pyobject(self, cname):
if (not self.is_complete()) or self.is_extension_type:
return "(PyObject *)" + cname
else:
return cname
def py_type_name(self):
return "object"
def __lt__(self, other):
"""
Make sure we sort highest, as instance checking on py_type_name
('object') is always true
"""
return False
def global_init_code(self, entry, code):
code.put_init_var_to_py_none(entry, nanny=False)
def check_for_null_code(self, cname):
return cname
def generate_incref(self, code, cname, nanny):
if nanny:
code.funcstate.needs_refnanny = True
code.putln("__Pyx_INCREF(%s);" % self.as_pyobject(cname))
else:
code.putln("Py_INCREF(%s);" % self.as_pyobject(cname))
def generate_xincref(self, code, cname, nanny):
if nanny:
code.funcstate.needs_refnanny = True
code.putln("__Pyx_XINCREF(%s);" % self.as_pyobject(cname))
else:
code.putln("Py_XINCREF(%s);" % self.as_pyobject(cname))
def generate_decref(self, code, cname, nanny, have_gil):
# have_gil is for the benefit of memoryviewslice - it's ignored here
assert have_gil
self._generate_decref(code, cname, nanny, null_check=False, clear=False)
def generate_xdecref(self, code, cname, nanny, have_gil):
# in this (and other) PyObjectType functions, have_gil is being
# passed to provide a common interface with MemoryviewSlice.
# It's ignored here
self._generate_decref(code, cname, nanny, null_check=True,
clear=False)
def generate_decref_clear(self, code, cname, clear_before_decref, nanny, have_gil):
self._generate_decref(code, cname, nanny, null_check=False,
clear=True, clear_before_decref=clear_before_decref)
def generate_xdecref_clear(self, code, cname, clear_before_decref=False, nanny=True, have_gil=None):
self._generate_decref(code, cname, nanny, null_check=True,
clear=True, clear_before_decref=clear_before_decref)
def generate_gotref(self, code, cname):
code.funcstate.needs_refnanny = True
code.putln("__Pyx_GOTREF(%s);" % self.as_pyobject(cname))
def generate_xgotref(self, code, cname):
code.funcstate.needs_refnanny = True
code.putln("__Pyx_XGOTREF(%s);" % self.as_pyobject(cname))
def generate_giveref(self, code, cname):
code.funcstate.needs_refnanny = True
code.putln("__Pyx_GIVEREF(%s);" % self.as_pyobject(cname))
def generate_xgiveref(self, code, cname):
code.funcstate.needs_refnanny = True
code.putln("__Pyx_XGIVEREF(%s);" % self.as_pyobject(cname))
def generate_decref_set(self, code, cname, rhs_cname):
code.funcstate.needs_refnanny = True
code.putln("__Pyx_DECREF_SET(%s, %s);" % (cname, rhs_cname))
def generate_xdecref_set(self, code, cname, rhs_cname):
code.funcstate.needs_refnanny = True
code.putln("__Pyx_XDECREF_SET(%s, %s);" % (cname, rhs_cname))
def _generate_decref(self, code, cname, nanny, null_check=False,
clear=False, clear_before_decref=False):
prefix = '__Pyx' if nanny else 'Py'
X = 'X' if null_check else ''
if nanny:
code.funcstate.needs_refnanny = True
if clear:
if clear_before_decref:
if not nanny:
X = '' # CPython doesn't have a Py_XCLEAR()
code.putln("%s_%sCLEAR(%s);" % (prefix, X, cname))
else:
code.putln("%s_%sDECREF(%s); %s = 0;" % (
prefix, X, self.as_pyobject(cname), cname))
else:
code.putln("%s_%sDECREF(%s);" % (
prefix, X, self.as_pyobject(cname)))
def nullcheck_string(self, cname):
return cname
builtin_types_that_cannot_create_refcycles = frozenset({
'object', 'bool', 'int', 'long', 'float', 'complex',
'bytearray', 'bytes', 'unicode', 'str', 'basestring',
})
builtin_types_with_trashcan = frozenset({
'dict', 'list', 'set', 'frozenset', 'tuple', 'type',
})
class BuiltinObjectType(PyObjectType):
# objstruct_cname string Name of PyObject struct
is_builtin_type = 1
has_attributes = 1
base_type = None
module_name = '__builtin__'
require_exact = 1
# fields that let it look like an extension type
vtabslot_cname = None
vtabstruct_cname = None
vtabptr_cname = None
typedef_flag = True
is_external = True
decl_type = 'PyObject'
def __init__(self, name, cname, objstruct_cname=None):
self.name = name
self.cname = cname
self.typeptr_cname = "(&%s)" % cname
self.objstruct_cname = objstruct_cname
self.is_gc_simple = name in builtin_types_that_cannot_create_refcycles
self.builtin_trashcan = name in builtin_types_with_trashcan
if name == 'type':
# Special case the type type, as many C API calls (and other
# libraries) actually expect a PyTypeObject* for type arguments.
self.decl_type = objstruct_cname
if name == 'Exception':
self.require_exact = 0
def set_scope(self, scope):
self.scope = scope
if scope:
scope.parent_type = self
def __str__(self):
return "%s object" % self.name
def __repr__(self):
return "<%s>"% self.cname
def default_coerced_ctype(self):
if self.name in ('bytes', 'bytearray'):
return c_char_ptr_type
elif self.name == 'bool':
return c_bint_type
elif self.name == 'float':
return c_double_type
return None
def assignable_from(self, src_type):
if isinstance(src_type, BuiltinObjectType):
if self.name == 'basestring':
return src_type.name in ('str', 'unicode', 'basestring')
else:
return src_type.name == self.name
elif src_type.is_extension_type:
# FIXME: This is an ugly special case that we currently
# keep supporting. It allows users to specify builtin
# types as external extension types, while keeping them
# compatible with the real builtin types. We already
# generate a warning for it. Big TODO: remove!
return (src_type.module_name == '__builtin__' and
src_type.name == self.name)
else:
return True
def typeobj_is_available(self):
return True
def attributes_known(self):
return True
def subtype_of(self, type):
return type.is_pyobject and type.assignable_from(self)
def type_check_function(self, exact=True):
type_name = self.name
if type_name == 'str':
type_check = 'PyString_Check'
elif type_name == 'basestring':
type_check = '__Pyx_PyBaseString_Check'
elif type_name == 'Exception':
type_check = '__Pyx_PyException_Check'
elif type_name == 'bytearray':
type_check = 'PyByteArray_Check'
elif type_name == 'frozenset':
type_check = 'PyFrozenSet_Check'
elif type_name == 'int':
# For backwards compatibility of (Py3) 'x: int' annotations in Py2, we also allow 'long' there.
type_check = '__Pyx_Py3Int_Check'
elif type_name == "memoryview":
# captialize doesn't catch the 'V'
type_check = "PyMemoryView_Check"
else:
type_check = 'Py%s_Check' % type_name.capitalize()
if exact and type_name not in ('bool', 'slice', 'Exception', 'memoryview'):
type_check += 'Exact'
return type_check
def isinstance_code(self, arg):
return '%s(%s)' % (self.type_check_function(exact=False), arg)
def type_test_code(self, arg, notnone=False, exact=True):
type_check = self.type_check_function(exact=exact)
check = 'likely(%s(%s))' % (type_check, arg)
if not notnone:
check += '||((%s) == Py_None)' % arg
if self.name == 'basestring':
name = '(PY_MAJOR_VERSION < 3 ? "basestring" : "str")'
else:
name = '"%s"' % self.name
return check + ' || __Pyx_RaiseUnexpectedTypeError(%s, %s)' % (name, arg)
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if pyrex or for_display:
base_code = self.name
else:
base_code = public_decl(self.decl_type, dll_linkage)
entity_code = "*%s" % entity_code
return self.base_declaration_code(base_code, entity_code)
def as_pyobject(self, cname):
if self.decl_type == 'PyObject':
return cname
else:
return "(PyObject *)" + cname
def cast_code(self, expr_code, to_object_struct = False):
return "((%s*)%s)" % (
to_object_struct and self.objstruct_cname or self.decl_type, # self.objstruct_cname may be None
expr_code)
def py_type_name(self):
return self.name
class PyExtensionType(PyObjectType):
#
# A Python extension type.
#
# name string
# scope CClassScope Attribute namespace
# typedef_flag boolean
# base_type PyExtensionType or None
# module_name string or None Qualified name of defining module
# objstruct_cname string Name of PyObject struct
# objtypedef_cname string Name of PyObject struct typedef
# typeobj_cname string or None C code fragment referring to type object
# typeptr_cname string or None Name of pointer to external type object
# vtabslot_cname string Name of C method table member
# vtabstruct_cname string Name of C method table struct
# vtabptr_cname string Name of pointer to C method table
# vtable_cname string Name of C method table definition
# early_init boolean Whether to initialize early (as opposed to during module execution).
# defered_declarations [thunk] Used to declare class hierarchies in order
# is_external boolean Defined in a extern block
# check_size 'warn', 'error', 'ignore' What to do if tp_basicsize does not match
# dataclass_fields OrderedDict nor None Used for inheriting from dataclasses
# multiple_bases boolean Does this class have multiple bases
# has_sequence_flag boolean Set Py_TPFLAGS_SEQUENCE
is_extension_type = 1
has_attributes = 1
early_init = 1
objtypedef_cname = None
dataclass_fields = None
multiple_bases = False
has_sequence_flag = False
def __init__(self, name, typedef_flag, base_type, is_external=0, check_size=None):
self.name = name
self.scope = None
self.typedef_flag = typedef_flag
if base_type is not None:
base_type.is_subclassed = True
self.base_type = base_type
self.module_name = None
self.objstruct_cname = None
self.typeobj_cname = None
self.typeptr_cname = None
self.vtabslot_cname = None
self.vtabstruct_cname = None
self.vtabptr_cname = None
self.vtable_cname = None
self.is_external = is_external
self.check_size = check_size or 'warn'
self.defered_declarations = []
def set_scope(self, scope):
self.scope = scope
if scope:
scope.parent_type = self
def needs_nonecheck(self):
return True
def subtype_of_resolved_type(self, other_type):
if other_type.is_extension_type or other_type.is_builtin_type:
return self is other_type or (
self.base_type and self.base_type.subtype_of(other_type))
else:
return other_type is py_object_type
def typeobj_is_available(self):
# Do we have a pointer to the type object?
return self.typeptr_cname
def typeobj_is_imported(self):
# If we don't know the C name of the type object but we do
# know which module it's defined in, it will be imported.
return self.typeobj_cname is None and self.module_name is not None
def assignable_from(self, src_type):
if self == src_type:
return True
if isinstance(src_type, PyExtensionType):
if src_type.base_type is not None:
return self.assignable_from(src_type.base_type)
if isinstance(src_type, BuiltinObjectType):
# FIXME: This is an ugly special case that we currently
# keep supporting. It allows users to specify builtin
# types as external extension types, while keeping them
# compatible with the real builtin types. We already
# generate a warning for it. Big TODO: remove!
return (self.module_name == '__builtin__' and
self.name == src_type.name)
return False
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0, deref = 0):
if pyrex or for_display:
base_code = self.name
else:
if self.typedef_flag:
objstruct = self.objstruct_cname
else:
objstruct = "struct %s" % self.objstruct_cname
base_code = public_decl(objstruct, dll_linkage)
if deref:
assert not entity_code
else:
entity_code = "*%s" % entity_code
return self.base_declaration_code(base_code, entity_code)
def type_test_code(self, py_arg, notnone=False):
none_check = "((%s) == Py_None)" % py_arg
type_check = "likely(__Pyx_TypeTest(%s, %s))" % (
py_arg, self.typeptr_cname)
if notnone:
return type_check
else:
return "likely(%s || %s)" % (none_check, type_check)
def attributes_known(self):
return self.scope is not None
def __str__(self):
return self.name
def __repr__(self):
return "<PyExtensionType %s%s>" % (self.scope.class_name,
("", " typedef")[self.typedef_flag])
def py_type_name(self):
if not self.module_name:
return self.name
return "__import__(%r, None, None, ['']).%s" % (self.module_name,
self.name)
class CType(PyrexType):
#
# Base class for all C types (non-reference-counted).
#
# to_py_function string C function for converting to Python object
# from_py_function string C function for constructing from Python object
#
to_py_function = None
to_py_utility_code = None
from_py_function = None
from_py_utility_code = None
exception_value = None
exception_check = 1
def create_to_py_utility_code(self, env):
if self.to_py_function is not None:
if self.to_py_utility_code is not None:
env.use_utility_code(self.to_py_utility_code)
return True
return False
def create_from_py_utility_code(self, env):
if self.from_py_function is not None:
if self.from_py_utility_code is not None:
env.use_utility_code(self.from_py_utility_code)
return True
return False
def can_coerce_to_pyobject(self, env):
return self.create_to_py_utility_code(env)
def can_coerce_from_pyobject(self, env):
return self.create_from_py_utility_code(env)
def error_condition(self, result_code):
conds = []
if self.is_string or self.is_pyunicode_ptr:
conds.append("(!%s)" % result_code)
elif self.exception_value is not None:
conds.append("(%s == (%s)%s)" % (result_code, self.sign_and_name(), self.exception_value))
if self.exception_check:
conds.append("PyErr_Occurred()")
if len(conds) > 0:
return " && ".join(conds)
else:
return 0
def to_py_call_code(self, source_code, result_code, result_type, to_py_function=None):
func = self.to_py_function if to_py_function is None else to_py_function
assert func
if self.is_string or self.is_cpp_string:
if result_type.is_builtin_type:
result_type_name = result_type.name
if result_type_name in ('bytes', 'str', 'unicode'):
func = func.replace("Object", result_type_name.title(), 1)
elif result_type_name == 'bytearray':
func = func.replace("Object", "ByteArray", 1)
return '%s = %s(%s)' % (
result_code,
func,
source_code or 'NULL')
def from_py_call_code(self, source_code, result_code, error_pos, code,
from_py_function=None, error_condition=None,
special_none_cvalue=None):
return self._assign_from_py_code(
source_code, result_code, error_pos, code, from_py_function, error_condition,
special_none_cvalue=special_none_cvalue)
class PythranExpr(CType):
# Pythran object of a given type
to_py_function = "__Pyx_pythran_to_python"
is_pythran_expr = True
writable = True
has_attributes = 1
def __init__(self, pythran_type, org_buffer=None):
self.org_buffer = org_buffer
self.pythran_type = pythran_type
self.name = self.pythran_type
self.cname = self.pythran_type
self.from_py_function = "from_python<%s>" % (self.pythran_type)
self.scope = None
def declaration_code(self, entity_code, for_display=0, dll_linkage=None, pyrex=0):
assert not pyrex
return "%s %s" % (self.cname, entity_code)
def attributes_known(self):
if self.scope is None:
from . import Symtab
# FIXME: fake C scope, might be better represented by a struct or C++ class scope
self.scope = scope = Symtab.CClassScope(
'', None, visibility="extern", parent_type=self
)
scope.directives = {}
scope.declare_var("ndim", c_long_type, pos=None, cname="value", is_cdef=True)
scope.declare_cproperty(
"shape", c_ptr_type(c_long_type), "__Pyx_PythranShapeAccessor",
doc="Pythran array shape",
visibility="extern",
nogil=True,
)
return True
def __eq__(self, other):
return isinstance(other, PythranExpr) and self.pythran_type == other.pythran_type
def __ne__(self, other):
return not (isinstance(other, PythranExpr) and self.pythran_type == other.pythran_type)
def __hash__(self):
return hash(self.pythran_type)
class CConstOrVolatileType(BaseType):
"A C const or volatile type"
subtypes = ['cv_base_type']
is_cv_qualified = 1
def __init__(self, base_type, is_const=0, is_volatile=0):
self.cv_base_type = base_type
self.is_const = is_const
self.is_volatile = is_volatile
if base_type.has_attributes and base_type.scope is not None:
from .Symtab import CConstOrVolatileScope
self.scope = CConstOrVolatileScope(base_type.scope, is_const, is_volatile)
def cv_string(self):
cvstring = ""
if self.is_const:
cvstring = "const " + cvstring
if self.is_volatile:
cvstring = "volatile " + cvstring
return cvstring
def __repr__(self):
return "<CConstOrVolatileType %s%r>" % (self.cv_string(), self.cv_base_type)
def __str__(self):
return self.declaration_code("", for_display=1)
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
cv = self.cv_string()
if for_display or pyrex:
return cv + self.cv_base_type.declaration_code(entity_code, for_display, dll_linkage, pyrex)
else:
return self.cv_base_type.declaration_code(cv + entity_code, for_display, dll_linkage, pyrex)
def specialize(self, values):
base_type = self.cv_base_type.specialize(values)
if base_type == self.cv_base_type:
return self
return CConstOrVolatileType(base_type,
self.is_const, self.is_volatile)
def deduce_template_params(self, actual):
return self.cv_base_type.deduce_template_params(actual)
def can_coerce_to_pyobject(self, env):
return self.cv_base_type.can_coerce_to_pyobject(env)
def can_coerce_from_pyobject(self, env):
return self.cv_base_type.can_coerce_from_pyobject(env)
def create_to_py_utility_code(self, env):
if self.cv_base_type.create_to_py_utility_code(env):
self.to_py_function = self.cv_base_type.to_py_function
return True
def same_as_resolved_type(self, other_type):
if other_type.is_cv_qualified:
return self.cv_base_type.same_as_resolved_type(other_type.cv_base_type)
# Accept cv LHS <- non-cv RHS.
return self.cv_base_type.same_as_resolved_type(other_type)
def __getattr__(self, name):
return getattr(self.cv_base_type, name)
def CConstType(base_type):
return CConstOrVolatileType(base_type, is_const=1)
class FusedType(CType):
"""
Represents a Fused Type. All it needs to do is keep track of the types
it aggregates, as it will be replaced with its specific version wherever
needed.
