"""
cppgen.py - AST pass to that prints C++ code
"""
import itertools
import json # for "C escaping"
from typing import overload, Union, Optional, Dict
import mypy
from mypy.visitor import ExpressionVisitor, StatementVisitor
from mypy.types import (Type, AnyType, NoneTyp, TupleType, Instance, NoneType,
Overloaded, CallableType, UnionType, UninhabitedType,
PartialType, TypeAliasType)
from mypy.nodes import (Expression, Statement, NameExpr, IndexExpr, MemberExpr,
TupleExpr, ExpressionStmt, IfStmt, StrExpr, SliceExpr,
FuncDef, UnaryExpr, OpExpr, ComparisonExpr, CallExpr,
IntExpr, ListExpr, DictExpr, ListComprehension)
from mycpp import format_strings
from mycpp.crash import catch_errors
from mycpp.util import log, join_name, split_py_name
from mycpp import util
from typing import Tuple, List
T = None
NAME_CONFLICTS = ('stdin', 'stdout', 'stderr')
class UnsupportedException(Exception):
pass
def _IsContextManager(class_name):
return class_name[-1].startswith('ctx_')
def _SkipAssignment(var_name):
"""
Skip at the top level:
_ = log
unused1 = log
Always skip:
x, _ = mytuple # no second var
"""
return var_name == '_' or var_name.startswith('unused')
def _GetCTypeForCast(type_expr):
""" MyPy cast() """
if isinstance(type_expr, MemberExpr):
subtype_name = '%s::%s' % (type_expr.expr.name, type_expr.name)
elif isinstance(type_expr, IndexExpr):
# List[word_t] would be a problem.
# But worked around it in osh/word_parse.py
#subtype_name = 'List<word_t>'
raise AssertionError()
elif isinstance(type_expr, StrExpr):
parts = type_expr.value.split('.')
subtype_name = '::'.join(parts)
else:
subtype_name = type_expr.name
# Hack for now
if subtype_name != 'int' and subtype_name != 'mops::BigInt':
subtype_name += '*'
return subtype_name
def _GetCastKind(module_path, cast_to_type):
"""Translate MyPy cast to C++ cast.
Prefer static_cast, but sometimes we need reinterpret_cast.
"""
cast_kind = 'static_cast'
# Hack for Id.Expr_CastedDummy in expr_to_ast.py
if 'expr_to_ast.py' in module_path:
for name in (
'ShArrayLiteral',
'CommandSub',
'BracedVarSub',
'DoubleQuoted',
'SingleQuoted',
# Another kind of hack, not because of CastDummy
'y_lhs_t',
):
if name in cast_to_type:
cast_kind = 'reinterpret_cast'
break
# The other side of Id.Expr_CastedDummy
if 'expr_parse.py' in module_path:
for name in ('Token', ):
if name in cast_to_type:
cast_kind = 'reinterpret_cast'
break
if 'process.py' in module_path and 'mylib::Writer' in cast_to_type:
cast_kind = 'reinterpret_cast'
return cast_kind
def _GetContainsFunc(t):
""" x in y """
contains_func = None
if isinstance(t, Instance):
type_name = t.type.fullname
if type_name == 'builtins.list':
contains_func = 'list_contains'
elif type_name == 'builtins.str':
contains_func = 'str_contains'
elif type_name == 'builtins.dict':
contains_func = 'dict_contains'
elif isinstance(t, UnionType):
# Special case for Optional[T] == Union[T, None]
if len(t.items) != 2:
raise NotImplementedError('Expected Optional, got %s' % t)
if not isinstance(t.items[1], NoneTyp):
raise NotImplementedError('Expected Optional, got %s' % t)
contains_func = _GetContainsFunc(t.items[0])
return contains_func # None checked later
def IsStr(t):
"""Helper to check if a type is a string."""
return isinstance(t, Instance) and t.type.fullname == 'builtins.str'
def _EqualsFunc(left_type):
if IsStr(left_type):
return 'str_equals'
if (isinstance(left_type, UnionType) and len(left_type.items) == 2 and
IsStr(left_type.items[0]) and
isinstance(left_type.items[1], NoneTyp)):
return 'maybe_str_equals'
return None
_EXPLICIT = ('builtins.str', 'builtins.list', 'builtins.dict')
def _CheckCondition(node, types):
"""
Ban
if (mystr)
if (mylist)
if (mydict)
They mean non-empty in Python.
"""
#log('NODE %s', node)
if isinstance(node, UnaryExpr) and node.op == 'not':
return _CheckCondition(node.expr, types)
if isinstance(node, OpExpr):
#log('OpExpr node %s %s', node, dir(node))
# if x > 0 and not mylist, etc.
return _CheckCondition(node.left, types) and _CheckCondition(
node.right, types)
t = types[node]
if isinstance(t, Instance):
type_name = t.type.fullname
if type_name in _EXPLICIT:
return False
elif isinstance(t, UnionType):
if len(t.items) == 2 and isinstance(t.items[1], NoneTyp):
t2 = t.items[0]
if t2.type.fullname in _EXPLICIT:
return False
return True
def CTypeIsManaged(c_type):
# type: (str) -> bool
"""For rooting and field masks."""
assert c_type != 'void'
if util.SMALL_STR:
if c_type == 'Str':
return True
# int, double, bool, scope_t enums, etc. are not managed
return c_type.endswith('*')
def GetCType(t, param=False, local=False):
"""Recursively translate MyPy type to C++ type."""
is_pointer = False
if isinstance(t, NoneTyp): # e.g. a function that doesn't return anything
return 'void'
elif isinstance(t, AnyType):
# Note: this usually results in another compile-time error. We should get
# rid of the 'Any' types.
c_type = 'void'
is_pointer = True
elif isinstance(t, PartialType):
# Note: bin/oil.py has some of these? Not sure why.
c_type = 'void'
is_pointer = True
# TODO: It seems better not to check for string equality, but that's what
# mypyc/genops.py does?
elif isinstance(t, Instance):
type_name = t.type.fullname
#log('** TYPE NAME %s', type_name)
if type_name == 'builtins.int':
c_type = 'int'
elif type_name == 'builtins.float':
c_type = 'double'
elif type_name == 'builtins.bool':
c_type = 'bool'
elif type_name == 'builtins.str':
if util.SMALL_STR:
c_type = 'Str'
is_pointer = False
else:
c_type = 'BigStr'
is_pointer = True
elif type_name == 'builtins.list':
assert len(t.args) == 1, t.args
type_param = t.args[0]
inner_c_type = GetCType(type_param)
c_type = 'List<%s>' % inner_c_type
is_pointer = True
elif type_name == 'builtins.dict':
params = []
for type_param in t.args:
params.append(GetCType(type_param))
c_type = 'Dict<%s>' % ', '.join(params)
is_pointer = True
elif 'BigInt' in type_name:
# also spelled mycpp.mylib.BigInt
c_type = 'mops::BigInt'
# Not a pointer!
elif type_name == 'typing.IO':
c_type = 'mylib::File'
is_pointer = True
elif type_name == 'typing.Iterator':
assert len(t.args) == 1, t.args
type_param = t.args[0]
inner_c_type = GetCType(type_param)
c_type = 'ListIter<%s>' % inner_c_type
else:
# note: fullname => 'parse.Lexer'; name => 'Lexer'
base_class_names = [b.type.fullname for b in t.type.bases]
#log('** base_class_names %s', base_class_names)
# Check base class for pybase.SimpleObj so we can output
# expr_asdl::tok_t instead of expr_asdl::tok_t*. That is a enum, while
# expr_t is a "regular base class".
# NOTE: Could we avoid the typedef? If it's SimpleObj, just generate
# tok_e instead?
if 'asdl.pybase.SimpleObj' not in base_class_names:
is_pointer = True
parts = t.type.fullname.split('.')