See http://wiki.cython.org/enhancements/fusedtypes
types [PyrexType] is the list of types to be fused
name str the name of the ctypedef
"""
is_fused = 1
exception_check = 0
def __init__(self, types, name=None):
# Use list rather than set to preserve order (list should be short).
flattened_types = []
for t in types:
if t.is_fused:
# recursively merge in subtypes
if isinstance(t, FusedType):
t_types = t.types
else:
# handle types that aren't a fused type themselves but contain fused types
# for example a C++ template where the template type is fused.
t_fused_types = t.get_fused_types()
t_types = []
for substitution in product(
*[fused_type.types for fused_type in t_fused_types]
):
t_types.append(
t.specialize(
{
fused_type: sub
for fused_type, sub in zip(
t_fused_types, substitution
)
}
)
)
for subtype in t_types:
if subtype not in flattened_types:
flattened_types.append(subtype)
elif t not in flattened_types:
flattened_types.append(t)
self.types = flattened_types
self.name = name
def declaration_code(self, entity_code, for_display = 0,
dll_linkage = None, pyrex = 0):
if pyrex or for_display:
return self.name
raise Exception("This may never happen, please report a bug")
def __repr__(self):
return 'FusedType(name=%r)' % self.name
def specialize(self, values):
if self in values:
return values[self]
else:
raise CannotSpecialize()
def get_fused_types(self, result=None, seen=None, include_function_return_type=False):
if result is None:
return [self]
if self not in seen:
result.append(self)
seen.add(self)
class CVoidType(CType):
#
# C "void" type
#
is_void = 1
to_py_function = "__Pyx_void_to_None"
def __repr__(self):
return "<CVoidType>"
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if pyrex or for_display:
base_code = "void"
else:
base_code = public_decl("void", dll_linkage)
return self.base_declaration_code(base_code, entity_code)
def is_complete(self):
return 0
class InvisibleVoidType(CVoidType):
#
# For use with C++ constructors and destructors return types.
# Acts like void, but does not print out a declaration.
#
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if pyrex or for_display:
base_code = "[void]"
else:
base_code = public_decl("", dll_linkage)
return self.base_declaration_code(base_code, entity_code)
class CNumericType(CType):
#
# Base class for all C numeric types.
#
# rank integer Relative size
# signed integer 0 = unsigned, 1 = unspecified, 2 = explicitly signed
#
is_numeric = 1
default_value = "0"
has_attributes = True
scope = None
sign_words = ("unsigned ", "", "signed ")
def __init__(self, rank, signed = 1):
self.rank = rank
if rank > 0 and signed == SIGNED:
# Signed is meaningless for anything but char, and complicates
# type promotion.
signed = 1
self.signed = signed
def sign_and_name(self):
s = self.sign_words[self.signed]
n = rank_to_type_name[self.rank]
return s + n
def is_simple_buffer_dtype(self):
return True
def __repr__(self):
return "<CNumericType %s>" % self.sign_and_name()
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
type_name = self.sign_and_name()
if pyrex or for_display:
base_code = type_name.replace('PY_LONG_LONG', 'long long')
else:
base_code = public_decl(type_name, dll_linkage)
base_code = StringEncoding.EncodedString(base_code)
return self.base_declaration_code(base_code, entity_code)
def attributes_known(self):
if self.scope is None:
from . import Symtab
self.scope = scope = Symtab.CClassScope(
'',
None,
visibility="extern",
parent_type=self)
scope.directives = {}
scope.declare_cfunction(
"conjugate",
CFuncType(self, [CFuncTypeArg("self", self, None)], nogil=True),
pos=None,
defining=1,
cname=" ")
return True
def __lt__(self, other):
"""Sort based on rank, preferring signed over unsigned"""
if other.is_numeric:
return self.rank > other.rank and self.signed >= other.signed
# Prefer numeric types over others
return True
def py_type_name(self):
if self.rank <= 4:
return "int"
return "float"
class ForbidUseClass:
def __repr__(self):
raise RuntimeError()
def __str__(self):
raise RuntimeError()
ForbidUse = ForbidUseClass()
class CIntLike(object):
"""Mixin for shared behaviour of C integers and enums.
"""
to_py_function = None
from_py_function = None
to_pyunicode_utility = None
default_format_spec = 'd'
def can_coerce_to_pyobject(self, env):
return True
def can_coerce_from_pyobject(self, env):
return True
def create_to_py_utility_code(self, env):
if type(self).to_py_function is None:
self.to_py_function = "__Pyx_PyInt_From_" + self.specialization_name()
env.use_utility_code(TempitaUtilityCode.load_cached(
"CIntToPy", "TypeConversion.c",
context={"TYPE": self.empty_declaration_code(),
"TO_PY_FUNCTION": self.to_py_function}))
return True
def create_from_py_utility_code(self, env):
if type(self).from_py_function is None:
self.from_py_function = "__Pyx_PyInt_As_" + self.specialization_name()
env.use_utility_code(TempitaUtilityCode.load_cached(
"CIntFromPy", "TypeConversion.c",
context={
"TYPE": self.empty_declaration_code(),
"FROM_PY_FUNCTION": self.from_py_function,
"IS_ENUM": self.is_enum,
}))
return True
@staticmethod
def _parse_format(format_spec):
padding = ' '
if not format_spec:
return ('d', 0, padding)
format_type = format_spec[-1]
if format_type in ('o', 'd', 'x', 'X'):
prefix = format_spec[:-1]
elif format_type.isdigit():
format_type = 'd'
prefix = format_spec
else:
return (None, 0, padding)
if not prefix:
return (format_type, 0, padding)
if prefix[0] == '-':
prefix = prefix[1:]
if prefix and prefix[0] == '0':
padding = '0'
prefix = prefix.lstrip('0')
if prefix.isdigit():
return (format_type, int(prefix), padding)
return (None, 0, padding)
def can_coerce_to_pystring(self, env, format_spec=None):
format_type, width, padding = self._parse_format(format_spec)
return format_type is not None and width <= 2**30
def convert_to_pystring(self, cvalue, code, format_spec=None):
if self.to_pyunicode_utility is None:
utility_code_name = "__Pyx_PyUnicode_From_" + self.specialization_name()
to_pyunicode_utility = TempitaUtilityCode.load_cached(
"CIntToPyUnicode", "TypeConversion.c",
context={"TYPE": self.empty_declaration_code(),
"TO_PY_FUNCTION": utility_code_name})
self.to_pyunicode_utility = (utility_code_name, to_pyunicode_utility)
else:
utility_code_name, to_pyunicode_utility = self.to_pyunicode_utility
code.globalstate.use_utility_code(to_pyunicode_utility)
format_type, width, padding_char = self._parse_format(format_spec)
return "%s(%s, %d, '%s', '%s')" % (utility_code_name, cvalue, width, padding_char, format_type)
class CIntType(CIntLike, CNumericType):
is_int = 1
typedef_flag = 0
exception_value = -1
def get_to_py_type_conversion(self):
if self.rank < list(rank_to_type_name).index('int'):
# This assumes sizeof(short) < sizeof(int)
return "PyInt_FromLong"
else:
# Py{Int|Long}_From[Unsigned]Long[Long]
Prefix = "Int"
SignWord = ""
TypeName = "Long"
if not self.signed:
Prefix = "Long"
SignWord = "Unsigned"
if self.rank >= list(rank_to_type_name).index('PY_LONG_LONG'):
Prefix = "Long"
TypeName = "LongLong"
return "Py%s_From%s%s" % (Prefix, SignWord, TypeName)
def assignable_from_resolved_type(self, src_type):
return src_type.is_int or src_type.is_enum or src_type is error_type
def invalid_value(self):
if rank_to_type_name[int(self.rank)] == 'char':
return "'?'"
else:
# We do not really know the size of the type, so return
# a 32-bit literal and rely on casting to final type. It will
# be negative for signed ints, which is good.
return "0xbad0bad0"
def overflow_check_binop(self, binop, env, const_rhs=False):
env.use_utility_code(UtilityCode.load("Common", "Overflow.c"))
type = self.empty_declaration_code()
name = self.specialization_name()
if binop == "lshift":
env.use_utility_code(TempitaUtilityCode.load_cached(
"LeftShift", "Overflow.c",
context={'TYPE': type, 'NAME': name, 'SIGNED': self.signed}))
else:
if const_rhs:
binop += "_const"
if type in ('int', 'long', 'long long'):
env.use_utility_code(TempitaUtilityCode.load_cached(
"BaseCaseSigned", "Overflow.c",
context={'INT': type, 'NAME': name}))
elif type in ('unsigned int', 'unsigned long', 'unsigned long long'):
env.use_utility_code(TempitaUtilityCode.load_cached(
"BaseCaseUnsigned", "Overflow.c",
context={'UINT': type, 'NAME': name}))
elif self.rank <= 1:
# sizeof(short) < sizeof(int)
return "__Pyx_%s_%s_no_overflow" % (binop, name)
else:
_load_overflow_base(env)
env.use_utility_code(TempitaUtilityCode.load_cached(
"SizeCheck", "Overflow.c",
context={'TYPE': type, 'NAME': name}))
env.use_utility_code(TempitaUtilityCode.load_cached(
"Binop", "Overflow.c",
context={'TYPE': type, 'NAME': name, 'BINOP': binop}))
return "__Pyx_%s_%s_checking_overflow" % (binop, name)
def _load_overflow_base(env):
env.use_utility_code(UtilityCode.load("Common", "Overflow.c"))
for type in ('int', 'long', 'long long'):
env.use_utility_code(TempitaUtilityCode.load_cached(
"BaseCaseSigned", "Overflow.c",
context={'INT': type, 'NAME': type.replace(' ', '_')}))
for type in ('unsigned int', 'unsigned long', 'unsigned long long'):
env.use_utility_code(TempitaUtilityCode.load_cached(
"BaseCaseUnsigned", "Overflow.c",
context={'UINT': type, 'NAME': type.replace(' ', '_')}))
class CAnonEnumType(CIntType):
is_enum = 1
def sign_and_name(self):
return 'int'
def specialization_name(self):
# ensure that the to/from Python functions don't conflict with
# "int"
return '__pyx_anon_enum'
class CReturnCodeType(CIntType):
to_py_function = "__Pyx_Owned_Py_None"
is_returncode = True
exception_check = False
default_format_spec = ''
def specialization_name(self):
# I don't think we should end up creating PyInt_As_int/PyInt_From_int functions
# for this type, but it's better they're distinct in case it happens.
return super(CReturnCodeType, self).specialization_name() + "return_code"
def can_coerce_to_pystring(self, env, format_spec=None):
return not format_spec
def convert_to_pystring(self, cvalue, code, format_spec=None):
return "__Pyx_NewRef(%s)" % code.globalstate.get_py_string_const(StringEncoding.EncodedString("None")).cname
class CBIntType(CIntType):
to_py_function = "__Pyx_PyBool_FromLong"
from_py_function = "__Pyx_PyObject_IsTrue"
exception_check = 1 # for C++ bool
default_format_spec = ''
def can_coerce_to_pystring(self, env, format_spec=None):
return not format_spec or super(CBIntType, self).can_coerce_to_pystring(env, format_spec)
def convert_to_pystring(self, cvalue, code, format_spec=None):
if format_spec:
return super(CBIntType, self).convert_to_pystring(cvalue, code, format_spec)
# NOTE: no caching here as the string constant cnames depend on the current module
utility_code_name = "__Pyx_PyUnicode_FromBInt_" + self.specialization_name()
to_pyunicode_utility = TempitaUtilityCode.load_cached(
"CBIntToPyUnicode", "TypeConversion.c", context={
"TRUE_CONST": code.globalstate.get_py_string_const(StringEncoding.EncodedString("True")).cname,
"FALSE_CONST": code.globalstate.get_py_string_const(StringEncoding.EncodedString("False")).cname,
"TO_PY_FUNCTION": utility_code_name,
})
code.globalstate.use_utility_code(to_pyunicode_utility)
return "%s(%s)" % (utility_code_name, cvalue)
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if for_display:
base_code = 'bool'
elif pyrex:
base_code = 'bint'
else:
base_code = public_decl('int', dll_linkage)
return self.base_declaration_code(base_code, entity_code)
def specialization_name(self):
return "bint"
def __repr__(self):
return "<CNumericType bint>"
def __str__(self):
return 'bint'
def py_type_name(self):
return "bool"
class CPyUCS4IntType(CIntType):
# Py_UCS4
is_unicode_char = True
# Py_UCS4 coerces from and to single character unicode strings (or
# at most two characters on 16bit Unicode builds), but we also
# allow Python integers as input. The value range for Py_UCS4
# is 0..1114111, which is checked when converting from an integer
# value.
to_py_function = "__Pyx_PyUnicode_FromOrdinal"
from_py_function = "__Pyx_PyObject_AsPy_UCS4"
def can_coerce_to_pystring(self, env, format_spec=None):
return False # does the right thing anyway
def create_from_py_utility_code(self, env):
env.use_utility_code(UtilityCode.load_cached("ObjectAsUCS4", "TypeConversion.c"))
return True
def sign_and_name(self):
return "Py_UCS4"
class CPyUnicodeIntType(CIntType):