c_type = '%s::%s' % (parts[-2], parts[-1])
elif isinstance(t, UninhabitedType):
# UninhabitedType has a NoReturn flag
c_type = 'void'
elif isinstance(t, TupleType):
inner_c_types = []
for inner_type in t.items:
c_type = GetCType(inner_type)
if c_type == 'void':
# Why does MyPy give us 'None' instead of type declared with type: ?
# log('**** items %s', t.items)
pass
inner_c_types.append(c_type)
c_type = 'Tuple%d<%s>' % (len(t.items), ', '.join(inner_c_types))
is_pointer = True
elif isinstance(t, UnionType):
# Special case for Optional[IOError_OSError]
# == Union[IOError, OSError, None]
num_items = len(t.items)
if num_items == 3:
t0 = t.items[0]
t1 = t.items[1]
t2 = t.items[2]
t0_name = t0.type.fullname
t1_name = t1.type.fullname
if t0_name != 'builtins.IOError':
raise NotImplementedError(
'Expected Union[IOError, OSError, None]: t0 = %s' %
t0_name)
if t1_name != 'builtins.OSError':
raise NotImplementedError(
'Expected Union[IOError, OSError, None]: t1 = %s' %
t1_name)
if not isinstance(t2, NoneTyp):
raise NotImplementedError(
'Expected Union[IOError, OSError, None]')
c_type = 'IOError_OSError'
is_pointer = True
elif num_items == 2:
t0 = t.items[0]
t1 = t.items[1]
c_type = None
if isinstance(t1, NoneTyp): # Optional[T0]
c_type = GetCType(t.items[0])
else:
# Detect type alias defined in core/error.py
# IOError_OSError = Union[IOError, OSError]
t0_name = t0.type.fullname
t1_name = t1.type.fullname
if (t0_name == 'builtins.IOError' and
t1_name == 'builtins.OSError'):
c_type = 'IOError_OSError'
is_pointer = True
if c_type is None:
raise NotImplementedError('Unexpected Union type %s' % t)
else:
raise NotImplementedError(
'Expected 2 or 3 items in Union, got %s' % num_items)
elif isinstance(t, CallableType):
# Function types are expanded
# Callable[[Parser, Token, int], arith_expr_t]
# -> arith_expr_t* (*f)(Parser*, Token*, int) nud;
ret_type = GetCType(t.ret_type)
arg_types = [GetCType(typ) for typ in t.arg_types]
c_type = '%s (*f)(%s)' % (ret_type, ', '.join(arg_types))
elif isinstance(t, TypeAliasType):
if 0:
log('***')
log('%s', t)
log('%s', dir(t))
log('%s', t.alias)
log('%s', dir(t.alias))
log('%s', t.alias.target)
log('***')
return GetCType(t.alias.target)
else:
raise NotImplementedError('MyPy type: %s %s' % (type(t), t))
if is_pointer:
if param or local:
c_type = 'Local<%s>' % c_type
else:
c_type += '*'
return c_type
def GetCReturnType(t) -> Tuple[str, bool, Optional[str]]:
"""
Returns a C string, whether the tuple-by-value optimization was applied, and
the C type of an extra output param if the function is a generator.
"""
c_ret_type = GetCType(t)
# Optimization: Return tupels BY VALUE
if isinstance(t, TupleType):
assert c_ret_type.endswith('*')
return c_ret_type[:-1], True, None
elif c_ret_type.startswith('ListIter<'):
assert len(t.args) == 1, t.args
inner_c_type = GetCType(t.args[0])
return 'void', False, 'List<%s>*' % inner_c_type
else:
return c_ret_type, False, None
def PythonStringLiteral(s: str) -> str:
"""
Returns a properly quoted string.
"""
# MyPy does bad escaping. Decode and push through json to get something
# workable in C++.
return json.dumps(format_strings.DecodeMyPyString(s))
class Generate(ExpressionVisitor[T], StatementVisitor[None]):
def __init__(self,
types: Dict[Expression, Type],
const_lookup,
f,
virtual=None,
local_vars=None,
ctx_member_vars=None,
decl=False,
forward_decl=False,
stack_roots_warn=None):
self.types = types
self.const_lookup = const_lookup
self.f = f
self.virtual = virtual
# We collect local_vars and ctx-member_vars in the DECL phase, and
# write them in the IMPL phase.
self.local_vars = local_vars # Dict[FuncDef node, list of type, var]
self.ctx_member_vars = ctx_member_vars # for rooting
self.decl = decl
self.forward_decl = forward_decl
self.stack_roots_warn = stack_roots_warn
self.unique_id = 0
self.indent = 0
self.local_var_list = [] # Collected at assignment
self.prepend_to_block = None # For writing vars after {
# Temporary lists to use as output params for generators
self.yield_accumulators = {
} # type: Dict[Union[Statement, FuncDef], Tuple[str, str]]
self.current_func_node = None
self.current_stmt_node = None
# DECL pass: self.current_member_vars holds the vars for the current
# class. It's cleared when we start looking at a class. Then we visit
# all the methods, and accumulate the types of everything that looks
# like self.foo = 1. Then we write C++ class member declarations at
# the end of the class.
self.current_member_vars: Dict[str, Type] = {}
self.current_class_name = None # for prototypes
self.current_method_name = None
self.imported_names = set() # MemberExpr -> module::Foo() or self->foo
# HACK for conditional import inside mylib.PYTHON
# in core/shell.py
self.imported_names.add('help_meta')
# So we can report multiple at once
# module path, line number, message
self.errors_keep_going: List[Tuple[str, int, str]] = []
self.writing_default_arg = False
def log(self, msg, *args):
ind_str = self.indent * ' '
log(ind_str + msg, *args)
def always_write(self, msg, *args):
"""Write unconditionally - forward decl, decl, def """
if args:
msg = msg % args
self.f.write(msg)
def always_write_ind(self, msg, *args):
ind_str = self.indent * ' '
self.always_write(ind_str + msg, *args)
def def_write(self, msg, *args):
"""Write only in definitions."""
if self.forward_decl:
return
if self.decl and not self.writing_default_arg:
return
if args:
msg = msg % args
self.f.write(msg)
def def_write_ind(self, msg, *args):
ind_str = self.indent * ' '
self.def_write(ind_str + msg, *args)
def decl_write(self, msg, *args):
"""Write only in the decl stage (not forward declarations)"""
if not self.decl:
return
if args:
msg = msg % args
self.f.write(msg)
def decl_write_ind(self, msg, *args):
ind_str = self.indent * ' '
self.decl_write(ind_str + msg, *args)
#
# COPIED from IRBuilder
#
@overload
def accept(self, node: Expression) -> T:
...
@overload
def accept(self, node: Statement) -> None:
...
def accept(self, node: Union[Statement, Expression]) -> Optional[T]:
with catch_errors(self.module_path, node.line):
if isinstance(node, Expression):
try:
res = node.accept(self)
#res = self.coerce(res, self.node_type(node), node.line)
# If we hit an error during compilation, we want to
# keep trying, so we can produce more error
# messages. Generate a temp of the right type to keep
# from causing more downstream trouble.
except UnsupportedException:
res = self.alloc_temp(self.node_type(node))
return res
else:
try:
node.accept(self)
except UnsupportedException:
pass
return None
def report_error(self, node: Union[Statement, Expression], msg: str):
err = (self.module_path, node.line, msg)
self.errors_keep_going.append(err)
def not_translated(self, node: Union[Statement, Expression], name: str):
self.report_error(node, '%s not translated' % name)
def not_python2(self, node: Union[Statement, Expression], name: str):
self.report_error(node, "%s: shouldn't get here in Python 2" % name)
# Not in superclasses:
def visit_mypy_file(self, o: 'mypy.nodes.MypyFile') -> T:
if util.ShouldSkipPyFile(o):
return
#self.log('')
#self.log('mypyfile %s', o.fullname)
mod_parts = o.fullname.split('.')
if self.forward_decl:
comment = 'forward declare'
elif self.decl:
comment = 'declare'
else:
comment = 'define'
self.always_write_ind('namespace %s { // %s\n', mod_parts[-1],
comment)
self.always_write('\n')
self.module_path = o.path
if self.forward_decl:
self.indent += 1
#self.log('defs %s', o.defs)
for node in o.defs:
# skip module docstring
if (isinstance(node, ExpressionStmt) and
isinstance(node.expr, StrExpr)):
continue
self.accept(node)
if self.forward_decl:
self.indent -= 1
self.always_write('\n')
self.always_write_ind('} // %s namespace %s\n', comment,
mod_parts[-1])
self.always_write('\n')
for path, line_num, msg in self.errors_keep_going:
self.log('%s:%s %s', path, line_num, msg)
# NOTE: Copied ExpressionVisitor and StatementVisitor nodes below!
# LITERALS
def visit_int_expr(self, o: 'mypy.nodes.IntExpr') -> T:
self.def_write(str(o.value))
def visit_float_expr(self, o: 'mypy.nodes.FloatExpr') -> T:
# e.g. for arg.t > 0.0
self.def_write(str(o.value))
def visit_str_expr(self, o: 'mypy.nodes.StrExpr') -> T:
self.def_write(self.const_lookup[o])
# UNHANDLED
def visit_bytes_expr(self, o: 'mypy.nodes.BytesExpr') -> T:
self.not_python2(o, 'bytes expr')
def visit_unicode_expr(self, o: 'mypy.nodes.UnicodeExpr') -> T:
self.not_translated(o, 'unicode expr')
def visit_complex_expr(self, o: 'mypy.nodes.ComplexExpr') -> T:
self.not_translated(o, 'complex expr')
def visit_ellipsis(self, o: 'mypy.nodes.EllipsisExpr') -> T:
# is this in .pyi files only?
self.not_translated(o, 'ellipsis')
# Expressions
def visit_star_expr(self, o: 'mypy.nodes.StarExpr') -> T:
# mycpp/examples/invalid_python.py doesn't hit this?
self.not_translated(o, 'star expr')
def visit_name_expr(self, o: 'mypy.nodes.NameExpr') -> T:
if o.name == 'None':
self.def_write('nullptr')
return
if o.name == 'True':
self.def_write('true')
return
if o.name == 'False':
self.def_write('false')
return
if o.name == 'self':
self.def_write('this')
return
if o.name in NAME_CONFLICTS:
self.report_error(
o,
"The name %r conflicts with C macros on some platforms; choose a different name"
% o.name)
return
self.def_write(o.name)
def visit_member_expr(self, o: 'mypy.nodes.MemberExpr') -> T:
if o.expr:
# Why do we not get some of the types? e.g. hnode.Record in asdl/runtime
# But this might suffice for the "Str_v" and "value_v" refactoring.
# We want to rewrite w.parts not to w->parts, but to w.parts() (method call)
is_small_str = False
if util.SMALL_STR:
lhs_type = self.types.get(o.expr)
if IsStr(lhs_type):
is_small_str = True
else:
#self.log('NOT a string %s %s', o.expr, o.name)
pass
"""
if lhs_type is not None and isinstance(lhs_type, Instance):
self.log('lhs_type %s expr %s name %s',
lhs_type.type.fullname, o.expr, o.name)
"""
is_asdl = o.name == 'CreateNull' # hack for MyType.CreateNull(alloc_lists=True)
is_module = (isinstance(o.expr, NameExpr) and
o.expr.name in self.imported_names)
# This is an approximate hack that assumes that locals don't shadow
# imported names. Might be a problem with names like 'word'?
if is_small_str:
op = '.'