# Py_UNICODE
is_unicode_char = True
# Py_UNICODE coerces from and to single character unicode strings,
# but we also allow Python integers as input. The value range for
# Py_UNICODE is 0..1114111, which is checked when converting from
# an integer value.
to_py_function = "__Pyx_PyUnicode_FromOrdinal"
from_py_function = "__Pyx_PyObject_AsPy_UNICODE"
def can_coerce_to_pystring(self, env, format_spec=None):
return False # does the right thing anyway
def create_from_py_utility_code(self, env):
env.use_utility_code(UtilityCode.load_cached("ObjectAsPyUnicode", "TypeConversion.c"))
return True
def sign_and_name(self):
return "Py_UNICODE"
class CPyHashTType(CIntType):
to_py_function = "__Pyx_PyInt_FromHash_t"
from_py_function = "__Pyx_PyInt_AsHash_t"
def sign_and_name(self):
return "Py_hash_t"
class CPySSizeTType(CIntType):
to_py_function = "PyInt_FromSsize_t"
from_py_function = "__Pyx_PyIndex_AsSsize_t"
def sign_and_name(self):
return "Py_ssize_t"
class CSSizeTType(CIntType):
to_py_function = "PyInt_FromSsize_t"
from_py_function = "PyInt_AsSsize_t"
def sign_and_name(self):
return "Py_ssize_t"
class CSizeTType(CIntType):
to_py_function = "__Pyx_PyInt_FromSize_t"
def sign_and_name(self):
return "size_t"
class CPtrdiffTType(CIntType):
def sign_and_name(self):
return "ptrdiff_t"
class CFloatType(CNumericType):
is_float = 1
to_py_function = "PyFloat_FromDouble"
from_py_function = "__pyx_PyFloat_AsDouble"
exception_value = -1
def __init__(self, rank, math_h_modifier = ''):
CNumericType.__init__(self, rank, 1)
self.math_h_modifier = math_h_modifier
if rank == RANK_FLOAT:
self.from_py_function = "__pyx_PyFloat_AsFloat"
def assignable_from_resolved_type(self, src_type):
return (src_type.is_numeric and not src_type.is_complex) or src_type is error_type
def invalid_value(self):
return Naming.PYX_NAN
class CComplexType(CNumericType):
is_complex = 1
has_attributes = 1
scope = None
@property
def to_py_function(self):
return "__pyx_PyComplex_FromComplex%s" % self.implementation_suffix
def __init__(self, real_type):
while real_type.is_typedef and not real_type.typedef_is_external:
real_type = real_type.typedef_base_type
self.funcsuffix = "_%s" % real_type.specialization_name()
if not real_type.is_float:
# neither C nor C++ supports non-floating complex numbers,
# so fall back the on Cython implementation.
self.implementation_suffix = "_Cy"
elif real_type.is_typedef and real_type.typedef_is_external:
# C can't handle typedefs in complex numbers,
# so in this case also fall back on the Cython implementation.
self.implementation_suffix = "_CyTypedef"
else:
self.implementation_suffix = ""
if real_type.is_float:
self.math_h_modifier = real_type.math_h_modifier
else:
self.math_h_modifier = "_UNUSED"
self.real_type = real_type
CNumericType.__init__(self, real_type.rank + 0.5, real_type.signed)
self.binops = {}
self.from_parts = "%s_from_parts" % self.specialization_name()
self.default_value = "%s(0, 0)" % self.from_parts
def __eq__(self, other):
if isinstance(self, CComplexType) and isinstance(other, CComplexType):
return self.real_type == other.real_type
else:
return False
def __ne__(self, other):
if isinstance(self, CComplexType) and isinstance(other, CComplexType):
return self.real_type != other.real_type
else:
return True
def __lt__(self, other):
if isinstance(self, CComplexType) and isinstance(other, CComplexType):
return self.real_type < other.real_type
else:
# this is arbitrary, but it makes sure we always have
# *some* kind of order
return False
def __hash__(self):
return ~hash(self.real_type)
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if pyrex or for_display:
real_code = self.real_type.declaration_code("", for_display, dll_linkage, pyrex)
base_code = "%s complex" % real_code
else:
base_code = public_decl(self.sign_and_name(), dll_linkage)
return self.base_declaration_code(base_code, entity_code)
def sign_and_name(self):
real_type_name = self.real_type.specialization_name()
real_type_name = real_type_name.replace('long__double','long_double')
real_type_name = real_type_name.replace('PY_LONG_LONG','long_long')
return Naming.type_prefix + real_type_name + "_complex"
def assignable_from(self, src_type):
# Temporary hack/feature disabling, see #441
if (not src_type.is_complex and src_type.is_numeric and src_type.is_typedef
and src_type.typedef_is_external):
return False
elif src_type.is_pyobject:
return True
else:
return super(CComplexType, self).assignable_from(src_type)
def assignable_from_resolved_type(self, src_type):
return (src_type.is_complex and self.real_type.assignable_from_resolved_type(src_type.real_type)
or src_type.is_numeric and self.real_type.assignable_from_resolved_type(src_type)
or src_type is error_type)
def attributes_known(self):
if self.scope is None:
from . import Symtab
self.scope = scope = Symtab.CClassScope(
'',
None,
visibility="extern",
parent_type=self)
scope.directives = {}
scope.declare_var("real", self.real_type, None, cname="real", is_cdef=True)
scope.declare_var("imag", self.real_type, None, cname="imag", is_cdef=True)
scope.declare_cfunction(
"conjugate",
CFuncType(self, [CFuncTypeArg("self", self, None)], nogil=True),
pos=None,
defining=1,
cname="__Pyx_c_conj%s" % self.funcsuffix)
return True
def _utility_code_context(self):
return {
'type': self.empty_declaration_code(),
'type_name': self.specialization_name(),
'real_type': self.real_type.empty_declaration_code(),
'func_suffix': self.funcsuffix,
'm': self.math_h_modifier,
'is_float': int(self.real_type.is_float),
'is_extern_float_typedef': int(
self.real_type.is_float and self.real_type.is_typedef and self.real_type.typedef_is_external)
}
def create_declaration_utility_code(self, env):
# This must always be run, because a single CComplexType instance can be shared
# across multiple compilations (the one created in the module scope)
if self.real_type.is_float:
env.use_utility_code(UtilityCode.load_cached('Header', 'Complex.c'))
utility_code_context = self._utility_code_context()
env.use_utility_code(UtilityCode.load_cached(
'RealImag' + self.implementation_suffix, 'Complex.c'))
env.use_utility_code(TempitaUtilityCode.load_cached(
'Declarations', 'Complex.c', utility_code_context))
env.use_utility_code(TempitaUtilityCode.load_cached(
'Arithmetic', 'Complex.c', utility_code_context))
return True
def can_coerce_to_pyobject(self, env):
return True
def can_coerce_from_pyobject(self, env):
return True
def create_to_py_utility_code(self, env):
env.use_utility_code(TempitaUtilityCode.load_cached(
'ToPy', 'Complex.c', self._utility_code_context()))
return True
def create_from_py_utility_code(self, env):
env.use_utility_code(TempitaUtilityCode.load_cached(
'FromPy', 'Complex.c', self._utility_code_context()))
self.from_py_function = "__Pyx_PyComplex_As_" + self.specialization_name()
return True
def lookup_op(self, nargs, op):
try:
return self.binops[nargs, op]
except KeyError:
pass
try:
op_name = complex_ops[nargs, op]
self.binops[nargs, op] = func_name = "__Pyx_c_%s%s" % (op_name, self.funcsuffix)
return func_name
except KeyError:
return None
def unary_op(self, op):
return self.lookup_op(1, op)
def binary_op(self, op):
return self.lookup_op(2, op)
def py_type_name(self):
return "complex"
def cast_code(self, expr_code):
return expr_code
def real_code(self, expr_code):
return "__Pyx_CREAL%s(%s)" % (self.implementation_suffix, expr_code)
def imag_code(self, expr_code):
return "__Pyx_CIMAG%s(%s)" % (self.implementation_suffix, expr_code)
complex_ops = {
(1, '-'): 'neg',
(1, 'zero'): 'is_zero',
(2, '+'): 'sum',
(2, '-'): 'diff',
(2, '*'): 'prod',
(2, '/'): 'quot',
(2, '**'): 'pow',
(2, '=='): 'eq',
}
class SoftCComplexType(CComplexType):
"""
a**b in Python can return either a complex or a float
depending on the sign of a. This "soft complex" type is
stored as a C complex (and so is a little slower than a
direct C double) but it prints/coerces to a float if
the imaginary part is 0. Therefore it provides a C
representation of the Python behaviour.
"""
to_py_function = "__pyx_Py_FromSoftComplex"
def __init__(self):
super(SoftCComplexType, self).__init__(c_double_type)
def declaration_code(self, entity_code, for_display=0, dll_linkage=None, pyrex=0):
base_result = super(SoftCComplexType, self).declaration_code(
entity_code,
for_display=for_display,
dll_linkage=dll_linkage,
pyrex=pyrex,
)
if for_display:
return "soft %s" % base_result
else:
return base_result
def create_to_py_utility_code(self, env):
env.use_utility_code(UtilityCode.load_cached('SoftComplexToPy', 'Complex.c'))
return True
def __repr__(self):
result = super(SoftCComplexType, self).__repr__()
assert result[-1] == ">"
return "%s (soft)%s" % (result[:-1], result[-1])
class CPyTSSTType(CType):
#
# PEP-539 "Py_tss_t" type
#
declaration_value = "Py_tss_NEEDS_INIT"
def __repr__(self):
return "<Py_tss_t>"
def declaration_code(self, entity_code,
for_display=0, dll_linkage=None, pyrex=0):
if pyrex or for_display:
base_code = "Py_tss_t"
else:
base_code = public_decl("Py_tss_t", dll_linkage)
return self.base_declaration_code(base_code, entity_code)
class CPointerBaseType(CType):
# common base type for pointer/array types
#
# base_type CType Reference type
subtypes = ['base_type']
def __init__(self, base_type):
self.base_type = base_type
if base_type.is_cv_qualified:
base_type = base_type.cv_base_type
for char_type in (c_char_type, c_uchar_type, c_schar_type):
if base_type.same_as(char_type):
self.is_string = 1
break
else:
if base_type.same_as(c_py_unicode_type):
self.is_pyunicode_ptr = 1
if self.is_string and not base_type.is_error:
if base_type.signed == 2:
self.to_py_function = "__Pyx_PyObject_FromCString"
if self.is_ptr:
self.from_py_function = "__Pyx_PyObject_As%sSString"
elif base_type.signed:
self.to_py_function = "__Pyx_PyObject_FromString"
if self.is_ptr:
self.from_py_function = "__Pyx_PyObject_As%sString"
else:
self.to_py_function = "__Pyx_PyObject_FromCString"
if self.is_ptr:
self.from_py_function = "__Pyx_PyObject_As%sUString"
if self.is_ptr:
self.from_py_function %= '' if self.base_type.is_const else 'Writable'
self.exception_value = "NULL"
elif self.is_pyunicode_ptr and not base_type.is_error:
self.to_py_function = "__Pyx_PyUnicode_FromUnicode"
self.to_py_utility_code = UtilityCode.load_cached(
"pyunicode_from_unicode", "StringTools.c")
if self.is_ptr:
self.from_py_function = "__Pyx_PyUnicode_AsUnicode"
self.exception_value = "NULL"
def py_type_name(self):
if self.is_string:
return "bytes"
elif self.is_pyunicode_ptr:
return "unicode"
else:
return super(CPointerBaseType, self).py_type_name()
def literal_code(self, value):
if self.is_string:
assert isinstance(value, str)
return '"%s"' % StringEncoding.escape_byte_string(value)
return str(value)
class CArrayType(CPointerBaseType):
# base_type CType Element type
# size integer or None Number of elements
is_array = 1
to_tuple_function = None
def __init__(self, base_type, size):
super(CArrayType, self).__init__(base_type)
self.size = size
def __eq__(self, other):
if isinstance(other, CType) and other.is_array and self.size == other.size:
return self.base_type.same_as(other.base_type)
return False
def __hash__(self):
return hash(self.base_type) + 28 # arbitrarily chosen offset
def __repr__(self):
return "<CArrayType %s %s>" % (self.size, repr(self.base_type))
def same_as_resolved_type(self, other_type):
return ((other_type.is_array and
self.base_type.same_as(other_type.base_type))
or other_type is error_type)
def assignable_from_resolved_type(self, src_type):
# C arrays are assigned by value, either Python containers or C arrays/pointers
if src_type.is_pyobject:
return True
if src_type.is_ptr or src_type.is_array:
return self.base_type.assignable_from(src_type.base_type)
return False
def element_ptr_type(self):
return c_ptr_type(self.base_type)
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if self.size is not None:
dimension_code = self.size
else:
dimension_code = ""
if entity_code.startswith("*"):
entity_code = "(%s)" % entity_code
return self.base_type.declaration_code(
"%s[%s]" % (entity_code, dimension_code),
for_display, dll_linkage, pyrex)
def as_argument_type(self):
return c_ptr_type(self.base_type)
def is_complete(self):
return self.size is not None
def specialize(self, values):
base_type = self.base_type.specialize(values)
if base_type == self.base_type:
return self
else:
return CArrayType(base_type, self.size)
def deduce_template_params(self, actual):
if isinstance(actual, CArrayType):
return self.base_type.deduce_template_params(actual.base_type)
else:
return {}
def can_coerce_to_pyobject(self, env):
return self.base_type.can_coerce_to_pyobject(env)
def can_coerce_from_pyobject(self, env):
return self.base_type.can_coerce_from_pyobject(env)
def create_to_py_utility_code(self, env):
if self.to_py_function is not None:
return self.to_py_function
if not self.base_type.create_to_py_utility_code(env):
return False
safe_typename = self.base_type.specialization_name()
to_py_function = "__Pyx_carray_to_py_%s" % safe_typename
to_tuple_function = "__Pyx_carray_to_tuple_%s" % safe_typename
from .UtilityCode import CythonUtilityCode
context = {
'cname': to_py_function,
'to_tuple_cname': to_tuple_function,
'base_type': self.base_type,
}
env.use_utility_code(CythonUtilityCode.load(
"carray.to_py", "CConvert.pyx",
outer_module_scope=env.global_scope(), # need access to types declared in module
context=context, compiler_directives=dict(env.global_scope().directives)))
self.to_tuple_function = to_tuple_function
self.to_py_function = to_py_function
return True
def to_py_call_code(self, source_code, result_code, result_type, to_py_function=None):
func = self.to_py_function if to_py_function is None else to_py_function
if self.is_string or self.is_pyunicode_ptr:
return '%s = %s(%s)' % (
result_code,
func,
source_code)
target_is_tuple = result_type.is_builtin_type and result_type.name == 'tuple'
return '%s = %s(%s, %s)' % (
result_code,
self.to_tuple_function if target_is_tuple else func,
source_code,
self.size)
def create_from_py_utility_code(self, env):
if self.from_py_function is not None:
return self.from_py_function
if not self.base_type.create_from_py_utility_code(env):
return False
from_py_function = "__Pyx_carray_from_py_%s" % self.base_type.specialization_name()
from .UtilityCode import CythonUtilityCode
context = {
'cname': from_py_function,
'base_type': self.base_type,
}
env.use_utility_code(CythonUtilityCode.load(
"carray.from_py", "CConvert.pyx",
outer_module_scope=env.global_scope(), # need access to types declared in module
context=context, compiler_directives=dict(env.global_scope().directives)))
self.from_py_function = from_py_function
return True
def from_py_call_code(self, source_code, result_code, error_pos, code,
from_py_function=None, error_condition=None,
special_none_cvalue=None):
assert not error_condition, '%s: %s' % (error_pos, error_condition)
assert not special_none_cvalue, '%s: %s' % (error_pos, special_none_cvalue) # not currently supported
call_code = "%s(%s, %s, %s)" % (
from_py_function or self.from_py_function,
source_code, result_code, self.size)
return code.error_goto_if_neg(call_code, error_pos)
def error_condition(self, result_code):
# It isn't possible to use CArrays as return type so the error_condition
# is irrelevant. Returning a falsy value does avoid an error when getting
# from_py_call_code from a typedef.
return ""
class CPtrType(CPointerBaseType):
# base_type CType Reference type
is_ptr = 1
default_value = "0"
exception_value = "NULL"
def __hash__(self):
return hash(self.base_type) + 27 # arbitrarily chosen offset
def __eq__(self, other):
if isinstance(other, CType) and other.is_ptr:
return self.base_type.same_as(other.base_type)
return False
def __ne__(self, other):
return not (self == other)
def __repr__(self):
return "<CPtrType %s>" % repr(self.base_type)
def same_as_resolved_type(self, other_type):
return ((other_type.is_ptr and
self.base_type.same_as(other_type.base_type))
or other_type is error_type)
def is_simple_buffer_dtype(self):
return True
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
#print "CPtrType.declaration_code: pointer to", self.base_type ###
return self.base_type.declaration_code(
"*%s" % entity_code,
for_display, dll_linkage, pyrex)
def assignable_from_resolved_type(self, other_type):
if other_type is error_type:
return 1
if other_type.is_null_ptr:
return 1
if self.base_type.is_cv_qualified:
self = CPtrType(self.base_type.cv_base_type)
if self.base_type.is_cfunction:
if other_type.is_ptr:
other_type = other_type.base_type.resolve()
if other_type.is_cfunction:
return self.base_type.pointer_assignable_from_resolved_type(other_type)
else:
return 0
if (self.base_type.is_cpp_class and other_type.is_ptr
and other_type.base_type.is_cpp_class and other_type.base_type.is_subclass(self.base_type)):
return 1
if other_type.is_array or other_type.is_ptr:
return self.base_type.is_void or self.base_type.same_as(other_type.base_type)
return 0
def assignment_failure_extra_info(self, src_type, src_name):
if self.base_type.is_cfunction and src_type.is_ptr:
src_type = src_type.base_type.resolve()
if self.base_type.is_cfunction and src_type.is_cfunction:
copied_src_type = copy.copy(src_type)
# make the exception values the same as us
copied_src_type.exception_check = self.base_type.exception_check
copied_src_type.exception_value = self.base_type.exception_value
if self.base_type.pointer_assignable_from_resolved_type(copied_src_type):
# the only reason we can't assign is because of exception incompatibility
msg = "Exception values are incompatible."