elif is_asdl or is_module:
op = '::'
else:
op = '->' # Everything is a pointer
self.accept(o.expr)
self.def_write(op)
if o.name == 'errno':
# e->errno -> e->errno_ to avoid conflict with C macro
self.def_write('errno_')
elif o.name in NAME_CONFLICTS:
self.report_error(
o,
"The name %r conflicts with C macros on some platforms; choose a different name"
% o.name)
else:
self.def_write('%s', o.name)
def visit_yield_from_expr(self, o: 'mypy.nodes.YieldFromExpr') -> T:
self.not_python2(o, 'yield from')
def visit_yield_expr(self, o: 'mypy.nodes.YieldExpr') -> T:
assert self.current_func_node in self.yield_accumulators
self.def_write_ind('%s->append(',
self.yield_accumulators[self.current_func_node][0])
self.accept(o.expr)
self.def_write(');\n')
def _WriteArgList(self, o):
self.def_write('(')
for i, arg in enumerate(o.args):
if i != 0:
self.def_write(', ')
self.accept(arg)
# Will be set if we're:
# a) accumulating the output of an iterator
# b) constructing an iterator with the result of (a)
if self.current_stmt_node in self.yield_accumulators:
if len(o.args) > 0:
self.def_write(', ')
arg_name, _ = self.yield_accumulators[self.current_stmt_node]
self.def_write('&%s', arg_name)
self.def_write(')')
def _IsInstantiation(self, o):
callee_name = o.callee.name
callee_type = self.types[o.callee]
# e.g. int() takes str, float, etc. It doesn't matter for translation.
if isinstance(callee_type, Overloaded):
if 0:
for item in callee_type.items():
self.log('item: %s', item)
if isinstance(callee_type, CallableType):
# If the function name is the same as the return type, then add
# 'Alloc<>'. f = Foo() => f = Alloc<Foo>().
ret_type = callee_type.ret_type
# str(i) doesn't need new. For now it's a free function.
# TODO: rename int_to_str? or BigStr::from_int()?
if (callee_name not in ('str', 'bool', 'float') and
'BigInt' not in callee_name and
isinstance(ret_type, Instance)):
ret_type_name = ret_type.type.name
# HACK: Const is the callee; expr.Const is the return type
if (ret_type_name == callee_name or
ret_type_name.endswith('__' + callee_name)):
return True
return False
def _ProbeExpr(self, o):
assert len(o.args) >= 2 and len(o.args) < 13, o.args
assert isinstance(o.args[0], mypy.nodes.StrExpr), o.args[0]
assert isinstance(o.args[1], mypy.nodes.StrExpr), o.args[1]
arity = len(o.args) - 2
macro = 'DTRACE_PROBE'
if arity > 0:
macro = 'DTRACE_PROBE%d' % arity
self.def_write('%s(%s, %s', macro, o.args[0].value, o.args[1].value)
for arg in o.args[2:]:
arg_type = self.types[arg]
self.def_write(', ')
if (isinstance(arg_type, Instance) and
arg_type.type.fullname == 'builtins.str'):
self.def_write('%s->data()' % arg.name)
else:
self.accept(arg)
self.def_write(')')
def _LogExpr(self, o):
args = o.args
if len(args) == 1: # log(CONST)
self.def_write('mylib::print_stderr(')
self.accept(args[0])
self.def_write(')')
return
quoted_fmt = PythonStringLiteral(args[0].value)
self.def_write('mylib::print_stderr(StrFormat(%s, ' % quoted_fmt)
for i, arg in enumerate(args[1:]):
if i != 0:
self.def_write(', ')
self.accept(arg)
self.def_write('))')
def visit_call_expr(self, o: 'mypy.nodes.CallExpr') -> T:
if o.callee.name == 'probe':
self._ProbeExpr(o)
return
if o.callee.name == 'isinstance':
self.report_error(o, 'isinstance() not allowed')
return
# return cast(ShArrayLiteral, tok)
# -> return static_cast<ShArrayLiteral*>(tok)
# TODO: Consolidate this with AssignmentExpr logic.
if o.callee.name == 'cast':
call = o
type_expr = call.args[0]
subtype_name = _GetCTypeForCast(type_expr)
cast_kind = _GetCastKind(self.module_path, subtype_name)
self.def_write('%s<%s>(', cast_kind, subtype_name)
self.accept(call.args[1]) # variable being casted
self.def_write(')')
return
# Translate printf-style varargs:
#
# log('foo %s', x)
# =>
# log(StrFormat('foo %s', x))
if o.callee.name == 'log':
self._LogExpr(o)
return
callee_name = o.callee.name
if isinstance(o.callee, MemberExpr) and callee_name == 'next':
self.accept(o.callee.expr)
self.def_write('.iterNext')
self._WriteArgList(o)
return
if self._IsInstantiation(o):
self.def_write('Alloc<')
self.accept(o.callee)
self.def_write('>')
self._WriteArgList(o)
return
# Namespace.
if callee_name == 'int': # int('foo') in Python conflicts with keyword
self.def_write('to_int')
elif callee_name == 'float':
self.def_write('to_float')
elif callee_name == 'bool':
self.def_write('to_bool')
else:
self.accept(o.callee) # could be f() or obj.method()
self._WriteArgList(o)
# TODO: look at keyword arguments!
#self.log(' arg_kinds %s', o.arg_kinds)
#self.log(' arg_names %s', o.arg_names)
def visit_op_expr(self, o: 'mypy.nodes.OpExpr') -> T:
# a + b when a and b are strings. (Can't use operator overloading
# because they're pointers.)
left_type = self.types[o.left]
right_type = self.types[o.right]
# NOTE: Need GetCType to handle Optional[BigStr*] in ASDL schemas.
# Could tighten it up later.
left_ctype = GetCType(left_type)
right_ctype = GetCType(right_type)
c_op = o.op
if left_ctype == right_ctype == 'int' and c_op == '//':
# integer division // -> /
c_op = '/'
# 'abc' + 'def'
if left_ctype == right_ctype == 'BigStr*' and c_op == '+':
self.def_write('str_concat(')
self.accept(o.left)
self.def_write(', ')
self.accept(o.right)
self.def_write(')')
return
# 'abc' * 3
if left_ctype == 'BigStr*' and right_ctype == 'int' and c_op == '*':
self.def_write('str_repeat(')
self.accept(o.left)
self.def_write(', ')
self.accept(o.right)
self.def_write(')')
return
# [None] * 3 => list_repeat(None, 3)
if left_ctype.startswith(
'List<') and right_ctype == 'int' and c_op == '*':
self.def_write('list_repeat(')
self.accept(o.left.items[0])
self.def_write(', ')
self.accept(o.right)
self.def_write(')')
return
# RHS can be primitive or tuple
if left_ctype == 'BigStr*' and c_op == '%':
self.def_write('StrFormat(')
if isinstance(o.left, StrExpr):
self.def_write(PythonStringLiteral(o.left.value))
else:
self.accept(o.left)
#log('right_type %s', right_type)
if isinstance(right_type, Instance):
fmt_types = [right_type]
elif isinstance(right_type, TupleType):
fmt_types = right_type.items
# Handle Optional[str]
elif (isinstance(right_type, UnionType) and
len(right_type.items) == 2 and
isinstance(right_type.items[1], NoneTyp)):
fmt_types = [right_type.items[0]]
else:
raise AssertionError(right_type)
# In the definition pass, write the call site.
if isinstance(right_type, TupleType):
for i, item in enumerate(o.right.items):
self.def_write(', ')
self.accept(item)
else: # '[%s]' % x
self.def_write(', ')
self.accept(o.right)
self.def_write(')')
return
# These parens are sometimes extra, but sometimes required. Example:
#
# if ((a and (false or true))) { # right
# vs.
# if (a and false or true)) { # wrong
self.def_write('(')
self.accept(o.left)
self.def_write(' %s ', c_op)
self.accept(o.right)
self.def_write(')')
def visit_comparison_expr(self, o: 'mypy.nodes.ComparisonExpr') -> T:
# Make sure it's binary
assert len(o.operators) == 1, o.operators
assert len(o.operands) == 2, o.operands
operator = o.operators[0]
left = o.operands[0]
right = o.operands[1]
# Assume is and is not are for None / nullptr comparison.
if operator == 'is': # foo is None => foo == nullptr
self.accept(o.operands[0])
self.def_write(' == ')
self.accept(o.operands[1])
return
if operator == 'is not': # foo is not None => foo != nullptr
self.accept(o.operands[0])
self.def_write(' != ')
self.accept(o.operands[1])
return
# TODO: Change Optional[T] to T for our purposes?
t0 = self.types[left]
t1 = self.types[right]
# 0: not a special case
# 1: str
# 2: Optional[str] which is Union[str, None]
left_type = 0 # not a special case
right_type = 0 # not a special case
if IsStr(t0):
left_type = 1
elif (isinstance(t0, UnionType) and len(t0.items) == 2 and
IsStr(t0.items[0]) and isinstance(t0.items[1], NoneTyp)):
left_type = 2
if IsStr(t1):
right_type = 1
elif (isinstance(t1, UnionType) and len(t1.items) == 2 and
IsStr(t1.items[0]) and isinstance(t1.items[1], NoneTyp)):
right_type = 2
#self.log('left_type %s right_type %s', left_type, right_type)
if left_type > 0 and right_type > 0 and operator in ('==', '!='):
if operator == '!=':
self.def_write('!(')
# NOTE: This could also be str_equals(left, right)? Does it make a
# difference?
if left_type > 1 or right_type > 1:
self.def_write('maybe_str_equals(')
else:
self.def_write('str_equals(')
self.accept(left)
self.def_write(', ')
self.accept(right)
self.def_write(')')
if operator == '!=':
self.def_write(')')
return
# Note: we could get rid of this altogether and rely on C++ function
# overloading. But somehow I like it more explicit, closer to C (even
# though we use templates).
contains_func = _GetContainsFunc(t1)
if operator == 'in':
if isinstance(right, TupleExpr):
left_type = self.types[left]
equals_func = _EqualsFunc(left_type)
# x in (1, 2, 3) => (x == 1 || x == 2 || x == 3)
self.def_write('(')
for i, item in enumerate(right.items):
if i != 0:
self.def_write(' || ')
if equals_func:
self.def_write('%s(' % equals_func)
self.accept(left)
self.def_write(', ')
self.accept(item)
self.def_write(')')
else:
self.accept(left)
self.def_write(' == ')
self.accept(item)
self.def_write(')')
return
assert contains_func, "RHS of 'in' has type %r" % t1
# x in mylist => list_contains(mylist, x)
self.def_write('%s(', contains_func)
self.accept(right)
self.def_write(', ')
self.accept(left)
self.def_write(')')
return
if operator == 'not in':
if isinstance(right, TupleExpr):
left_type = self.types[left]
equals_func = _EqualsFunc(left_type)
# x not in (1, 2, 3) => (x != 1 && x != 2 && x != 3)
self.def_write('(')
for i, item in enumerate(right.items):
if i != 0:
self.def_write(' && ')
if equals_func:
self.def_write('!%s(' % equals_func)
self.accept(left)
self.def_write(', ')
self.accept(item)
self.def_write(')')
else:
self.accept(left)
self.def_write(' != ')
self.accept(item)
self.def_write(')')
return
assert contains_func, t1
# x not in mylist => !list_contains(mylist, x)
self.def_write('!%s(', contains_func)
self.accept(right)
self.def_write(', ')
self.accept(left)
self.def_write(')')
return
# Default case
self.accept(o.operands[0])
self.def_write(' %s ', o.operators[0])
self.accept(o.operands[1])
def visit_cast_expr(self, o: 'mypy.nodes.CastExpr') -> T:
pass
def visit_reveal_expr(self, o: 'mypy.nodes.RevealExpr') -> T:
pass
def visit_super_expr(self, o: 'mypy.nodes.SuperExpr') -> T:
self.not_translated(o, 'super expr')
def visit_assignment_expr(self, o: 'mypy.nodes.AssignmentExpr') -> T:
self.not_translated(o, 'assign expr')
def visit_unary_expr(self, o: 'mypy.nodes.UnaryExpr') -> T:
# e.g. a[-1] or 'not x'
if o.op == 'not':
op_str = '!'
else:
op_str = o.op
self.def_write(op_str)
self.accept(o.expr)
def _WriteListElements(self, items, sep=', '):
# sep may be 'COMMA' for a macro
self.def_write('{')
for i, item in enumerate(items):
if i != 0:
self.def_write(sep)
self.accept(item)
self.def_write('}')
def visit_list_expr(self, o: 'mypy.nodes.ListExpr') -> T:
list_type = self.types[o]
#self.log('**** list_type = %s', list_type)
c_type = GetCType(list_type)
item_type = list_type.args[0] # int for List[int]
item_c_type = GetCType(item_type)
assert c_type.endswith('*'), c_type
c_type = c_type[:-1] # HACK TO CLEAN UP
if len(o.items) == 0:
self.def_write('Alloc<%s>()' % c_type)
else:
self.def_write('NewList<%s>(std::initializer_list<%s>' %
(item_c_type, item_c_type))
self._WriteListElements(o.items)
self.def_write(')')
def visit_dict_expr(self, o: 'mypy.nodes.DictExpr') -> T:
dict_type = self.types[o]
c_type = GetCType(dict_type)
assert c_type.endswith('*'), c_type
c_type = c_type[:-1] # HACK TO CLEAN UP
key_type, val_type = dict_type.args
key_c_type = GetCType(key_type)
val_c_type = GetCType(val_type)
self.def_write('Alloc<%s>(' % c_type)
#self.def_write('NewDict<%s, %s>(' % (key_c_type, val_c_type))
if o.items:
keys = [k for k, _ in o.items]
values = [v for _, v in o.items]
self.def_write('std::initializer_list<%s>' % key_c_type)
self._WriteListElements(keys)
self.def_write(', ')
self.def_write('std::initializer_list<%s>' % val_c_type)
self._WriteListElements(values)
self.def_write(')')
def visit_tuple_expr(self, o: 'mypy.nodes.TupleExpr') -> T:
tuple_type = self.types[o]
c_type = GetCType(tuple_type)
assert c_type.endswith('*'), c_type
c_type = c_type[:-1] # HACK TO CLEAN UP
self.def_write('(Alloc<%s>(' % c_type)
for i, item in enumerate(o.items):
if i != 0:
self.def_write(', ')
self.accept(item)
self.def_write('))')
def visit_set_expr(self, o: 'mypy.nodes.SetExpr') -> T:
self.not_translated(o, 'set expr')
def visit_index_expr(self, o: 'mypy.nodes.IndexExpr') -> T:
self.accept(o.base)
#base_type = self.types[o.base]
#self.log('*** BASE TYPE %s', base_type)
if isinstance(o.index, SliceExpr):
self.accept(o.index) # method call
else:
# it's hard syntactically to do (*a)[0], so do it this way.
if util.SMALL_STR:
self.def_write('.at(')
else:
self.def_write('->at(')
self.accept(o.index)
self.def_write(')')
def visit_type_application(self, o: 'mypy.nodes.TypeApplication') -> T:
pass
def visit_lambda_expr(self, o: 'mypy.nodes.LambdaExpr') -> T:
pass
def visit_list_comprehension(self, o: 'mypy.nodes.ListComprehension') -> T:
pass
def visit_set_comprehension(self, o: 'mypy.nodes.SetComprehension') -> T:
pass
def visit_dictionary_comprehension(
self, o: 'mypy.nodes.DictionaryComprehension') -> T:
pass
def visit_generator_expr(self, o: 'mypy.nodes.GeneratorExpr') -> T:
pass
def visit_slice_expr(self, o: 'mypy.nodes.SliceExpr') -> T:
self.def_write('->slice(')
if o.begin_index:
self.accept(o.begin_index)
else:
self.def_write('0') # implicit beginning
if o.end_index:
self.def_write(', ')
self.accept(o.end_index)
if o.stride:
if not o.begin_index or not o.end_index:
raise AssertionError(
'Stride only supported with beginning and ending index')
self.def_write(', ')
self.accept(o.stride)
self.def_write(')')
def visit_conditional_expr(self, o: 'mypy.nodes.ConditionalExpr') -> T:
if not _CheckCondition(o.cond, self.types):
self.report_error(
o,
"Use explicit len(obj) or 'obj is not None' for mystr, mylist, mydict"
)
return
# 0 if b else 1 -> b ? 0 : 1
self.accept(o.cond)
self.def_write(' ? ')
self.accept(o.if_expr)
self.def_write(' : ')
self.accept(o.else_expr)
def visit_backquote_expr(self, o: 'mypy.nodes.BackquoteExpr') -> T:
pass
def visit_type_var_expr(self, o: 'mypy.nodes.TypeVarExpr') -> T:
pass
def visit_type_alias_expr(self, o: 'mypy.nodes.TypeAliasExpr') -> T:
pass
def visit_namedtuple_expr(self, o: 'mypy.nodes.NamedTupleExpr') -> T:
pass
def visit_enum_call_expr(self, o: 'mypy.nodes.EnumCallExpr') -> T:
pass
def visit_typeddict_expr(self, o: 'mypy.nodes.TypedDictExpr') -> T:
pass
def visit_newtype_expr(self, o: 'mypy.nodes.NewTypeExpr') -> T:
pass
def visit__promote_expr(self, o: 'mypy.nodes.PromoteExpr') -> T:
pass
def visit_await_expr(self, o: 'mypy.nodes.AwaitExpr') -> T:
pass
def visit_temp_node(self, o: 'mypy.nodes.TempNode') -> T:
pass
def _write_tuple_unpacking(self,
temp_name,
lval_items,
item_types,
is_return=False):
"""Used by assignment and for loops."""