if not self.base_type.exception_check and not self.base_type.exception_value:
if src_name is None:
src_name = "the value being assigned"
else:
src_name = "'{}'".format(src_name)
msg += " Suggest adding 'noexcept' to the type of {0}.".format(src_name)
return msg
return super(CPtrType, self).assignment_failure_extra_info(src_type, src_name)
def specialize(self, values):
base_type = self.base_type.specialize(values)
if base_type == self.base_type:
return self
else:
return CPtrType(base_type)
def deduce_template_params(self, actual):
if isinstance(actual, CPtrType):
return self.base_type.deduce_template_params(actual.base_type)
else:
return {}
def invalid_value(self):
return "1"
def find_cpp_operation_type(self, operator, operand_type=None):
if self.base_type.is_cpp_class:
return self.base_type.find_cpp_operation_type(operator, operand_type)
return None
def get_fused_types(self, result=None, seen=None, include_function_return_type=False):
# For function pointers, include the return type - unlike for fused functions themselves,
# where the return type cannot be an independent fused type (i.e. is derived or non-fused).
return super(CPointerBaseType, self).get_fused_types(result, seen, include_function_return_type=True)
class CNullPtrType(CPtrType):
is_null_ptr = 1
class CReferenceBaseType(BaseType):
is_fake_reference = 0
# Common base type for C reference and C++ rvalue reference types.
subtypes = ['ref_base_type']
def __init__(self, base_type):
self.ref_base_type = base_type
def __repr__(self):
return "<%r %s>" % (self.__class__.__name__, self.ref_base_type)
def specialize(self, values):
base_type = self.ref_base_type.specialize(values)
if base_type == self.ref_base_type:
return self
else:
return type(self)(base_type)
def deduce_template_params(self, actual):
return self.ref_base_type.deduce_template_params(actual)
def __getattr__(self, name):
return getattr(self.ref_base_type, name)
class CReferenceType(CReferenceBaseType):
is_reference = 1
def __str__(self):
return "%s &" % self.ref_base_type
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
#print "CReferenceType.declaration_code: pointer to", self.base_type ###
return self.ref_base_type.declaration_code(
"&%s" % entity_code,
for_display, dll_linkage, pyrex)
class CFakeReferenceType(CReferenceType):
is_fake_reference = 1
def __str__(self):
return "%s [&]" % self.ref_base_type
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
#print "CReferenceType.declaration_code: pointer to", self.base_type ###
return "__Pyx_FakeReference<%s> %s" % (self.ref_base_type.empty_declaration_code(), entity_code)
class CppRvalueReferenceType(CReferenceBaseType):
is_rvalue_reference = 1
def __str__(self):
return "%s &&" % self.ref_base_type
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
return self.ref_base_type.declaration_code(
"&&%s" % entity_code,
for_display, dll_linkage, pyrex)
class CFuncType(CType):
# return_type CType
# args [CFuncTypeArg]
# has_varargs boolean
# exception_value string
# exception_check boolean True if PyErr_Occurred check needed
# calling_convention string Function calling convention
# nogil boolean Can be called without gil
# with_gil boolean Acquire gil around function body
# templates [string] or None
# cached_specialized_types [CFuncType] cached specialized versions of the CFuncType if defined in a pxd
# from_fused boolean Indicates whether this is a specialized
# C function
# is_strict_signature boolean function refuses to accept coerced arguments
# (used for optimisation overrides)
# is_const_method boolean
# is_static_method boolean
# op_arg_struct CPtrType Pointer to optional argument struct
is_cfunction = 1
original_sig = None
cached_specialized_types = None
from_fused = False
is_const_method = False
op_arg_struct = None
subtypes = ['return_type', 'args']
def __init__(self, return_type, args, has_varargs = 0,
exception_value = None, exception_check = 0, calling_convention = "",
nogil = 0, with_gil = 0, is_overridable = 0, optional_arg_count = 0,
is_const_method = False, is_static_method=False,
templates = None, is_strict_signature = False):
self.return_type = return_type
self.args = args
self.has_varargs = has_varargs
self.optional_arg_count = optional_arg_count
self.exception_value = exception_value
self.exception_check = exception_check
self.calling_convention = calling_convention
self.nogil = nogil
self.with_gil = with_gil
self.is_overridable = is_overridable
self.is_const_method = is_const_method
self.is_static_method = is_static_method
self.templates = templates
self.is_strict_signature = is_strict_signature
def __repr__(self):
arg_reprs = list(map(repr, self.args))
if self.has_varargs:
arg_reprs.append("...")
if self.exception_value:
except_clause = " %r" % self.exception_value
else:
except_clause = ""
if self.exception_check:
except_clause += "?"
return "<CFuncType %s %s[%s]%s>" % (
repr(self.return_type),
self.calling_convention_prefix(),
",".join(arg_reprs),
except_clause)
def with_with_gil(self, with_gil):
if with_gil == self.with_gil:
return self
else:
return CFuncType(
self.return_type, self.args, self.has_varargs,
self.exception_value, self.exception_check,
self.calling_convention, self.nogil,
with_gil,
self.is_overridable, self.optional_arg_count,
self.is_const_method, self.is_static_method,
self.templates, self.is_strict_signature)
def calling_convention_prefix(self):
cc = self.calling_convention
if cc:
return cc + " "
else:
return ""
def as_argument_type(self):
return c_ptr_type(self)
def same_c_signature_as(self, other_type, as_cmethod = 0):
return self.same_c_signature_as_resolved_type(
other_type.resolve(), as_cmethod)
def same_c_signature_as_resolved_type(self, other_type, as_cmethod=False, as_pxd_definition=False,
exact_semantics=True):
# If 'exact_semantics' is false, allow any equivalent C signatures
# if the Cython semantics are compatible, i.e. the same or wider for 'other_type'.
#print "CFuncType.same_c_signature_as_resolved_type:", \
# self, other_type, "as_cmethod =", as_cmethod ###
if other_type is error_type:
return 1
if not other_type.is_cfunction:
return 0
if self.is_overridable != other_type.is_overridable:
return 0
nargs = len(self.args)
if nargs != len(other_type.args):
return 0
# When comparing C method signatures, the first argument
# is exempt from compatibility checking (the proper check
# is performed elsewhere).
for i in range(as_cmethod, nargs):
if not self.args[i].type.same_as(other_type.args[i].type):
return 0
if self.has_varargs != other_type.has_varargs:
return 0
if self.optional_arg_count != other_type.optional_arg_count:
return 0
if as_pxd_definition:
# A narrowing of the return type declared in the pxd is allowed.
if not self.return_type.subtype_of_resolved_type(other_type.return_type):
return 0
else:
if not self.return_type.same_as(other_type.return_type):
return 0
if not self.same_calling_convention_as(other_type):
return 0
if exact_semantics:
if self.exception_check != other_type.exception_check:
return 0
if not self._same_exception_value(other_type.exception_value):
return 0
elif not self._is_exception_compatible_with(other_type):
return 0
return 1
def _same_exception_value(self, other_exc_value):
if self.exception_value == other_exc_value:
return 1
if self.exception_check != '+':
return 0
if not self.exception_value or not other_exc_value:
return 0
if self.exception_value.type != other_exc_value.type:
return 0
if self.exception_value.entry and other_exc_value.entry:
if self.exception_value.entry.cname != other_exc_value.entry.cname:
return 0
if self.exception_value.name != other_exc_value.name:
return 0
return 1
def compatible_signature_with(self, other_type, as_cmethod = 0):
return self.compatible_signature_with_resolved_type(other_type.resolve(), as_cmethod)
def compatible_signature_with_resolved_type(self, other_type, as_cmethod):
#print "CFuncType.same_c_signature_as_resolved_type:", \
# self, other_type, "as_cmethod =", as_cmethod ###
if other_type is error_type:
return 1
if not other_type.is_cfunction:
return 0
if not self.is_overridable and other_type.is_overridable:
return 0
nargs = len(self.args)
if nargs - self.optional_arg_count != len(other_type.args) - other_type.optional_arg_count:
return 0
if self.optional_arg_count < other_type.optional_arg_count:
return 0
# When comparing C method signatures, the first argument
# is exempt from compatibility checking (the proper check
# is performed elsewhere).
for i in range(as_cmethod, len(other_type.args)):
if not self.args[i].type.same_as(
other_type.args[i].type):
return 0
if self.has_varargs != other_type.has_varargs:
return 0
if not self.return_type.subtype_of_resolved_type(other_type.return_type):
return 0
if not self.same_calling_convention_as(other_type):
return 0
if self.nogil != other_type.nogil:
return 0
if not self._is_exception_compatible_with(other_type):
return 0
self.original_sig = other_type.original_sig or other_type
return 1
def _is_exception_compatible_with(self, other_type):
# narrower exception checks are ok, but prevent mismatches
if self.exception_check == '+' and other_type.exception_check != '+':
# must catch C++ exceptions if we raise them
return 0
if not other_type.exception_check or other_type.exception_value is not None:
# There's no problem if this type doesn't emit exceptions but the other type checks
if other_type.exception_check and not (self.exception_check or self.exception_value):
return 1
# if other does not *always* check exceptions, self must comply
if not self._same_exception_value(other_type.exception_value):
return 0
if self.exception_check and self.exception_check != other_type.exception_check:
# a redundant exception check doesn't make functions incompatible, but a missing one does
return 0
return 1
def narrower_c_signature_than(self, other_type, as_cmethod = 0):
return self.narrower_c_signature_than_resolved_type(other_type.resolve(), as_cmethod)
def narrower_c_signature_than_resolved_type(self, other_type, as_cmethod):
if other_type is error_type:
return 1
if not other_type.is_cfunction:
return 0
nargs = len(self.args)
if nargs != len(other_type.args):
return 0
for i in range(as_cmethod, nargs):
if not self.args[i].type.subtype_of_resolved_type(other_type.args[i].type):
return 0
else:
self.args[i].needs_type_test = other_type.args[i].needs_type_test \
or not self.args[i].type.same_as(other_type.args[i].type)
if self.has_varargs != other_type.has_varargs:
return 0
if self.optional_arg_count != other_type.optional_arg_count:
return 0
if not self.return_type.subtype_of_resolved_type(other_type.return_type):
return 0
if not self.exception_check and other_type.exception_check:
# a redundant exception check doesn't make functions incompatible, but a missing one does
return 0
if not self._same_exception_value(other_type.exception_value):
return 0
return 1
def same_calling_convention_as(self, other):
## XXX Under discussion ...
## callspec_words = ("__stdcall", "__cdecl", "__fastcall")
## cs1 = self.calling_convention
## cs2 = other.calling_convention
## if (cs1 in callspec_words or
## cs2 in callspec_words):
## return cs1 == cs2
## else:
## return True
sc1 = self.calling_convention == '__stdcall'
sc2 = other.calling_convention == '__stdcall'
return sc1 == sc2
def same_as_resolved_type(self, other_type, as_cmethod=False):
return self.same_c_signature_as_resolved_type(other_type, as_cmethod=as_cmethod) \
and self.nogil == other_type.nogil
def pointer_assignable_from_resolved_type(self, rhs_type):
# Accept compatible exception/nogil declarations for the RHS.
if rhs_type is error_type:
return 1
if not rhs_type.is_cfunction:
return 0
return rhs_type.same_c_signature_as_resolved_type(self, exact_semantics=False) \
and not (self.nogil and not rhs_type.nogil)
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0,
with_calling_convention = 1):
arg_decl_list = []
for arg in self.args[:len(self.args)-self.optional_arg_count]:
arg_decl_list.append(
arg.type.declaration_code("", for_display, pyrex = pyrex))
if self.is_overridable:
arg_decl_list.append("int %s" % Naming.skip_dispatch_cname)
if self.optional_arg_count:
if self.op_arg_struct:
arg_decl_list.append(self.op_arg_struct.declaration_code(Naming.optional_args_cname))
else:
# op_arg_struct may not be initialized at this point if this class is being used
# to prepare a Python error message or similar. In this case, just omit the args.
assert for_display
if self.has_varargs:
arg_decl_list.append("...")
arg_decl_code = ", ".join(arg_decl_list)
if not arg_decl_code and not pyrex:
arg_decl_code = "void"
trailer = ""
if (pyrex or for_display) and not self.return_type.is_pyobject:
if self.exception_value and self.exception_check:
trailer = " except? %s" % self.exception_value
elif self.exception_value and not self.exception_check:
trailer = " except %s" % self.exception_value
elif not self.exception_value and not self.exception_check:
trailer = " noexcept"
elif self.exception_check == '+':
trailer = " except +"
elif self.exception_check and for_display:
# not spelled out by default, unless for human eyes
trailer = " except *"
if self.nogil:
trailer += " nogil"
if not with_calling_convention:
cc = ''
else:
cc = self.calling_convention_prefix()
if (not entity_code and cc) or entity_code.startswith("*"):
entity_code = "(%s%s)" % (cc, entity_code)
cc = ""
if self.is_const_method:
trailer += " const"
return self.return_type.declaration_code(
"%s%s(%s)%s" % (cc, entity_code, arg_decl_code, trailer),
for_display, dll_linkage, pyrex)
def function_header_code(self, func_name, arg_code):
if self.is_const_method:
trailer = " const"
else:
trailer = ""
return "%s%s(%s)%s" % (self.calling_convention_prefix(),
func_name, arg_code, trailer)
def signature_string(self):
s = self.empty_declaration_code()
return s
def signature_cast_string(self):
s = self.declaration_code("(*)", with_calling_convention=False)
return '(%s)' % s
def specialize(self, values):
result = CFuncType(self.return_type.specialize(values),
[arg.specialize(values) for arg in self.args],
has_varargs = self.has_varargs,
exception_value = self.exception_value,
exception_check = self.exception_check,
calling_convention = self.calling_convention,
nogil = self.nogil,
with_gil = self.with_gil,
is_overridable = self.is_overridable,
optional_arg_count = self.optional_arg_count,
is_const_method = self.is_const_method,
is_static_method = self.is_static_method,
templates = self.templates)
result.from_fused = self.is_fused
return result
def opt_arg_cname(self, arg_name):
return self.op_arg_struct.base_type.scope.lookup(arg_name).cname
# Methods that deal with Fused Types
# All but map_with_specific_entries should be called only on functions
# with fused types (and not on their corresponding specific versions).
def get_all_specialized_permutations(self, fused_types=None):
"""
Permute all the types. For every specific instance of a fused type, we
want all other specific instances of all other fused types.
It returns an iterable of two-tuples of the cname that should prefix
the cname of the function, and a dict mapping any fused types to their
respective specific types.
"""
assert self.is_fused
if fused_types is None:
fused_types = self.get_fused_types()
return get_all_specialized_permutations(fused_types)
def get_all_specialized_function_types(self):
"""
Get all the specific function types of this one.