for i, (lval_item, item_type) in enumerate(zip(lval_items,
item_types)):
#self.log('*** %s :: %s', lval_item, item_type)
if isinstance(lval_item, NameExpr):
if _SkipAssignment(lval_item.name):
continue
item_c_type = GetCType(item_type)
# declare it at the top of the function
if self.decl:
self.local_var_list.append((lval_item.name, item_c_type))
self.def_write_ind('%s', lval_item.name)
else:
# Could be MemberExpr like self.foo, self.bar = baz
self.def_write_ind('')
self.accept(lval_item)
# Tuples that are return values aren't pointers
op = '.' if is_return else '->'
self.def_write(' = %s%sat%d();\n', temp_name, op, i) # RHS
def _ListComprehensionImpl(self, o, lval, c_type):
"""
Special case for list comprehensions. Note that the LHS MUST be on the
LHS, so we can append to it.
y = [i+1 for i in x[1:] if i]
=>
y = []
for i in x[1:]:
if i:
y.append(i+1)
(but in C++)
"""
gen = o.rvalue.generator # GeneratorExpr
left_expr = gen.left_expr
index_expr = gen.indices[0]
seq = gen.sequences[0]
cond = gen.condlists[0]
# BUG: can't use this to filter
# results = [x for x in results]
if isinstance(seq, NameExpr) and seq.name == lval.name:
raise AssertionError(
"Can't use var %r in list comprehension because it would "
"be overwritten" % lval.name)
# Write empty container as initialization.
assert c_type.endswith('*'), c_type # Hack
self.def_write('Alloc<%s>();\n' % c_type[:-1])
over_type = self.types[seq]
if over_type.type.fullname == 'builtins.list':
c_type = GetCType(over_type)
assert c_type.endswith('*'), c_type
c_iter_type = c_type.replace('List', 'ListIter',
1)[:-1] # remove *
else:
# Example: assoc == Optional[Dict[str, str]]
c_iter_type = 'TODO_ASSOC'
self.def_write_ind('for (%s it(', c_iter_type)
self.accept(seq)
self.def_write('); !it.Done(); it.Next()) {\n')
item_type = over_type.args[0] # get 'int' from 'List<int>'
if isinstance(item_type, Instance):
self.def_write_ind(' %s ', GetCType(item_type))
# TODO(StackRoots): for ch in 'abc'
self.accept(index_expr)
self.def_write(' = it.Value();\n')
elif isinstance(item_type, TupleType): # for x, y in pairs
c_item_type = GetCType(item_type)
if isinstance(index_expr, TupleExpr):
temp_name = 'tup%d' % self.unique_id
self.unique_id += 1
self.def_write_ind(' %s %s = it.Value();\n', c_item_type,
temp_name)
self.indent += 1
self._write_tuple_unpacking(temp_name, index_expr.items,
item_type.items)
self.indent -= 1
else:
raise AssertionError()
else:
raise AssertionError('Unexpected type %s' % item_type)
if cond:
self.indent += 1
self.def_write_ind('if (')
self.accept(cond[0]) # Just the first one
self.def_write(') {\n')
self.def_write_ind(' %s->append(', lval.name)
self.accept(left_expr)
self.def_write(');\n')
if cond:
self.def_write_ind('}\n')
self.indent -= 1
self.def_write_ind('}\n')
def _AssignNewDictImpl(self, lval):
"""
d = NewDict() # type: Dict[int, int]
-> auto* d = NewDict<int, int>();
- NewDict exists in Python, it makes ordered dictionaries
- We translate it here because we need type inference
I think we could get rid of NewDict in C++, and have it only in
Python.
We used to have the "allocating in a constructor" rooting
problem, but I believe that's gone now.
"""
lval_type = self.types[lval]
# Fix for Dict[str, value]? in ASDL
#self.log('lval type %s', lval_type)
if (isinstance(lval_type, UnionType) and len(lval_type.items) == 2 and
isinstance(lval_type.items[1], NoneTyp)):
lval_type = lval_type.items[0]
c_type = GetCType(lval_type)
if self.decl:
self.local_var_list.append((lval.name, c_type))
assert c_type.endswith('*')
# Hack for declaration vs. definition. TODO: clean this up
prefix = '' if self.current_func_node else 'auto* '
self.def_write_ind('%s%s = Alloc<%s>();\n', prefix, lval.name,
c_type[:-1])
def _AssignCastImpl(self, o, lval):
"""
is_downcast_and_shadow idiom:
src = cast(source__SourcedFile, UP_src)
-> source__SourcedFile* src = static_cast<source__SourcedFile>(UP_src)
"""
assert isinstance(lval, NameExpr)
call = o.rvalue
type_expr = call.args[0]
subtype_name = _GetCTypeForCast(type_expr)
cast_kind = _GetCastKind(self.module_path, subtype_name)
is_downcast_and_shadow = False
to_cast = call.args[1]
if isinstance(to_cast, NameExpr):
if to_cast.name.startswith('UP_'):
is_downcast_and_shadow = True
if is_downcast_and_shadow:
# Declare NEW local variable inside case, which shadows it
self.def_write_ind('%s %s = %s<%s>(', subtype_name, lval.name,
cast_kind, subtype_name)
else:
# Normal variable
if self.decl:
self.local_var_list.append((lval.name, subtype_name))
self.def_write_ind('%s = %s<%s>(', lval.name, cast_kind,
subtype_name)
self.accept(call.args[1]) # variable being casted
self.def_write(');\n')
def _IteratorImpl(self, o, lval, rval_type):
# We're calling a generator. Create a temporary List<T> on the stack
# to accumulate the results in one big batch, then wrap it in
# ListIter<T>.
assert len(rval_type.args) == 1, rval_type.args
c_type = GetCType(rval_type)
type_param = rval_type.args[0]
inner_c_type = GetCType(type_param)
iter_buf = ('_iter_buf_%s' % lval.name, 'List<%s>*' % inner_c_type)
self.def_write_ind('List<%s> %s;\n', inner_c_type, iter_buf[0])
self.current_stmt_node = o
self.yield_accumulators[o] = iter_buf
self.def_write_ind('')
self.accept(o.rvalue)
self.current_stmt_node = None
self.def_write(';\n')
self.def_write_ind('%s %s(&%s);\n', c_type, lval.name, iter_buf[0])
def visit_assignment_stmt(self, o: 'mypy.nodes.AssignmentStmt') -> T:
# Declare constant strings. They have to be at the top level.
if self.decl and self.indent == 0 and len(o.lvalues) == 1:
lval = o.lvalues[0]
c_type = GetCType(self.types[lval])
if not _SkipAssignment(lval.name):
self.always_write('extern %s %s;\n', c_type, lval.name)
# I think there are more than one when you do a = b = 1, which I never
# use.
assert len(o.lvalues) == 1, o.lvalues
lval = o.lvalues[0]
# GLOBAL CONSTANTS - Avoid Alloc<T>, since that can't be done until main().
if self.indent == 0:
assert isinstance(lval, NameExpr), lval
if _SkipAssignment(lval.name):
return
#self.log(' GLOBAL: %s', lval.name)
lval_type = self.types[lval]
# Global
# L = [1, 2] # type: List[int]
if isinstance(o.rvalue, ListExpr):
item_type = lval_type.args[0]
item_c_type = GetCType(item_type)
# Any constant strings will have already been written
# TODO: Assert that every item is a constant?
self.def_write('GLOBAL_LIST(%s, %s, %d, ', lval.name,
item_c_type, len(o.rvalue.items))
self._WriteListElements(o.rvalue.items, sep=' COMMA ')
self.def_write(');\n')
return
# Global
# D = {"foo": "bar"} # type: Dict[str, str]
if isinstance(o.rvalue, DictExpr):
key_type, val_type = lval_type.args
key_c_type = GetCType(key_type)
val_c_type = GetCType(val_type)
dict_expr = o.rvalue
self.def_write('GLOBAL_DICT(%s, %s, %s, %d, ', lval.name,
key_c_type, val_c_type, len(dict_expr.items))
keys = [k for k, _ in dict_expr.items]
values = [v for _, v in dict_expr.items]
self._WriteListElements(keys, sep=' COMMA ')
self.def_write(', ')
self._WriteListElements(values, sep=' COMMA ')
self.def_write(');\n')
return
# We could do GcGlobal<> for ASDL classes, but Oils doesn't use them
if isinstance(o.rvalue, CallExpr):
self.report_error(
o,
"Can't initialize objects at the top level, only BigStr List Dict"
)
return
# myconst = 1 << 3 => myconst = 1 << 3 is currently allowed
#
# Non-top-level
#
if isinstance(o.rvalue, CallExpr):
callee = o.rvalue.callee
if callee.name == 'NewDict':
self._AssignNewDictImpl(lval)
return
if callee.name == 'cast':
self._AssignCastImpl(o, lval)
return
rval_type = self.types[o.rvalue]
if (isinstance(rval_type, Instance) and
rval_type.type.fullname == 'typing.Iterator'):
self._IteratorImpl(o, lval, rval_type)
return
if isinstance(lval, NameExpr):
lval_type = self.types[lval]
#c_type = GetCType(lval_type, local=self.indent != 0)
c_type = GetCType(lval_type)
# for "hoisting" to the top of the function
if self.current_func_node:
self.def_write_ind('%s = ', lval.name)
if self.decl:
self.local_var_list.append((lval.name, c_type))
else:
# globals always get a type -- they're not mutated
self.def_write_ind('%s %s = ', c_type, lval.name)
if isinstance(o.rvalue, ListComprehension):
self._ListComprehensionImpl(o, lval, c_type)
return
self.accept(o.rvalue)
self.def_write(';\n')
return
if isinstance(lval, MemberExpr): # self.x = foo
self.def_write_ind('')
self.accept(lval)
self.def_write(' = ')
self.accept(o.rvalue)
self.def_write(';\n')
if self.current_method_name in ('__init__', 'Reset'):
# Collect statements that look like self.foo = 1
# Only do this in __init__ so that a derived class mutating a field
# from the base class doesn't cause duplicate C++ fields. (C++
# allows two fields of the same name!)
#
# HACK for WordParser: also include Reset(). We could change them
# all up front but I kinda like this.
if (isinstance(lval.expr, NameExpr) and
lval.expr.name == 'self'):
#log(' lval.name %s', lval.name)
lval_type = self.types[lval]
c_type = GetCType(lval_type)
is_managed = CTypeIsManaged(c_type)
self.current_member_vars[lval.name] = (lval_type, c_type,
is_managed)
return
if isinstance(lval, IndexExpr): # a[x] = 1
# d->set(x, 1) for both List and Dict
self.def_write_ind('')
self.accept(lval.base)
self.def_write('->set(')
self.accept(lval.index)
self.def_write(', ')
self.accept(o.rvalue)
self.def_write(');\n')
return
if isinstance(lval, TupleExpr):
# An assignment to an n-tuple turns into n+1 statements. Example:
#
# x, y = mytuple
#
# Tuple2<int, BigStr*> tup1 = mytuple
# int x = tup1->at0()
# BigStr* y = tup1->at1()
rvalue_type = self.types[o.rvalue]
# type alias upgrade for MyPy 0.780
if isinstance(rvalue_type, TypeAliasType):
rvalue_type = rvalue_type.alias.target
c_type = GetCType(rvalue_type)
is_return = (isinstance(o.rvalue, CallExpr) and
o.rvalue.callee.name != "next")
if is_return:
assert c_type.endswith('*')
c_type = c_type[:-1]
temp_name = 'tup%d' % self.unique_id
self.unique_id += 1
self.def_write_ind('%s %s = ', c_type, temp_name)
self.accept(o.rvalue)
self.def_write(';\n')
self._write_tuple_unpacking(temp_name,
lval.items,
rvalue_type.items,
is_return=is_return)
return
raise AssertionError(lval)
def _write_body(self, body):
"""Write a block without the { }."""
for stmt in body:
# Ignore things that look like docstrings
if (isinstance(stmt, ExpressionStmt) and
isinstance(stmt.expr, StrExpr)):
continue
#log('-- %d', self.indent)
self.accept(stmt)
def visit_for_stmt(self, o: 'mypy.nodes.ForStmt') -> T:
if 0:
self.log('ForStmt')
self.log(' index_type %s', o.index_type)
self.log(' inferred_item_type %s', o.inferred_item_type)
self.log(' inferred_iterator_type %s', o.inferred_iterator_type)
func_name = None # does the loop look like 'for x in func():' ?
if (isinstance(o.expr, CallExpr) and
isinstance(o.expr.callee, NameExpr)):
func_name = o.expr.callee.name
# special case: 'for i in xrange(3)'
if func_name == 'xrange':
index_name = o.index.name
args = o.expr.args
num_args = len(args)
if num_args == 1: # xrange(end)
self.def_write_ind('for (int %s = 0; %s < ', index_name,
index_name)
self.accept(args[0])
self.def_write('; ++%s) ', index_name)
elif num_args == 2: # xrange(being, end)
self.def_write_ind('for (int %s = ', index_name)
self.accept(args[0])
self.def_write('; %s < ', index_name)
self.accept(args[1])
self.def_write('; ++%s) ', index_name)
elif num_args == 3: # xrange(being, end, step)
# Special case to detect a constant -1. This is a static
# heuristic, because it could be negative dynamically. TODO:
# mylib.reverse_xrange() or something?
step = args[2]
if isinstance(step, UnaryExpr) and step.op == '-':
comparison_op = '>'
else:
comparison_op = '<'
self.def_write_ind('for (int %s = ', index_name)
self.accept(args[0])
self.def_write('; %s %s ', index_name, comparison_op)
self.accept(args[1])
self.def_write('; %s += ', index_name)
self.accept(step)
self.def_write(') ')
else:
raise AssertionError()
self.accept(o.body)
return
reverse = False
# for i, x in enumerate(...):
index0_name = None
if func_name == 'enumerate':
assert isinstance(o.index, TupleExpr), o.index
index0 = o.index.items[0]
assert isinstance(index0, NameExpr), index0
index0_name = index0.name # generate int i = 0; ; ++i
# type of 'x' in 'for i, x in enumerate(...)'
item_type = o.inferred_item_type.items[1]
index_expr = o.index.items[1]
# enumerate(mylist) turns into iteration over mylist with variable i
assert len(o.expr.args) == 1, o.expr.args
iterated_over = o.expr.args[0]
elif func_name == 'reversed':
# NOTE: enumerate() and reversed() can't be mixed yet. But you CAN
# reverse iter over tuples.
item_type = o.inferred_item_type
index_expr = o.index
args = o.expr.args
assert len(args) == 1, args
iterated_over = args[0]
reverse = True # use different iterate
elif func_name == 'iteritems':
item_type = o.inferred_item_type
index_expr = o.index
args = o.expr.args
assert len(args) == 1, args
# This should be a dict
iterated_over = args[0]
#log('------------ ITERITEMS OVER %s', iterated_over)
else:
item_type = o.inferred_item_type
index_expr = o.index
iterated_over = o.expr
over_type = self.types[iterated_over]
if isinstance(over_type, TypeAliasType):
over_type = over_type.alias.target
#self.log(' iterating over type %s', over_type)
#self.log(' iterating over type %s', over_type.type.fullname)
over_dict = False
yield_acc = None
if over_type.type.fullname == 'builtins.list':
c_type = GetCType(over_type)
assert c_type.endswith('*'), c_type
c_iter_type = c_type.replace('List', 'ListIter',
1)[:-1] # remove *
# ReverseListIter!
if reverse:
c_iter_type = 'Reverse' + c_iter_type
elif over_type.type.fullname == 'builtins.dict':
# Iterator
c_type = GetCType(over_type)
assert c_type.endswith('*'), c_type
c_iter_type = c_type.replace('Dict', 'DictIter',
1)[:-1] # remove *
over_dict = True
assert not reverse
elif over_type.type.fullname == 'builtins.str':
c_iter_type = 'StrIter'
assert not reverse # can't reverse iterate over string yet
elif over_type.type.fullname == 'typing.Iterator':
# We're iterating over a generator. Create a temporary List<T> on the stack
# to accumulate the results in one big batch.
c_iter_type = GetCType(over_type)
assert len(over_type.args) == 1, over_type.args
inner_c_type = GetCType(over_type.args[0])
yield_acc = ('_for_yield_acc%d' % self.unique_id,
'List<%s>*' % inner_c_type)
self.unique_id += 1
self.def_write_ind('List<%s> %s;\n', inner_c_type, yield_acc[0])
self.def_write_ind('')
self.yield_accumulators[o] = yield_acc
self.current_stmt_node = o
self.accept(iterated_over)
self.current_stmt_node = None
self.def_write(';\n')
else: # assume it's like d.iteritems()? Iterator type
assert False, over_type
if index0_name:
# can't initialize two things in a for loop, so do it on a separate line
if self.decl:
self.local_var_list.append((index0_name, 'int'))
self.def_write_ind('%s = 0;\n', index0_name)
index_update = ', ++%s' % index0_name
else:
index_update = ''
self.def_write_ind('for (%s it(', c_iter_type)
if yield_acc:
self.def_write('&%s', yield_acc[0])
else:
self.accept(iterated_over) # the thing being iterated over
self.def_write('); !it.Done(); it.Next()%s) {\n', index_update)