"""
assert self.is_fused
if self.entry.fused_cfunction:
return [n.type for n in self.entry.fused_cfunction.nodes]
elif self.cached_specialized_types is not None:
return self.cached_specialized_types
result = []
permutations = self.get_all_specialized_permutations()
new_cfunc_entries = []
for cname, fused_to_specific in permutations:
new_func_type = self.entry.type.specialize(fused_to_specific)
if self.optional_arg_count:
# Remember, this method is set by CFuncDeclaratorNode
self.declare_opt_arg_struct(new_func_type, cname)
new_entry = copy.deepcopy(self.entry)
new_func_type.specialize_entry(new_entry, cname)
new_entry.type = new_func_type
new_func_type.entry = new_entry
result.append(new_func_type)
new_cfunc_entries.append(new_entry)
cfunc_entries = self.entry.scope.cfunc_entries
try:
cindex = cfunc_entries.index(self.entry)
except ValueError:
cfunc_entries.extend(new_cfunc_entries)
else:
cfunc_entries[cindex:cindex+1] = new_cfunc_entries
self.cached_specialized_types = result
return result
def get_fused_types(self, result=None, seen=None, subtypes=None, include_function_return_type=False):
"""Return fused types in the order they appear as parameter types"""
return super(CFuncType, self).get_fused_types(
result, seen,
# for function pointer types, we consider the result type; for plain function
# types we don't (because it must be derivable from the arguments)
subtypes=self.subtypes if include_function_return_type else ['args'])
def specialize_entry(self, entry, cname):
assert not self.is_fused
specialize_entry(entry, cname)
def can_coerce_to_pyobject(self, env):
# duplicating the decisions from create_to_py_utility_code() here avoids writing out unused code
if self.has_varargs or self.optional_arg_count:
return False
if self.to_py_function is not None:
return self.to_py_function
for arg in self.args:
if not arg.type.is_pyobject and not arg.type.can_coerce_to_pyobject(env):
return False
if not self.return_type.is_pyobject and not self.return_type.can_coerce_to_pyobject(env):
return False
return True
def create_to_py_utility_code(self, env):
# FIXME: it seems we're trying to coerce in more cases than we should
if self.to_py_function is not None:
return self.to_py_function
if not self.can_coerce_to_pyobject(env):
return False
from .UtilityCode import CythonUtilityCode
# include argument names into the c function name to ensure cname is unique
# between functions with identical types but different argument names
from .Symtab import punycodify_name
def arg_name_part(arg):
return "%s%s" % (len(arg.name), punycodify_name(arg.name)) if arg.name else "0"
arg_names = [ arg_name_part(arg) for arg in self.args ]
arg_names = cap_length("_".join(arg_names))
safe_typename = type_identifier(self, pyrex=True)
# Note that the length here is slightly bigger than twice the default cap in
# "cap_length" (since the length is capped in both arg_names and the type_identifier)
# but since this is significantly shorter than compilers should be able to handle,
# that is acceptable.
to_py_function = "__Pyx_CFunc_%s_to_py_%s" % (safe_typename, arg_names)
for arg in self.args:
if not arg.type.is_pyobject and not arg.type.create_from_py_utility_code(env):
return False
if not self.return_type.is_pyobject and not self.return_type.create_to_py_utility_code(env):
return False
def declared_type(ctype):
type_displayname = str(ctype.declaration_code("", for_display=True))
if ctype.is_pyobject:
arg_ctype = type_name = type_displayname
if ctype.is_builtin_type:
arg_ctype = ctype.name
elif not ctype.is_extension_type:
type_name = 'object'
type_displayname = None
else:
type_displayname = repr(type_displayname)
elif ctype is c_bint_type:
type_name = arg_ctype = 'bint'
else:
type_name = arg_ctype = type_displayname
if ctype is c_double_type:
type_displayname = 'float'
else:
type_displayname = repr(type_displayname)
return type_name, arg_ctype, type_displayname
class Arg(object):
def __init__(self, arg_name, arg_type):
self.name = arg_name
self.type = arg_type
self.type_cname, self.ctype, self.type_displayname = declared_type(arg_type)
if self.return_type.is_void:
except_clause = 'except *'
elif self.return_type.is_pyobject:
except_clause = ''
elif self.exception_value:
except_clause = ('except? %s' if self.exception_check else 'except %s') % self.exception_value
else:
except_clause = 'except *'
context = {
'cname': to_py_function,
'args': [Arg(arg.name or 'arg%s' % ix, arg.type) for ix, arg in enumerate(self.args)],
'return_type': Arg('return', self.return_type),
'except_clause': except_clause,
}
# FIXME: directives come from first defining environment and do not adapt for reuse
env.use_utility_code(CythonUtilityCode.load(
"cfunc.to_py", "CConvert.pyx",
outer_module_scope=env.global_scope(), # need access to types declared in module
context=context, compiler_directives=dict(env.global_scope().directives)))
self.to_py_function = to_py_function
return True
def specialize_entry(entry, cname):
"""
Specialize an entry of a copied fused function or method
"""
entry.is_fused_specialized = True
entry.name = get_fused_cname(cname, entry.name)
if entry.is_cmethod:
entry.cname = entry.name
if entry.is_inherited:
entry.cname = StringEncoding.EncodedString(
"%s.%s" % (Naming.obj_base_cname, entry.cname))
else:
entry.cname = get_fused_cname(cname, entry.cname)
if entry.func_cname:
entry.func_cname = get_fused_cname(cname, entry.func_cname)
if entry.final_func_cname:
entry.final_func_cname = get_fused_cname(cname, entry.final_func_cname)
def get_fused_cname(fused_cname, orig_cname):
"""
Given the fused cname id and an original cname, return a specialized cname
"""
assert fused_cname and orig_cname
return StringEncoding.EncodedString('%s%s%s' % (Naming.fused_func_prefix,
fused_cname, orig_cname))
def unique(somelist):
seen = set()
result = []
for obj in somelist:
if obj not in seen:
result.append(obj)
seen.add(obj)
return result
def get_all_specialized_permutations(fused_types):
return _get_all_specialized_permutations(unique(fused_types))
def _get_all_specialized_permutations(fused_types, id="", f2s=()):
fused_type, = fused_types[0].get_fused_types()
result = []
for newid, specific_type in enumerate(fused_type.types):
# f2s = dict(f2s, **{ fused_type: specific_type })
f2s = dict(f2s)
f2s.update({ fused_type: specific_type })
if id:
cname = '%s_%s' % (id, newid)
else:
cname = str(newid)
if len(fused_types) > 1:
result.extend(_get_all_specialized_permutations(
fused_types[1:], cname, f2s))
else:
result.append((cname, f2s))
return result
def specialization_signature_string(fused_compound_type, fused_to_specific):
"""
Return the signature for a specialization of a fused type. e.g.
floating[:] ->
'float' or 'double'
cdef fused ft:
float[:]
double[:]
ft ->
'float[:]' or 'double[:]'
integral func(floating) ->
'int (*func)(float)' or ...
"""
fused_types = fused_compound_type.get_fused_types()
if len(fused_types) == 1:
fused_type = fused_types[0]
else:
fused_type = fused_compound_type
return fused_type.specialize(fused_to_specific).typeof_name()
def get_specialized_types(type):
"""
Return a list of specialized types in their declared order.
"""
assert type.is_fused
if isinstance(type, FusedType):
result = list(type.types)
for specialized_type in result:
specialized_type.specialization_string = specialized_type.typeof_name()
else:
result = []
for cname, f2s in get_all_specialized_permutations(type.get_fused_types()):
specialized_type = type.specialize(f2s)
specialized_type.specialization_string = (
specialization_signature_string(type, f2s))
result.append(specialized_type)
return result
class CFuncTypeArg(BaseType):
# name string
# cname string
# type PyrexType
# pos source file position
# FIXME: is this the right setup? should None be allowed here?
not_none = False
or_none = False
accept_none = True
accept_builtin_subtypes = False
annotation = None
subtypes = ['type']
def __init__(self, name, type, pos, cname=None, annotation=None):
self.name = name
if cname is not None:
self.cname = cname
else:
self.cname = Naming.var_prefix + name
if annotation is not None:
self.annotation = annotation
self.type = type
self.pos = pos
self.needs_type_test = False # TODO: should these defaults be set in analyse_types()?
def __repr__(self):
return "%s:%s" % (self.name, repr(self.type))
def declaration_code(self, for_display = 0):
return self.type.declaration_code(self.cname, for_display)
def specialize(self, values):
return CFuncTypeArg(self.name, self.type.specialize(values), self.pos, self.cname)
def is_forwarding_reference(self):
if self.type.is_rvalue_reference:
if (isinstance(self.type.ref_base_type, TemplatePlaceholderType)
and not self.type.ref_base_type.is_cv_qualified):
return True
return False
class ToPyStructUtilityCode(object):
requires = None
def __init__(self, type, forward_decl, env):
self.type = type
self.header = "static PyObject* %s(%s)" % (type.to_py_function,
type.declaration_code('s'))
self.forward_decl = forward_decl
self.env = env
def __eq__(self, other):
return isinstance(other, ToPyStructUtilityCode) and self.header == other.header
def __hash__(self):
return hash(self.header)
def get_tree(self, **kwargs):
pass
def put_code(self, output):
code = output['utility_code_def']
proto = output['utility_code_proto']
code.putln("%s {" % self.header)
code.putln("PyObject* res;")
code.putln("PyObject* member;")
code.putln("res = __Pyx_PyDict_NewPresized(%d); if (unlikely(!res)) return NULL;" %
len(self.type.scope.var_entries))
for member in self.type.scope.var_entries:
nameconst_cname = code.get_py_string_const(member.name, identifier=True)
code.putln("%s; if (unlikely(!member)) goto bad;" % (
member.type.to_py_call_code('s.%s' % member.cname, 'member', member.type)))
code.putln("if (unlikely(PyDict_SetItem(res, %s, member) < 0)) goto bad;" % nameconst_cname)
code.putln("Py_DECREF(member);")
code.putln("return res;")
code.putln("bad:")
code.putln("Py_XDECREF(member);")
code.putln("Py_DECREF(res);")
code.putln("return NULL;")
code.putln("}")
# This is a bit of a hack, we need a forward declaration
# due to the way things are ordered in the module...
if self.forward_decl:
proto.putln(self.type.empty_declaration_code() + ';')
proto.putln(self.header + ";")
def inject_tree_and_scope_into(self, module_node):
pass
class CStructOrUnionType(CType):
# name string
# cname string
# kind string "struct" or "union"
# scope StructOrUnionScope, or None if incomplete
# typedef_flag boolean
# packed boolean
# entry Entry
is_struct_or_union = 1
has_attributes = 1
exception_check = True
def __init__(self, name, kind, scope, typedef_flag, cname, packed=False, in_cpp=False):
self.name = name
self.cname = cname
self.kind = kind
self.scope = scope
self.typedef_flag = typedef_flag
self.is_struct = kind == 'struct'
self.to_py_function = "%s_to_py_%s" % (
Naming.convert_func_prefix, self.specialization_name())
self.from_py_function = "%s_from_py_%s" % (
Naming.convert_func_prefix, self.specialization_name())
self.exception_check = True
self._convert_to_py_code = None
self._convert_from_py_code = None
self.packed = packed
self.needs_cpp_construction = self.is_struct and in_cpp
def can_coerce_to_pyobject(self, env):
if self._convert_to_py_code is False:
return None # tri-state-ish
if env.outer_scope is None:
return False
if self._convert_to_py_code is None:
is_union = not self.is_struct
unsafe_union_types = set()
safe_union_types = set()
for member in self.scope.var_entries:
member_type = member.type
if not member_type.can_coerce_to_pyobject(env):
self.to_py_function = None
self._convert_to_py_code = False
return False
if is_union:
if member_type.is_ptr or member_type.is_cpp_class:
unsafe_union_types.add(member_type)
else:
safe_union_types.add(member_type)
if unsafe_union_types and (safe_union_types or len(unsafe_union_types) > 1):
# unsafe mix of safe and unsafe to convert types
self.from_py_function = None
self._convert_from_py_code = False
return False
return True
def create_to_py_utility_code(self, env):
if not self.can_coerce_to_pyobject(env):
return False
if self._convert_to_py_code is None:
for member in self.scope.var_entries:
member.type.create_to_py_utility_code(env)
forward_decl = self.entry.visibility != 'extern' and not self.typedef_flag
self._convert_to_py_code = ToPyStructUtilityCode(self, forward_decl, env)
env.use_utility_code(self._convert_to_py_code)
return True
def can_coerce_from_pyobject(self, env):
if env.outer_scope is None or self._convert_from_py_code is False:
return False
for member in self.scope.var_entries:
if not member.type.can_coerce_from_pyobject(env):
return False
return True
def create_from_py_utility_code(self, env):
if env.outer_scope is None:
return False
if self._convert_from_py_code is False:
return None # tri-state-ish
if self._convert_from_py_code is None:
if not self.scope.var_entries:
# There are obviously missing fields; don't allow instantiation
# where absolutely no content is provided.
return False
for member in self.scope.var_entries:
if not member.type.create_from_py_utility_code(env):
self.from_py_function = None
self._convert_from_py_code = False
return False
context = dict(
struct_type=self,
var_entries=self.scope.var_entries,
funcname=self.from_py_function,
)
env.use_utility_code(UtilityCode.load_cached("RaiseUnexpectedTypeError", "ObjectHandling.c"))
from .UtilityCode import CythonUtilityCode
self._convert_from_py_code = CythonUtilityCode.load(
"FromPyStructUtility" if self.is_struct else "FromPyUnionUtility",
"CConvert.pyx",
outer_module_scope=env.global_scope(), # need access to types declared in module
context=context)
env.use_utility_code(self._convert_from_py_code)
return True
def __repr__(self):
return "<CStructOrUnionType %s %s%s>" % (
self.name, self.cname,
("", " typedef")[self.typedef_flag])
def declaration_code(self, entity_code,
for_display=0, dll_linkage=None, pyrex=0):
if pyrex or for_display:
base_code = self.name
else:
if self.typedef_flag:
base_code = self.cname
else:
base_code = "%s %s" % (self.kind, self.cname)
base_code = public_decl(base_code, dll_linkage)
return self.base_declaration_code(base_code, entity_code)
def __eq__(self, other):
try:
return (isinstance(other, CStructOrUnionType) and
self.name == other.name)
except AttributeError:
return False
def __lt__(self, other):
try:
return self.name < other.name
except AttributeError:
# this is arbitrary, but it makes sure we always have
# *some* kind of order
return False
def __hash__(self):
return hash(self.cname) ^ hash(self.kind)
def is_complete(self):
return self.scope is not None
def attributes_known(self):
return self.is_complete()
def can_be_complex(self):
# Does the struct consist of exactly two identical floats?
fields = self.scope.var_entries
if len(fields) != 2: return False
a, b = fields
return (a.type.is_float and b.type.is_float and
a.type.empty_declaration_code() ==
b.type.empty_declaration_code())
def struct_nesting_depth(self):
child_depths = [x.type.struct_nesting_depth()
for x in self.scope.var_entries]
return max(child_depths) + 1
def cast_code(self, expr_code):
if self.is_struct:
return expr_code
return super(CStructOrUnionType, self).cast_code(expr_code)
cpp_string_conversions = ("std::string",)
builtin_cpp_conversions = {
# type element template params
"std::pair": 2,
"std::vector": 1,
"std::list": 1,
"std::set": 1,
"std::unordered_set": 1,
"std::map": 2,
"std::unordered_map": 2,
"std::complex": 1,
}
class CppClassType(CType):