# for x in it: ...
# for i, x in enumerate(pairs): ...
if isinstance(item_type, Instance) or index0_name:
c_item_type = GetCType(item_type)
self.def_write_ind(' %s ', c_item_type)
self.accept(index_expr)
if over_dict:
self.def_write(' = it.Key();\n')
else:
self.def_write(' = it.Value();\n')
# Register loop variable as a stack root.
# Note we have mylib.Collect() in CommandEvaluator::_Execute(), and
# it's called in a loop by _ExecuteList(). Although the 'child'
# variable is already live by other means.
# TODO: Test how much this affects performance.
if CTypeIsManaged(c_item_type):
self.def_write_ind(' StackRoot _for(&')
self.accept(index_expr)
self.def_write_ind(');\n')
elif isinstance(item_type, TupleType): # for x, y in pairs
if over_dict:
assert isinstance(o.index, TupleExpr), o.index
index_items = o.index.items
assert len(index_items) == 2, index_items
assert len(item_type.items) == 2, item_type.items
key_type = GetCType(item_type.items[0])
val_type = GetCType(item_type.items[1])
#log('** %s key_type %s', item_type.items[0], key_type)
#log('** %s val_type %s', item_type.items[1], val_type)
# TODO(StackRoots): k, v
self.def_write_ind(' %s %s = it.Key();\n', key_type,
index_items[0].name)
self.def_write_ind(' %s %s = it.Value();\n', val_type,
index_items[1].name)
else:
# Example:
# for (ListIter it(mylist); !it.Done(); it.Next()) {
# Tuple2<int, BigStr*> tup1 = it.Value();
# int i = tup1->at0();
# BigStr* s = tup1->at1();
# log("%d %s", i, s);
# }
c_item_type = GetCType(item_type)
if isinstance(o.index, TupleExpr):
# TODO(StackRoots)
temp_name = 'tup%d' % self.unique_id
self.unique_id += 1
self.def_write_ind(' %s %s = it.Value();\n', c_item_type,
temp_name)
self.indent += 1
self._write_tuple_unpacking(temp_name, o.index.items,
item_type.items)
self.indent -= 1
else:
self.def_write_ind(' %s %s = it.Value();\n', c_item_type,
o.index.name)
#self.def_write_ind(' StackRoots _for(&%s)\n;', o.index.name)
else:
raise AssertionError('Unexpected type %s' % item_type)
# Copy of visit_block, without opening {
self.indent += 1
block = o.body
self._write_body(block.body)
self.indent -= 1
self.def_write_ind('}\n')
if o.else_body:
raise AssertionError("can't translate for-else")
def _write_cases(self, switch_expr, cases, default_block):
""" Write a list of (expr, block) pairs """
for expr, body in cases:
assert expr is not None, expr
if not isinstance(expr, CallExpr):
self.report_error(expr,
'Expected call like case(x), got %s' % expr)
return
for i, arg in enumerate(expr.args):
if i != 0:
self.def_write('\n')
self.def_write_ind('case ')
self.accept(arg)
self.def_write(': ')
self.accept(body)
self.def_write_ind(' break;\n')
if default_block is None:
# an error occurred
return
if default_block is False:
# This is too restrictive
#self.report_error(switch_expr,
# 'switch got no else: for default block')
return
self.def_write_ind('default: ')
self.accept(default_block)
# don't write 'break'
def _write_switch(self, expr, o):
"""Write a switch statement over integers."""
assert len(expr.args) == 1, expr.args
self.def_write_ind('switch (')
self.accept(expr.args[0])
self.def_write(') {\n')
assert len(o.body.body) == 1, o.body.body
if_node = o.body.body[0]
assert isinstance(if_node, IfStmt), if_node
self.indent += 1
cases = []
default_block = util._collect_cases(self.module_path,
if_node,
cases,
errors=self.errors_keep_going)
self._write_cases(expr, cases, default_block)
self.indent -= 1
self.def_write_ind('}\n')
def _write_tag_switch(self, expr, o):
"""Write a switch statement over ASDL types."""
assert len(expr.args) == 1, expr.args
self.def_write_ind('switch (')
self.accept(expr.args[0])
self.def_write('->tag()) {\n')
assert len(o.body.body) == 1, o.body.body
if_node = o.body.body[0]
assert isinstance(if_node, IfStmt), if_node
self.indent += 1
cases = []
default_block = util._collect_cases(self.module_path,
if_node,
cases,
errors=self.errors_keep_going)
self._write_cases(expr, cases, default_block)
self.indent -= 1
self.def_write_ind('}\n')
def _str_switch_cases(self, cases):
cases2 = []
for expr, body in cases:
if not isinstance(expr, CallExpr):
# non-fatal check from _collect_cases
break
args = expr.args
if len(args) != 1:
self.report_error(
expr,
'str_switch can only have case("x"), not case("x", "y")' %
args)
break
if not isinstance(args[0], StrExpr):
self.report_error(
expr,
'str_switch can only be used with constant strings, got %s'
% args[0])
break
s = args[0].value
cases2.append((len(s), s, body))
# Sort by string length
cases2.sort(key=lambda pair: pair[0])
grouped = itertools.groupby(cases2, key=lambda pair: pair[0])
return grouped
def _write_str_switch(self, expr, o):
"""Write a switch statement over strings."""
assert len(expr.args) == 1, expr.args
switch_expr = expr # for later error
switch_var = expr.args[0]
if not isinstance(switch_var, NameExpr):
self.report_error(
expr.args[0],
'str_switch(x) accepts only a variable name, got %s' %
switch_var)
return
self.def_write_ind('switch (len(%s)) {\n' % switch_var.name)
# There can only be one thing under 'with str_switch'
assert len(o.body.body) == 1, o.body.body
if_node = o.body.body[0]
assert isinstance(if_node, IfStmt), if_node
self.indent += 1
cases = []
default_block = util._collect_cases(self.module_path,
if_node,
cases,
errors=self.errors_keep_going)
grouped_cases = self._str_switch_cases(cases)
# Warning: this consumes internal iterator
#self.log('grouped %s', list(grouped_cases))
for str_len, group in grouped_cases:
self.def_write_ind('case %s: {\n' % str_len)
if_num = 0
for _, case_str, block in group:
self.indent += 1
else_str = '' if if_num == 0 else 'else '
self.def_write_ind('%sif (str_equals_c(%s, %s, %d)) ' %
(else_str, switch_var.name,
PythonStringLiteral(case_str), str_len))
self.accept(block)
self.indent -= 1
if_num += 1
self.indent += 1
self.def_write_ind('else {\n')
self.def_write_ind(' goto str_switch_default;\n')
self.def_write_ind('}\n')
self.indent -= 1
self.def_write_ind('}\n')
self.def_write_ind(' break;\n')
if default_block is None:
# an error occurred
return
if default_block is False:
self.report_error(switch_expr,
'str_switch got no else: for default block')
return
self.def_write('\n')
self.def_write_ind('str_switch_default:\n')
self.def_write_ind('default: ')
self.accept(default_block)
self.indent -= 1
self.def_write_ind('}\n')
def visit_with_stmt(self, o: 'mypy.nodes.WithStmt') -> T:
"""
Translate only blocks of this form:
with switch(x) as case:
if case(0):
print('zero')
elif case(1, 2, 3):
print('low')
else:
print('other')
switch(x) {
case 0:
# TODO: need casting here
print('zero')
break;
case 1:
case 2:
case 3:
print('low')
break;
default:
print('other')
break;
}
Or:
with ctx_Bar(bar, x, y):
x()
{
ctx_Bar(bar, x, y)
x();
}
"""
#log('WITH')
#log('expr %s', o.expr)
#log('target %s', o.target)
assert len(o.expr) == 1, o.expr
expr = o.expr[0]
assert isinstance(expr, CallExpr), expr
callee_name = expr.callee.name
if callee_name == 'switch':
self._write_switch(expr, o)
elif callee_name == 'str_switch':
self._write_str_switch(expr, o)
elif callee_name == 'tagswitch':
self._write_tag_switch(expr, o)
else:
assert isinstance(expr, CallExpr), expr
self.def_write_ind('{ // with\n')
self.indent += 1
self.def_write_ind('')
self.accept(expr.callee)
# FIX: Use braced initialization to avoid most-vexing parse when
# there are 0 args!
self.def_write(' ctx{')
for i, arg in enumerate(expr.args):
if i != 0:
self.def_write(', ')
self.accept(arg)
self.def_write('};\n\n')
#self.def_write_ind('')
self._write_body(o.body.body)
self.indent -= 1
self.def_write_ind('}\n')
def visit_del_stmt(self, o: 'mypy.nodes.DelStmt') -> T:
d = o.expr
if isinstance(d, IndexExpr):
self.def_write_ind('')
self.accept(d.base)
if isinstance(d.index, SliceExpr):
# del mylist[:] -> mylist->clear()
sl = d.index
assert sl.begin_index is None, sl
assert sl.end_index is None, sl
self.def_write('->clear()')
else:
# del mydict[mykey] raises KeyError, which we don't want
raise AssertionError(
'Use mylib.maybe_remove(d, key) instead of del d[key]')
self.def_write(';\n')
def _WriteFuncParams(self,
arg_types,
arguments,
update_locals=False,
write_defaults=False):
"""Write params for function/method signatures.