# name string
# cname string
# scope CppClassScope
# templates [string] or None
is_cpp_class = 1
has_attributes = 1
needs_cpp_construction = 1
exception_check = True
namespace = None
# For struct-like declaration.
kind = "struct"
packed = False
typedef_flag = False
subtypes = ['templates']
def __init__(self, name, scope, cname, base_classes, templates=None, template_type=None):
self.name = name
self.cname = cname
self.scope = scope
self.base_classes = base_classes
self.operators = []
self.templates = templates
self.template_type = template_type
self.num_optional_templates = sum(is_optional_template_param(T) for T in templates or ())
if templates:
self.specializations = {tuple(zip(templates, templates)): self}
else:
self.specializations = {}
self.is_cpp_string = cname in cpp_string_conversions
def use_conversion_utility(self, from_or_to):
pass
def maybe_unordered(self):
if 'unordered' in self.cname:
return 'unordered_'
else:
return ''
def can_coerce_from_pyobject(self, env):
if self.cname in builtin_cpp_conversions:
template_count = builtin_cpp_conversions[self.cname]
for ix, T in enumerate(self.templates or []):
if ix >= template_count:
break
if T.is_pyobject or not T.can_coerce_from_pyobject(env):
return False
return True
elif self.cname in cpp_string_conversions:
return True
return False
def create_from_py_utility_code(self, env):
if self.from_py_function is not None:
return True
if self.cname in builtin_cpp_conversions or self.cname in cpp_string_conversions:
X = "XYZABC"
tags = []
context = {}
for ix, T in enumerate(self.templates or []):
if ix >= builtin_cpp_conversions[self.cname]:
break
if T.is_pyobject or not T.create_from_py_utility_code(env):
return False
tags.append(T.specialization_name())
context[X[ix]] = T
if self.cname in cpp_string_conversions:
cls = 'string'
tags = type_identifier(self),
else:
cls = self.cname[5:]
cname = '__pyx_convert_%s_from_py_%s' % (cls, '__and_'.join(tags))
context.update({
'cname': cname,
'maybe_unordered': self.maybe_unordered(),
'type': self.cname,
})
# Override directives that should not be inherited from user code.
from .UtilityCode import CythonUtilityCode
directives = CythonUtilityCode.filter_inherited_directives(env.directives)
env.use_utility_code(CythonUtilityCode.load(
cls.replace('unordered_', '') + ".from_py", "CppConvert.pyx",
context=context, compiler_directives=directives))
self.from_py_function = cname
return True
def can_coerce_to_pyobject(self, env):
if self.cname in builtin_cpp_conversions or self.cname in cpp_string_conversions:
for ix, T in enumerate(self.templates or []):
if ix >= builtin_cpp_conversions[self.cname]:
break
if T.is_pyobject or not T.can_coerce_to_pyobject(env):
return False
return True
def create_to_py_utility_code(self, env):
if self.to_py_function is not None:
return True
if self.cname in builtin_cpp_conversions or self.cname in cpp_string_conversions:
X = "XYZABC"
tags = []
context = {}
for ix, T in enumerate(self.templates or []):
if ix >= builtin_cpp_conversions[self.cname]:
break
if not T.create_to_py_utility_code(env):
return False
tags.append(T.specialization_name())
context[X[ix]] = T
if self.cname in cpp_string_conversions:
cls = 'string'
prefix = 'PyObject_' # gets specialised by explicit type casts in CoerceToPyTypeNode
tags = type_identifier(self),
else:
cls = self.cname[5:]
prefix = ''
cname = "__pyx_convert_%s%s_to_py_%s" % (prefix, cls, "____".join(tags))
context.update({
'cname': cname,
'maybe_unordered': self.maybe_unordered(),
'type': self.cname,
})
from .UtilityCode import CythonUtilityCode
# Override directives that should not be inherited from user code.
directives = CythonUtilityCode.filter_inherited_directives(env.directives)
env.use_utility_code(CythonUtilityCode.load(
cls.replace('unordered_', '') + ".to_py", "CppConvert.pyx",
context=context, compiler_directives=directives))
self.to_py_function = cname
return True
def is_template_type(self):
return self.templates is not None and self.template_type is None
def get_fused_types(self, result=None, seen=None, include_function_return_type=False):
if result is None:
result = []
seen = set()
if self.namespace:
self.namespace.get_fused_types(result, seen)
if self.templates:
for T in self.templates:
T.get_fused_types(result, seen)
return result
def specialize_here(self, pos, env, template_values=None):
if not self.is_template_type():
error(pos, "'%s' type is not a template" % self)
return error_type
if len(self.templates) - self.num_optional_templates <= len(template_values) < len(self.templates):
num_defaults = len(self.templates) - len(template_values)
partial_specialization = self.declaration_code('', template_params=template_values)
# Most of the time we don't need to declare anything typed to these
# default template arguments, but when we do there's no way in C++
# to reference this directly. However, it is common convention to
# provide a typedef in the template class that resolves to each
# template type. For now, allow the user to specify this name as
# the template parameter.
# TODO: Allow typedefs in cpp classes and search for it in this
# classes scope as a concrete name we could use.
template_values = template_values + [
TemplatePlaceholderType(
"%s::%s" % (partial_specialization, param.name), True)
for param in self.templates[-num_defaults:]]
if len(self.templates) != len(template_values):
error(pos, "%s templated type receives %d arguments, got %d" %
(self.name, len(self.templates), len(template_values)))
return error_type
has_object_template_param = False
for value in template_values:
if value.is_pyobject or value.needs_refcounting:
has_object_template_param = True
type_description = "Python object" if value.is_pyobject else "Reference-counted"
error(pos,
"%s type '%s' cannot be used as a template argument" % (
type_description, value))
if has_object_template_param:
return error_type
return self.specialize(dict(zip(self.templates, template_values)))
def specialize(self, values):
if not self.templates and not self.namespace:
return self
if self.templates is None:
self.templates = []
key = tuple(values.items())
if key in self.specializations:
return self.specializations[key]
template_values = [t.specialize(values) for t in self.templates]
specialized = self.specializations[key] = \
CppClassType(self.name, None, self.cname, [], template_values, template_type=self)
# Need to do these *after* self.specializations[key] is set
# to avoid infinite recursion on circular references.
specialized.base_classes = [b.specialize(values) for b in self.base_classes]
if self.namespace is not None:
specialized.namespace = self.namespace.specialize(values)
specialized.scope = self.scope.specialize(values, specialized)
if self.cname == 'std::vector':
# vector<bool> is special cased in the C++ standard, and its
# accessors do not necessarily return references to the underlying
# elements (which may be bit-packed).
# http://www.cplusplus.com/reference/vector/vector-bool/
# Here we pretend that the various methods return bool values
# (as the actual returned values are coercible to such, and
# we don't support call expressions as lvalues).
T = values.get(self.templates[0], None)
if T and not T.is_fused and T.empty_declaration_code() == 'bool':
for bit_ref_returner in ('at', 'back', 'front'):
if bit_ref_returner in specialized.scope.entries:
specialized.scope.entries[bit_ref_returner].type.return_type = T
return specialized
def deduce_template_params(self, actual):
if actual.is_cv_qualified:
actual = actual.cv_base_type
if actual.is_reference:
actual = actual.ref_base_type
if self == actual:
return {}
elif actual.is_cpp_class:
self_template_type = self
while getattr(self_template_type, 'template_type', None):
self_template_type = self_template_type.template_type
def all_bases(cls):
yield cls
for parent in cls.base_classes:
for base in all_bases(parent):
yield base
for actual_base in all_bases(actual):
template_type = actual_base
while getattr(template_type, 'template_type', None):
template_type = template_type.template_type
if (self_template_type.empty_declaration_code()
== template_type.empty_declaration_code()):
return reduce(
merge_template_deductions,
[formal_param.deduce_template_params(actual_param)
for (formal_param, actual_param)
in zip(self.templates, actual_base.templates)],
{})
else:
return {}
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0,
template_params = None):
if template_params is None:
template_params = self.templates
if self.templates:
template_strings = [param.declaration_code('', for_display, None, pyrex)
for param in template_params
if not is_optional_template_param(param) and not param.is_fused]
if for_display:
brackets = "[%s]"
else:
brackets = "<%s> "
templates = brackets % ",".join(template_strings)
else:
templates = ""
if pyrex or for_display:
base_code = "%s%s" % (self.name, templates)
else:
base_code = "%s%s" % (self.cname, templates)
if self.namespace is not None:
base_code = "%s::%s" % (self.namespace.empty_declaration_code(), base_code)
base_code = public_decl(base_code, dll_linkage)
return self.base_declaration_code(base_code, entity_code)
def cpp_optional_declaration_code(self, entity_code, dll_linkage=None, template_params=None):
return "__Pyx_Optional_Type<%s> %s" % (
self.declaration_code("", False, dll_linkage, False,
template_params),
entity_code)
def is_subclass(self, other_type):
if self.same_as_resolved_type(other_type):
return 1
for base_class in self.base_classes:
if base_class.is_subclass(other_type):
return 1
return 0
def subclass_dist(self, super_type):
if self.same_as_resolved_type(super_type):
return 0
elif not self.base_classes:
return float('inf')
else:
return 1 + min(b.subclass_dist(super_type) for b in self.base_classes)
def same_as_resolved_type(self, other_type):
if other_type.is_cpp_class:
if self == other_type:
return 1
# This messy logic is needed due to GH Issue #1852.
elif (self.cname == other_type.cname and
(self.template_type and other_type.template_type
or self.templates
or other_type.templates)):
if self.templates == other_type.templates:
return 1
for t1, t2 in zip(self.templates, other_type.templates):
if is_optional_template_param(t1) and is_optional_template_param(t2):
break
if not t1.same_as_resolved_type(t2):
return 0
return 1
return 0
def assignable_from_resolved_type(self, other_type):
# TODO: handle operator=(...) here?
if other_type is error_type:
return True
elif other_type.is_cpp_class:
return other_type.is_subclass(self)
elif other_type.is_string and self.cname in cpp_string_conversions:
return True
def attributes_known(self):
return self.scope is not None
def find_cpp_operation_type(self, operator, operand_type=None):
operands = [self]
if operand_type is not None:
operands.append(operand_type)
# pos == None => no errors
operator_entry = self.scope.lookup_operator_for_types(None, operator, operands)
if not operator_entry:
return None
func_type = operator_entry.type
if func_type.is_ptr:
func_type = func_type.base_type
return func_type.return_type
def get_constructor(self, pos):
constructor = self.scope.lookup('<init>')
if constructor is not None:
return constructor
# Otherwise: automatically declare no-args default constructor.
# Make it "nogil" if the base classes allow it.
nogil = True
for base in self.base_classes:
base_constructor = base.scope.lookup('<init>')
if base_constructor and not base_constructor.type.nogil:
nogil = False
break
func_type = CFuncType(self, [], exception_check='+', nogil=nogil)
return self.scope.declare_cfunction(u'<init>', func_type, pos)
def check_nullary_constructor(self, pos, msg="stack allocated"):
constructor = self.scope.lookup(u'<init>')
if constructor is not None and best_match([], constructor.all_alternatives()) is None:
error(pos, "C++ class must have a nullary constructor to be %s" % msg)
def cpp_optional_check_for_null_code(self, cname):
# only applies to c++ classes that are being declared as std::optional
return "(%s.has_value())" % cname
class EnumMixin(object):
"""
Common implementation details for C and C++ enums.
"""
def create_enum_to_py_utility_code(self, env):
from .UtilityCode import CythonUtilityCode
self.to_py_function = "__Pyx_Enum_%s_to_py" % type_identifier(self)
if self.entry.scope != env.global_scope():
module_name = self.entry.scope.qualified_name
else:
module_name = None
directives = CythonUtilityCode.filter_inherited_directives(
env.global_scope().directives)
if any(value_entry.enum_int_value is None for value_entry in self.entry.enum_values):
# We're at a high risk of making a switch statement with equal values in
# (because we simply can't tell, and enums are often used like that).
# So turn off the switch optimization to be safe.
# (Note that for now Cython doesn't do the switch optimization for
# scoped enums anyway)
directives['optimize.use_switch'] = False
if self.is_cpp_enum:
underlying_type_str = self.underlying_type.empty_declaration_code()
else:
underlying_type_str = "int"
env.use_utility_code(CythonUtilityCode.load(
"EnumTypeToPy", "CpdefEnums.pyx",
context={"funcname": self.to_py_function,
"name": self.name,
"items": tuple(self.values),
"underlying_type": underlying_type_str,
"module_name": module_name,
"is_flag": not self.is_cpp_enum,
},
outer_module_scope=self.entry.scope, # ensure that "name" is findable
compiler_directives = directives,
))
class CppScopedEnumType(CType, EnumMixin):
# name string
# doc string or None
# cname string
is_cpp_enum = True
def __init__(self, name, cname, underlying_type, namespace=None, doc=None):
self.name = name
self.doc = doc
self.cname = cname
self.values = []
self.underlying_type = underlying_type
self.namespace = namespace
def __str__(self):
return self.name
def declaration_code(self, entity_code,
for_display=0, dll_linkage=None, pyrex=0):
if pyrex or for_display:
type_name = self.name
else:
if self.namespace:
type_name = "%s::%s" % (
self.namespace.empty_declaration_code(),
self.cname
)
else:
type_name = "__PYX_ENUM_CLASS_DECL %s" % self.cname
type_name = public_decl(type_name, dll_linkage)
return self.base_declaration_code(type_name, entity_code)
def create_from_py_utility_code(self, env):
if self.from_py_function:
return True
if self.underlying_type.create_from_py_utility_code(env):
self.from_py_function = '(%s)%s' % (
self.cname, self.underlying_type.from_py_function
)
return True
def create_to_py_utility_code(self, env):
if self.to_py_function is not None:
return True
if self.entry.create_wrapper:
self.create_enum_to_py_utility_code(env)
return True
if self.underlying_type.create_to_py_utility_code(env):
# Using a C++11 lambda here, which is fine since
# scoped enums are a C++11 feature
self.to_py_function = '[](const %s& x){return %s((%s)x);}' % (
self.cname,
self.underlying_type.to_py_function,
self.underlying_type.empty_declaration_code()
)
return True
def create_type_wrapper(self, env):
from .UtilityCode import CythonUtilityCode
rst = CythonUtilityCode.load(
"CppScopedEnumType", "CpdefEnums.pyx",
context={
"name": self.name,
"cname": self.cname.split("::")[-1],
"items": tuple(self.values),
"underlying_type": self.underlying_type.empty_declaration_code(),
"enum_doc": self.doc,
"static_modname": env.qualified_name,
},
outer_module_scope=env.global_scope())
env.use_utility_code(rst)
class TemplatePlaceholderType(CType):
def __init__(self, name, optional=False):
self.name = name
self.optional = optional
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if entity_code:
return self.name + " " + entity_code
else:
return self.name
def specialize(self, values):
if self in values:
return values[self]
else:
return self
def deduce_template_params(self, actual):
return {self: actual}
def same_as_resolved_type(self, other_type):
if isinstance(other_type, TemplatePlaceholderType):
return self.name == other_type.name
else:
return 0
def __hash__(self):
return hash(self.name)
def __cmp__(self, other):
if isinstance(other, TemplatePlaceholderType):
return cmp(self.name, other.name)
else:
return cmp(type(self), type(other))
def __eq__(self, other):
if isinstance(other, TemplatePlaceholderType):
return self.name == other.name
else:
return False
def is_optional_template_param(type):
return isinstance(type, TemplatePlaceholderType) and type.optional
class CEnumType(CIntLike, CType, EnumMixin):
# name string
# doc string or None
# cname string or None
# typedef_flag boolean
# values [string], populated during declaration analysis
is_enum = 1
signed = 1
rank = -1 # Ranks below any integer type
def __init__(self, name, cname, typedef_flag, namespace=None, doc=None):
self.name = name
self.doc = doc
self.cname = cname
self.values = []
self.typedef_flag = typedef_flag
self.namespace = namespace
self.default_value = "(%s) 0" % self.empty_declaration_code()
def __str__(self):
return self.name
def __repr__(self):
return "<CEnumType %s %s%s>" % (self.name, self.cname,
("", " typedef")[self.typedef_flag])
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if pyrex or for_display:
base_code = self.name
else:
if self.namespace:
base_code = "%s::%s" % (
self.namespace.empty_declaration_code(), self.cname)
elif self.typedef_flag:
base_code = self.cname
else:
base_code = "enum %s" % self.cname
base_code = public_decl(base_code, dll_linkage)
return self.base_declaration_code(base_code, entity_code)
def specialize(self, values):
if self.namespace:
namespace = self.namespace.specialize(values)
if namespace != self.namespace:
return CEnumType(
self.name, self.cname, self.typedef_flag, namespace)
return self
def create_type_wrapper(self, env):
from .UtilityCode import CythonUtilityCode
# Generate "int"-like conversion function
old_to_py_function = self.to_py_function
self.to_py_function = None
CIntLike.create_to_py_utility_code(self, env)
enum_to_pyint_func = self.to_py_function
self.to_py_function = old_to_py_function # we don't actually want to overwrite this
env.use_utility_code(CythonUtilityCode.load(
"EnumType", "CpdefEnums.pyx",
context={"name": self.name,
"items": tuple(self.values),
"enum_doc": self.doc,
"enum_to_pyint_func": enum_to_pyint_func,
"static_modname": env.qualified_name,
},
outer_module_scope=env.global_scope()))
def create_to_py_utility_code(self, env):
if self.to_py_function is not None:
return self.to_py_function
if not self.entry.create_wrapper:
return super(CEnumType, self).create_to_py_utility_code(env)
self.create_enum_to_py_utility_code(env)
return True
class CTupleType(CType):
# components [PyrexType]
is_ctuple = True
def __init__(self, cname, components):
self.cname = cname
self.components = components
self.size = len(components)
self.to_py_function = "%s_to_py_%s" % (Naming.convert_func_prefix, self.cname)
self.from_py_function = "%s_from_py_%s" % (Naming.convert_func_prefix, self.cname)
self.exception_check = True
self._convert_to_py_code = None
self._convert_from_py_code = None
# equivalent_type must be set now because it isn't available at import time
from .Builtin import tuple_type
self.equivalent_type = tuple_type
def __str__(self):
return "(%s)" % ", ".join(str(c) for c in self.components)
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
if pyrex or for_display:
return "%s %s" % (str(self), entity_code)
else:
return self.base_declaration_code(self.cname, entity_code)
def can_coerce_to_pyobject(self, env):
for component in self.components:
if not component.can_coerce_to_pyobject(env):
return False
return True
def can_coerce_from_pyobject(self, env):
for component in self.components:
if not component.can_coerce_from_pyobject(env):
return False
return True
def create_to_py_utility_code(self, env):
if self._convert_to_py_code is False:
return None # tri-state-ish
if self._convert_to_py_code is None:
for component in self.components:
if not component.create_to_py_utility_code(env):
self.to_py_function = None
self._convert_to_py_code = False
return False
context = dict(
struct_type_decl=self.empty_declaration_code(),
components=self.components,
funcname=self.to_py_function,
size=len(self.components)
)
self._convert_to_py_code = TempitaUtilityCode.load(
"ToPyCTupleUtility", "TypeConversion.c", context=context)
env.use_utility_code(self._convert_to_py_code)
return True
def create_from_py_utility_code(self, env):
if self._convert_from_py_code is False:
return None # tri-state-ish
if self._convert_from_py_code is None:
for component in self.components:
if not component.create_from_py_utility_code(env):
self.from_py_function = None
self._convert_from_py_code = False
return False
context = dict(
struct_type_decl=self.empty_declaration_code(),
components=self.components,
funcname=self.from_py_function,
size=len(self.components)
)
self._convert_from_py_code = TempitaUtilityCode.load(
"FromPyCTupleUtility", "TypeConversion.c", context=context)
env.use_utility_code(self._convert_from_py_code)
return True
def cast_code(self, expr_code):
return expr_code
def c_tuple_type(components):
components = tuple(components)
cname = Naming.ctuple_type_prefix + type_list_identifier(components)
tuple_type = CTupleType(cname, components)
return tuple_type
class UnspecifiedType(PyrexType):
# Used as a placeholder until the type can be determined.
is_unspecified = 1
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
return "<unspecified>"
def same_as_resolved_type(self, other_type):
return False
class ErrorType(PyrexType):
# Used to prevent propagation of error messages.
is_error = 1
exception_value = "0"
exception_check = 0
to_py_function = "dummy"
from_py_function = "dummy"
def create_to_py_utility_code(self, env):
return True
def create_from_py_utility_code(self, env):
return True
def declaration_code(self, entity_code,
for_display = 0, dll_linkage = None, pyrex = 0):
return "<error>"
def same_as_resolved_type(self, other_type):
return 1
def error_condition(self, result_code):
return "dummy"
class PythonTypeConstructorMixin(object):
"""Used to help Cython interpret indexed types from the typing module (or similar)
"""
modifier_name = None
def set_python_type_constructor_name(self, name):
self.python_type_constructor_name = name
def specialize_here(self, pos, env, template_values=None):
# for a lot of the typing classes it doesn't really matter what the template is
# (i.e. typing.Dict[int] is really just a dict)
return self
def __repr__(self):
if self.base_type:
return "%s[%r]" % (self.name, self.base_type)
else:
return self.name
def is_template_type(self):
return True
class BuiltinTypeConstructorObjectType(BuiltinObjectType, PythonTypeConstructorMixin):
"""
builtin types like list, dict etc which can be subscripted in annotations
"""
def __init__(self, name, cname, objstruct_cname=None):
super(BuiltinTypeConstructorObjectType, self).__init__(
name, cname, objstruct_cname=objstruct_cname)
self.set_python_type_constructor_name(name)
class PythonTupleTypeConstructor(BuiltinTypeConstructorObjectType):
def specialize_here(self, pos, env, template_values=None):
if (template_values and None not in template_values and
not any(v.is_pyobject for v in template_values)):
entry = env.declare_tuple_type(pos, template_values)
if entry:
entry.used = True
return entry.type
return super(PythonTupleTypeConstructor, self).specialize_here(pos, env, template_values)
class SpecialPythonTypeConstructor(PyObjectType, PythonTypeConstructorMixin):
"""
For things like ClassVar, Optional, etc, which are not types and disappear during type analysis.
"""
def __init__(self, name):
super(SpecialPythonTypeConstructor, self).__init__()
self.set_python_type_constructor_name(name)
self.modifier_name = name
def __repr__(self):
return self.name
def resolve(self):
return self
def specialize_here(self, pos, env, template_values=None):
if len(template_values) != 1:
error(pos, "'%s' takes exactly one template argument." % self.name)
return error_type
if template_values[0] is None:
# FIXME: allowing unknown types for now since we don't recognise all Python types.
return None
# Replace this type with the actual 'template' argument.
return template_values[0].resolve()
rank_to_type_name = (
"char", # 0
"short", # 1
"int", # 2
"long", # 3
"PY_LONG_LONG", # 4
"float", # 5
"double", # 6
"long double", # 7
)
RANK_INT = rank_to_type_name.index('int')
RANK_LONG = rank_to_type_name.index('long')
RANK_FLOAT = rank_to_type_name.index('float')
UNSIGNED = 0
SIGNED = 2
error_type = ErrorType()
unspecified_type = UnspecifiedType()
py_object_type = PyObjectType()
c_void_type = CVoidType()
c_uchar_type = CIntType(0, UNSIGNED)
c_ushort_type = CIntType(1, UNSIGNED)
c_uint_type = CIntType(2, UNSIGNED)
c_ulong_type = CIntType(3, UNSIGNED)
c_ulonglong_type = CIntType(4, UNSIGNED)
c_char_type = CIntType(0)
c_short_type = CIntType(1)
c_int_type = CIntType(2)
c_long_type = CIntType(3)
c_longlong_type = CIntType(4)
c_schar_type = CIntType(0, SIGNED)
c_sshort_type = CIntType(1, SIGNED)
c_sint_type = CIntType(2, SIGNED)
c_slong_type = CIntType(3, SIGNED)
c_slonglong_type = CIntType(4, SIGNED)
c_float_type = CFloatType(5, math_h_modifier='f')
c_double_type = CFloatType(6)
c_longdouble_type = CFloatType(7, math_h_modifier='l')
c_float_complex_type = CComplexType(c_float_type)
c_double_complex_type = CComplexType(c_double_type)
c_longdouble_complex_type = CComplexType(c_longdouble_type)
soft_complex_type = SoftCComplexType()
c_anon_enum_type = CAnonEnumType(-1)
c_returncode_type = CReturnCodeType(RANK_INT)
c_bint_type = CBIntType(RANK_INT)
c_py_unicode_type = CPyUnicodeIntType(RANK_INT-0.5, UNSIGNED)
c_py_ucs4_type = CPyUCS4IntType(RANK_LONG-0.5, UNSIGNED)
c_py_hash_t_type = CPyHashTType(RANK_LONG+0.5, SIGNED)
c_py_ssize_t_type = CPySSizeTType(RANK_LONG+0.5, SIGNED)
c_ssize_t_type = CSSizeTType(RANK_LONG+0.5, SIGNED)
c_size_t_type = CSizeTType(RANK_LONG+0.5, UNSIGNED)
c_ptrdiff_t_type = CPtrdiffTType(RANK_LONG+0.75, SIGNED)
c_null_ptr_type = CNullPtrType(c_void_type)
c_void_ptr_type = CPtrType(c_void_type)
c_void_ptr_ptr_type = CPtrType(c_void_ptr_type)
c_char_ptr_type = CPtrType(c_char_type)
c_const_char_ptr_type = CPtrType(CConstType(c_char_type))
c_uchar_ptr_type = CPtrType(c_uchar_type)
c_const_uchar_ptr_type = CPtrType(CConstType(c_uchar_type))
c_char_ptr_ptr_type = CPtrType(c_char_ptr_type)
c_int_ptr_type = CPtrType(c_int_type)
c_py_unicode_ptr_type = CPtrType(c_py_unicode_type)
c_const_py_unicode_ptr_type = CPtrType(CConstType(c_py_unicode_type))
c_py_ssize_t_ptr_type = CPtrType(c_py_ssize_t_type)
c_ssize_t_ptr_type = CPtrType(c_ssize_t_type)
c_size_t_ptr_type = CPtrType(c_size_t_type)
# GIL state
c_gilstate_type = CEnumType("PyGILState_STATE", "PyGILState_STATE", True)
c_threadstate_type = CStructOrUnionType("PyThreadState", "struct", None, 1, "PyThreadState")
c_threadstate_ptr_type = CPtrType(c_threadstate_type)
# PEP-539 "Py_tss_t" type
c_pytss_t_type = CPyTSSTType()
# the Py_buffer type is defined in Builtin.py
c_py_buffer_type = CStructOrUnionType("Py_buffer", "struct", None, 1, "Py_buffer")
c_py_buffer_ptr_type = CPtrType(c_py_buffer_type)
# Not sure whether the unsigned versions and 'long long' should be in there
# long long requires C99 and might be slow, and would always get preferred
# when specialization happens through calling and not indexing
cy_integral_type = FusedType([c_short_type, c_int_type, c_long_type],
name="integral")
# Omitting long double as it might be slow
cy_floating_type = FusedType([c_float_type, c_double_type], name="floating")
cy_numeric_type = FusedType([c_short_type,
c_int_type,
c_long_type,
c_float_type,
c_double_type,
c_float_complex_type,
c_double_complex_type], name="numeric")
# buffer-related structs
c_buf_diminfo_type = CStructOrUnionType("__Pyx_Buf_DimInfo", "struct",
None, 1, "__Pyx_Buf_DimInfo")
c_pyx_buffer_type = CStructOrUnionType("__Pyx_Buffer", "struct", None, 1, "__Pyx_Buffer")
c_pyx_buffer_ptr_type = CPtrType(c_pyx_buffer_type)
c_pyx_buffer_nd_type = CStructOrUnionType("__Pyx_LocalBuf_ND", "struct",
None, 1, "__Pyx_LocalBuf_ND")
cython_memoryview_type = CStructOrUnionType("__pyx_memoryview_obj", "struct",
None, 0, "__pyx_memoryview_obj")
memoryviewslice_type = CStructOrUnionType("memoryviewslice", "struct",
None, 1, "__Pyx_memviewslice")
modifiers_and_name_to_type = {
#(signed, longness, name) : type
(0, 0, "char"): c_uchar_type,
(1, 0, "char"): c_char_type,
(2, 0, "char"): c_schar_type,
(0, -1, "int"): c_ushort_type,
(0, 0, "int"): c_uint_type,
(0, 1, "int"): c_ulong_type,
(0, 2, "int"): c_ulonglong_type,
(1, -1, "int"): c_short_type,
(1, 0, "int"): c_int_type,
(1, 1, "int"): c_long_type,
(1, 2, "int"): c_longlong_type,
(2, -1, "int"): c_sshort_type,
(2, 0, "int"): c_sint_type,
(2, 1, "int"): c_slong_type,
(2, 2, "int"): c_slonglong_type,
(1, 0, "float"): c_float_type,
(1, 0, "double"): c_double_type,
(1, 1, "double"): c_longdouble_type,
(1, 0, "complex"): c_double_complex_type, # C: float, Python: double => Python wins
(1, 0, "floatcomplex"): c_float_complex_type,
(1, 0, "doublecomplex"): c_double_complex_type,
(1, 1, "doublecomplex"): c_longdouble_complex_type,
#
(1, 0, "void"): c_void_type,
(1, 0, "Py_tss_t"): c_pytss_t_type,
(1, 0, "bint"): c_bint_type,
(0, 0, "Py_UNICODE"): c_py_unicode_type,
(0, 0, "Py_UCS4"): c_py_ucs4_type,
(2, 0, "Py_hash_t"): c_py_hash_t_type,
(2, 0, "Py_ssize_t"): c_py_ssize_t_type,
(2, 0, "ssize_t") : c_ssize_t_type,
(0, 0, "size_t") : c_size_t_type,
(2, 0, "ptrdiff_t") : c_ptrdiff_t_type,
(1, 0, "object"): py_object_type,
}
def is_promotion(src_type, dst_type):
# It's hard to find a hard definition of promotion, but empirical
# evidence suggests that the below is all that's allowed.
if src_type.is_numeric:
if dst_type.same_as(c_int_type):
unsigned = (not src_type.signed)
return (src_type.is_enum or
(src_type.is_int and
unsigned + src_type.rank < dst_type.rank))
elif dst_type.same_as(c_double_type):
return src_type.is_float and src_type.rank <= dst_type.rank
return False
def best_match(arg_types, functions, pos=None, env=None, args=None):
"""
Given a list args of arguments and a list of functions, choose one
to call which seems to be the "best" fit for this list of arguments.
This function is used, e.g., when deciding which overloaded method
to dispatch for C++ classes.
We first eliminate functions based on arity, and if only one
function has the correct arity, we return it. Otherwise, we weight
functions based on how much work must be done to convert the
arguments, with the following priorities:
* identical types or pointers to identical types
* promotions
* non-Python types
That is, we prefer functions where no arguments need converted,
and failing that, functions where only promotions are required, and
so on.
If no function is deemed a good fit, or if two or more functions have
the same weight, we return None (as there is no best match). If pos
is not None, we also generate an error.
"""
# TODO: args should be a list of types, not a list of Nodes.
actual_nargs = len(arg_types)
candidates = []
errors = []
for func in functions:
error_mesg = ""
func_type = func.type
if func_type.is_ptr:
func_type = func_type.base_type
# Check function type
if not func_type.is_cfunction:
if not func_type.is_error and pos is not None:
error_mesg = "Calling non-function type '%s'" % func_type
errors.append((func, error_mesg))
continue
# Check no. of args
max_nargs = len(func_type.args)
min_nargs = max_nargs - func_type.optional_arg_count
if actual_nargs < min_nargs or (not func_type.has_varargs and actual_nargs > max_nargs):
if max_nargs == min_nargs and not func_type.has_varargs:
expectation = max_nargs
elif actual_nargs < min_nargs:
expectation = "at least %s" % min_nargs
else:
expectation = "at most %s" % max_nargs
error_mesg = "Call with wrong number of arguments (expected %s, got %s)" \
% (expectation, actual_nargs)
errors.append((func, error_mesg))
continue
if func_type.templates:
# For any argument/parameter pair A/P, if P is a forwarding reference,
# use lvalue-reference-to-A for deduction in place of A when the
# function call argument is an lvalue. See:
# https://en.cppreference.com/w/cpp/language/template_argument_deduction#Deduction_from_a_function_call
arg_types_for_deduction = list(arg_types)
if func.type.is_cfunction and args:
for i, formal_arg in enumerate(func.type.args):
if formal_arg.is_forwarding_reference():
if args[i].is_lvalue():
arg_types_for_deduction[i] = c_ref_type(arg_types[i])
deductions = reduce(
merge_template_deductions,
[pattern.type.deduce_template_params(actual) for (pattern, actual) in zip(func_type.args, arg_types_for_deduction)],
{})
if deductions is None:
errors.append((func, "Unable to deduce type parameters for %s given (%s)" % (
func_type, ', '.join(map(str, arg_types_for_deduction)))))
elif len(deductions) < len(func_type.templates):
errors.append((func, "Unable to deduce type parameter %s" % (
", ".join([param.name for param in set(func_type.templates) - set(deductions.keys())]))))
else:
type_list = [deductions[param] for param in func_type.templates]
from .Symtab import Entry
specialization = Entry(
name = func.name + "[%s]" % ",".join([str(t) for t in type_list]),
cname = func.cname + "<%s>" % ",".join([t.empty_declaration_code() for t in type_list]),
type = func_type.specialize(deductions),
pos = func.pos)
candidates.append((specialization, specialization.type))
else:
candidates.append((func, func_type))
# Optimize the most common case of no overloading...
if len(candidates) == 1:
return candidates[0][0]
elif len(candidates) == 0:
if pos is not None:
func, errmsg = errors[0]
if len(errors) == 1 or [1 for func, e in errors if e == errmsg]:
error(pos, errmsg)
else:
error(pos, "no suitable method found")
return None
possibilities = []
bad_types = []
needed_coercions = {}
for index, (func, func_type) in enumerate(candidates):
score = [0,0,0,0,0,0,0]
for i in range(min(actual_nargs, len(func_type.args))):
src_type = arg_types[i]
dst_type = func_type.args[i].type
assignable = dst_type.assignable_from(src_type)
# Now take care of unprefixed string literals. So when you call a cdef
# function that takes a char *, the coercion will mean that the
# type will simply become bytes. We need to do this coercion
# manually for overloaded and fused functions
if not assignable:
c_src_type = None
if src_type.is_pyobject:
if src_type.is_builtin_type and src_type.name == 'str' and dst_type.resolve().is_string:
c_src_type = dst_type.resolve()
else:
c_src_type = src_type.default_coerced_ctype()
elif src_type.is_pythran_expr:
c_src_type = src_type.org_buffer
if c_src_type is not None:
assignable = dst_type.assignable_from(c_src_type)
if assignable:
src_type = c_src_type
needed_coercions[func] = (i, dst_type)
if assignable:
if src_type == dst_type or dst_type.same_as(src_type):
pass # score 0
elif func_type.is_strict_signature:
break # exact match requested but not found
elif is_promotion(src_type, dst_type):
score[2] += 1
elif ((src_type.is_int and dst_type.is_int) or
(src_type.is_float and dst_type.is_float)):
score[2] += abs(dst_type.rank + (not dst_type.signed) -
(src_type.rank + (not src_type.signed))) + 1
elif dst_type.is_ptr and src_type.is_ptr:
if dst_type.base_type == c_void_type:
score[4] += 1
elif src_type.base_type.is_cpp_class and src_type.base_type.is_subclass(dst_type.base_type):
score[6] += src_type.base_type.subclass_dist(dst_type.base_type)
else:
score[5] += 1
elif not src_type.is_pyobject:
score[1] += 1
else:
score[0] += 1
else:
error_mesg = "Invalid conversion from '%s' to '%s'" % (src_type, dst_type)
bad_types.append((func, error_mesg))
break
else:
possibilities.append((score, index, func)) # so we can sort it
if possibilities:
possibilities.sort()
if len(possibilities) > 1:
score1 = possibilities[0][0]
score2 = possibilities[1][0]
if score1 == score2:
if pos is not None:
error(pos, "ambiguous overloaded method")
return None
function = possibilities[0][-1]
if function in needed_coercions and env:
arg_i, coerce_to_type = needed_coercions[function]
args[arg_i] = args[arg_i].coerce_to(coerce_to_type, env)
return function
if pos is not None:
if len(bad_types) == 1:
error(pos, bad_types[0][1])
else:
error(pos, "no suitable method found")
return None
def merge_template_deductions(a, b):
if a is None or b is None:
return None
all = a
for param, value in b.items():
if param in all:
if a[param] != b[param]:
return None
else:
all[param] = value
return all
def widest_numeric_type(type1, type2):
"""Given two numeric types, return the narrowest type encompassing both of them.
"""
if type1.is_reference:
type1 = type1.ref_base_type
if type2.is_reference:
type2 = type2.ref_base_type
if type1.is_cv_qualified:
type1 = type1.cv_base_type
if type2.is_cv_qualified:
type2 = type2.cv_base_type
if type1 == type2:
widest_type = type1
elif type1.is_complex or type2.is_complex:
def real_type(ntype):
if ntype.is_complex:
return ntype.real_type
return ntype
widest_type = CComplexType(
widest_numeric_type(
real_type(type1),
real_type(type2)))
if type1 is soft_complex_type or type2 is soft_complex_type:
type1_is_other_complex = type1 is not soft_complex_type and type1.is_complex
type2_is_other_complex = type2 is not soft_complex_type and type2.is_complex
if (not type1_is_other_complex and not type2_is_other_complex and
widest_type.real_type == soft_complex_type.real_type):
# ensure we can do an actual "is" comparison
# (this possibly goes slightly wrong when mixing long double and soft complex)
widest_type = soft_complex_type
elif type1.is_enum and type2.is_enum:
widest_type = c_int_type
elif type1.rank < type2.rank:
widest_type = type2
elif type1.rank > type2.rank:
widest_type = type1
elif type1.signed < type2.signed:
widest_type = type1
elif type1.signed > type2.signed:
widest_type = type2
elif type1.is_typedef > type2.is_typedef:
widest_type = type1
else:
widest_type = type2
return widest_type
def numeric_type_fits(small_type, large_type):
return widest_numeric_type(small_type, large_type) == large_type
def independent_spanning_type(type1, type2):
# Return a type assignable independently from both type1 and
# type2, but do not require any interoperability between the two.
# For example, in "True * 2", it is safe to assume an integer
# result type (so spanning_type() will do the right thing),
# whereas "x = True or 2" must evaluate to a type that can hold
# both a boolean value and an integer, so this function works
# better.
if type1.is_reference ^ type2.is_reference:
if type1.is_reference:
type1 = type1.ref_base_type
else:
type2 = type2.ref_base_type
resolved_type1 = type1.resolve()
resolved_type2 = type2.resolve()
if resolved_type1 == resolved_type2:
return type1
elif ((resolved_type1 is c_bint_type or resolved_type2 is c_bint_type)
and (type1.is_numeric and type2.is_numeric)):
# special case: if one of the results is a bint and the other
# is another C integer, we must prevent returning a numeric
# type so that we do not lose the ability to coerce to a
# Python bool if we have to.
return py_object_type
span_type = _spanning_type(type1, type2)
if span_type is None:
return error_type
return span_type
def spanning_type(type1, type2):
# Return a type assignable from both type1 and type2, or
# py_object_type if no better type is found. Assumes that the
# code that calls this will try a coercion afterwards, which will
# fail if the types cannot actually coerce to a py_object_type.
if type1 == type2:
return type1
elif type1 is py_object_type or type2 is py_object_type:
return py_object_type
elif type1 is c_py_unicode_type or type2 is c_py_unicode_type:
# Py_UNICODE behaves more like a string than an int
return py_object_type
span_type = _spanning_type(type1, type2)
if span_type is None:
return py_object_type
return span_type
def _spanning_type(type1, type2):
if type1.is_numeric and type2.is_numeric:
return widest_numeric_type(type1, type2)
elif type1.is_builtin_type and type1.name == 'float' and type2.is_numeric:
return widest_numeric_type(c_double_type, type2)
elif type2.is_builtin_type and type2.name == 'float' and type1.is_numeric:
return widest_numeric_type(type1, c_double_type)
elif type1.is_extension_type and type2.is_extension_type:
return widest_extension_type(type1, type2)
elif type1.is_pyobject or type2.is_pyobject:
return py_object_type
elif type1.assignable_from(type2):
if type1.is_extension_type and type1.typeobj_is_imported():
# external types are unsafe, so we use PyObject instead
return py_object_type
return type1
elif type2.assignable_from(type1):
if type2.is_extension_type and type2.typeobj_is_imported():
# external types are unsafe, so we use PyObject instead
return py_object_type
return type2
elif type1.is_ptr and type2.is_ptr:
if type1.base_type.is_cpp_class and type2.base_type.is_cpp_class:
common_base = widest_cpp_type(type1.base_type, type2.base_type)
if common_base:
return CPtrType(common_base)
# incompatible pointers, void* will do as a result
return c_void_ptr_type
else:
return None
def widest_extension_type(type1, type2):
if type1.typeobj_is_imported() or type2.typeobj_is_imported():
return py_object_type
while True:
if type1.subtype_of(type2):
return type2
elif type2.subtype_of(type1):
return type1
type1, type2 = type1.base_type, type2.base_type
if type1 is None or type2 is None:
return py_object_type
def widest_cpp_type(type1, type2):
@cached_function
def bases(type):
all = set()
for base in type.base_classes:
all.add(base)
all.update(bases(base))
return all
common_bases = bases(type1).intersection(bases(type2))
common_bases_bases = reduce(set.union, [bases(b) for b in common_bases], set())
candidates = [b for b in common_bases if b not in common_bases_bases]
if len(candidates) == 1:
return candidates[0]
else:
# Fall back to void* for now.
return None
def simple_c_type(signed, longness, name):
# Find type descriptor for simple type given name and modifiers.
# Returns None if arguments don't make sense.
return modifiers_and_name_to_type.get((signed, longness, name))
def parse_basic_type(name):
base = None
if name.startswith('p_'):
base = parse_basic_type(name[2:])
elif name.startswith('p'):
base = parse_basic_type(name[1:])
elif name.endswith('*'):
base = parse_basic_type(name[:-1])
if base:
return CPtrType(base)
#
basic_type = simple_c_type(1, 0, name)
if basic_type:
return basic_type
#
signed = 1
longness = 0
if name == 'Py_UNICODE':
signed = 0
elif name == 'Py_UCS4':
signed = 0
elif name == 'Py_hash_t':
signed = 2
elif name == 'Py_ssize_t':
signed = 2
elif name == 'ssize_t':
signed = 2
elif name == 'size_t':
signed = 0
elif name == 'ptrdiff_t':
signed = 2
else:
if name.startswith('u'):
name = name[1:]
signed = 0
elif (name.startswith('s') and
not name.startswith('short')):
name = name[1:]
signed = 2
longness = 0
while name.startswith('short'):
name = name.replace('short', '', 1).strip()
longness -= 1
while name.startswith('long'):
name = name.replace('long', '', 1).strip()
longness += 1
if longness != 0 and not name:
name = 'int'
return simple_c_type(signed, longness, name)
def _construct_type_from_base(cls, base_type, *args):
if base_type is error_type:
return error_type
return cls(base_type, *args)
def c_array_type(base_type, size):
# Construct a C array type.
return _construct_type_from_base(CArrayType, base_type, size)
def c_ptr_type(base_type):
# Construct a C pointer type.
if base_type.is_reference:
base_type = base_type.ref_base_type
return _construct_type_from_base(CPtrType, base_type)
def c_ref_type(base_type):
# Construct a C reference type
return _construct_type_from_base(CReferenceType, base_type)
def cpp_rvalue_ref_type(base_type):
# Construct a C++ rvalue reference type
return _construct_type_from_base(CppRvalueReferenceType, base_type)
def c_const_type(base_type):
# Construct a C const type.
return _construct_type_from_base(CConstType, base_type)
def c_const_or_volatile_type(base_type, is_const, is_volatile):
# Construct a C const/volatile type.
return _construct_type_from_base(CConstOrVolatileType, base_type, is_const, is_volatile)
def same_type(type1, type2):
return type1.same_as(type2)
def assignable_from(type1, type2):
return type1.assignable_from(type2)
def typecast(to_type, from_type, expr_code):
# Return expr_code cast to a C type which can be
# assigned to to_type, assuming its existing C type
# is from_type.
if (to_type is from_type or
(not to_type.is_pyobject and assignable_from(to_type, from_type))):
return expr_code
elif (to_type is py_object_type and from_type and
from_type.is_builtin_type and from_type.name != 'type'):
# no cast needed, builtins are PyObject* already
return expr_code
else:
#print "typecast: to", to_type, "from", from_type ###
return to_type.cast_code(expr_code)
def type_list_identifier(types):
return cap_length('__and_'.join(type_identifier(type) for type in types))
_special_type_characters = {
'__': '__dunder',
'const ': '__const_',
' ': '__space_',
'*': '__ptr',
'&': '__ref',
'&&': '__fwref',
'[': '__lArr',
']': '__rArr',
'<': '__lAng',
'>': '__rAng',
'(': '__lParen',
')': '__rParen',
',': '__comma_',
'...': '__EL',
'::': '__in_',
':': '__D',
}
_escape_special_type_characters = partial(re.compile(
# join substrings in reverse order to put longer matches first, e.g. "::" before ":"
" ?(%s) ?" % "|".join(re.escape(s) for s in sorted(_special_type_characters, reverse=True))
).sub, lambda match: _special_type_characters[match.group(1)])
def type_identifier(type, pyrex=False):
scope = None
decl = type.empty_declaration_code(pyrex=pyrex)
entry = getattr(type, "entry", None)
if entry and entry.scope:
scope = entry.scope
return type_identifier_from_declaration(decl, scope=scope)
_type_identifier_cache = {}
def type_identifier_from_declaration(decl, scope = None):
key = (decl, scope)
safe = _type_identifier_cache.get(key)
if safe is None:
safe = decl
if scope:
safe = scope.mangle(prefix="", name=safe)
safe = re.sub(' +', ' ', safe)
safe = re.sub(' ?([^a-zA-Z0-9_]) ?', r'\1', safe)
safe = _escape_special_type_characters(safe)
safe = cap_length(re.sub('[^a-zA-Z0-9_]', lambda x: '__%X' % ord(x.group(0)), safe))
_type_identifier_cache[key] = safe
return safe
def cap_length(s, max_len=63):
if len(s) <= max_len:
return s
hash_prefix = hashlib.sha256(s.encode('ascii')).hexdigest()[:6]
return '%s__%s__etc' % (hash_prefix, s[:max_len-17])
def write_noexcept_performance_hint(pos, env,
function_name=None, void_return=False, is_call=False,
is_from_pxd=False):
if function_name:
# we need it escaped everywhere we use it
function_name = "'%s'" % function_name
if is_call:
on_what = "after calling %s " % (function_name or 'function')
elif function_name:
on_what = "on %s " % function_name
else:
on_what =''
msg = (
"Exception check %swill always require the GIL to be acquired."
) % on_what
the_function = function_name if function_name else "the function"
if is_call and not function_name:
the_function = the_function + " you are calling"
solutions = ["Declare %s as 'noexcept' if you control the definition and "
"you're sure you don't want the function to raise exceptions."
% the_function]
if void_return:
solutions.append(
"Use an 'int' return type on %s to allow an error code to be returned." %
the_function)
if is_from_pxd and not void_return:
solutions.append(
"Declare any exception value explicitly for functions in pxd files.")
if len(solutions) == 1:
msg = "%s %s" % (msg, solutions[0])
else:
solutions = ["\t%s. %s" % (i+1, s) for i, s in enumerate(solutions)]
msg = "%s\nPossible solutions:\n%s" % (msg, "\n".join(solutions))
performance_hint(pos, msg, env)
def remove_cv_ref(tp, remove_fakeref=False):
# named by analogy with c++ std::remove_cv_ref
last_tp = None
# The while-loop is probably unnecessary, but I'm not confident
# of the order or how careful we are prevent nesting.
while tp != last_tp:
last_tp = tp
if tp.is_cv_qualified:
tp = tp.cv_base_type
if tp.is_reference and (not tp.is_fake_reference or remove_fakeref):
tp = tp.ref_base_type
return tp