Optionally mutate self.local_vars, and optionally write default arguments.
"""
if write_defaults:
# Check if default args are valid first
num_defaults = 0
for arg in arguments:
if arg.initializer:
t = self.types[arg.initializer]
valid = False
if isinstance(t, NoneType):
valid = True
if isinstance(t, Instance):
# Allowing strings since they're immutable, e.g.
# prefix='' seems OK
if t.type.fullname in ('builtins.bool', 'builtins.int',
'builtins.float',
'builtins.str'):
valid = True
# ASDL enums lex_mode_t, scope_t, ...
if t.type.fullname.endswith('_t'):
valid = True
# Hack for loc__Missing. Should detect the general case.
if t.type.fullname.endswith('loc__Missing'):
valid = True
if not valid:
self.report_error(
arg,
'Invalid default arg %r of type %s (not None, bool, int, float, ASDL enum)'
% (arg.initializer, t))
return
num_defaults += 1
if num_defaults > 1:
name = '[TODO]'
#if class_name:
# name = '%s::%s' % (class_name, func_name)
#else:
# name = func_name
# Report on first arg
self.report_error(
arg, '%s has %d default arguments. Only 1 is allowed' %
(name, num_defaults))
return
first = True # first NOT including self
for arg_type, arg in zip(arg_types, arguments):
if not first:
self.always_write(', ')
# TODO: Turn this on. Having stdlib problems, e.g.
# examples/cartesian.
c_type = GetCType(arg_type, param=False)
arg_name = arg.variable.name
# C++ has implicit 'this'
if arg_name == 'self':
continue
self.always_write('%s %s', c_type, arg_name)
if write_defaults and arg.initializer:
self.always_write(' = ')
# Silly mechanism to activate self.def_write()
self.writing_default_arg = True
self.accept(arg.initializer)
self.writing_default_arg = False
first = False
# Params are locals. There are 4 callers to _WriteFuncParams and we
# only do it in one place. TODO: Check if locals are used in
# __init__ after allocation.
if update_locals:
self.local_var_list.append((arg_name, c_type))
# We can't use __str__ on these Argument objects? That seems like an
# oversight
#self.log('%r', arg)
if 0:
self.log('Argument %s', arg.variable)
self.log(' type_annotation %s', arg.type_annotation)
# I think these are for default values
self.log(' initializer %s', arg.initializer)
self.log(' kind %s', arg.kind)
# Will be set if we're declaring or defining a function that returns
# Iterator[T].
if self.current_func_node in self.yield_accumulators:
if not first:
self.always_write(', ')
arg_name, c_type = self.yield_accumulators[self.current_func_node]
self.always_write('%s %s', c_type, arg_name)
def visit_func_def(self, o: 'mypy.nodes.FuncDef') -> T:
if o.name == '__repr__': # Don't translate
return
# No function prototypes when forward declaring.
if self.forward_decl:
self.virtual.OnMethod(self.current_class_name, o.name)
return
func_name = o.name
virtual = ''
if self.decl:
self.local_var_list = [] # Make a new instance to collect from
self.local_vars[o] = self.local_var_list
if self.virtual.IsVirtual(self.current_class_name, o.name):
virtual = 'virtual '
if not self.decl and self.current_class_name:
# definition looks like
# void Class::method(...);
func_name = join_name((self.current_class_name[-1], o.name))
else:
# declaration inside class { }
func_name = o.name
self.def_write('\n')
c_ret_type, _, c_iter_list_type = GetCReturnType(o.type.ret_type)
if c_iter_list_type is not None:
# The function is a generator. Add an output param that references an
# accumulator for the results.
self.yield_accumulators[o] = ('_out_yield_acc', c_iter_list_type)
# Avoid C++ warnings by prepending [[noreturn]]
noreturn = ''
if func_name in ('e_die', 'e_die_status', 'e_strict', 'e_usage',
'p_die'):
noreturn = '[[noreturn]] '
self.always_write_ind('%s%s%s %s(', noreturn, virtual, c_ret_type,
func_name)
self.current_func_node = o
self._WriteFuncParams(
o.type.arg_types,
o.arguments,
update_locals=True,
# write default values in the declaration only
write_defaults=self.decl)
if self.decl:
self.always_write(');\n')
else:
self.def_write(') ')
if not self.forward_decl:
arg_names = [arg.variable.name for arg in o.arguments]
#log('arg_names %s', arg_names)
#log('local_vars %s', self.local_vars[o])
self.prepend_to_block = [
(lval_name, c_type, lval_name in arg_names)
for (lval_name, c_type) in self.local_vars[o]
]
self.accept(o.body)
self.current_func_node = None
def visit_overloaded_func_def(self,
o: 'mypy.nodes.OverloadedFuncDef') -> T:
pass
def _TracingMetadataDecl(self, o, field_gc, mask_bits):
if mask_bits:
self.always_write_ind('\n')
self.always_write_ind('static constexpr uint32_t field_mask() {\n')
self.always_write_ind(' return ')
for i, b in enumerate(mask_bits):
if i != 0:
self.always_write('\n')
self.always_write_ind(' | ')
self.always_write(b)
self.always_write(';\n')
self.always_write_ind('}\n')
obj_tag, obj_arg = field_gc
if obj_tag == 'HeapTag::FixedSize':
obj_mask = obj_arg
obj_header = 'ObjHeader::ClassFixed(%s, sizeof(%s))' % (obj_mask,
o.name)
elif obj_tag == 'HeapTag::Scanned':
num_pointers = obj_arg
obj_header = 'ObjHeader::ClassScanned(%s, sizeof(%s))' % (
num_pointers, o.name)
else:
raise AssertionError(o.name)
self.always_write('\n')
self.always_write_ind('static constexpr ObjHeader obj_header() {\n')
self.always_write_ind(' return %s;\n' % obj_header)
self.always_write_ind('}\n')
def _MemberDecl(self, o, base_class_name):
# List of field mask expressions
mask_bits = []
if self.virtual.CanReorderFields(split_py_name(o.fullname)):
# No inheritance, so we are free to REORDER member vars, putting
# pointers at the front.
pointer_members = []
non_pointer_members = []
for name in self.current_member_vars:
_, c_type, is_managed = self.current_member_vars[name]
if is_managed:
pointer_members.append(name)
else:
non_pointer_members.append(name)
# So we declare them in the right order
sorted_member_names = pointer_members + non_pointer_members
field_gc = ('HeapTag::Scanned', len(pointer_members))
else:
# Has inheritance
# The field mask of a derived class is unioned with its base's
# field mask.
if base_class_name:
mask_bits.append(
'%s::field_mask()' %
join_name(base_class_name, strip_package=True))
for name in sorted(self.current_member_vars):
_, c_type, is_managed = self.current_member_vars[name]
if is_managed:
mask_bits.append('maskbit(offsetof(%s, %s))' %
(o.name, name))
# A base class with no fields has kZeroMask.
if not base_class_name and not mask_bits:
mask_bits.append('kZeroMask')
sorted_member_names = sorted(self.current_member_vars)
field_gc = ('HeapTag::FixedSize', 'field_mask()')
# Write member variables
#log('MEMBERS for %s: %s', o.name, list(self.current_member_vars.keys()))
if len(self.current_member_vars):
if base_class_name:
self.always_write('\n') # separate from functions
for name in sorted_member_names:
_, c_type, _ = self.current_member_vars[name]
self.always_write_ind('%s %s;\n', c_type, name)
if _IsContextManager(self.current_class_name):
# Copy ctx member vars out of this class
assert self.ctx_member_vars is not None
self.ctx_member_vars[o] = dict(self.current_member_vars)
else:
self._TracingMetadataDecl(o, field_gc, mask_bits)
self.always_write('\n')
self.always_write_ind('DISALLOW_COPY_AND_ASSIGN(%s)\n', o.name)
self.indent -= 1
self.always_write_ind('};\n')
self.always_write('\n')
def _ClassDefDecl(self, o, base_class_name):
self.current_member_vars.clear() # make a new list
self.always_write_ind('class %s', o.name) # block after this
# e.g. class TextOutput : public ColorOutput
if base_class_name:
self.always_write(' : public %s',
join_name(base_class_name, strip_package=True))
self.always_write(' {\n')
self.always_write_ind(' public:\n')
block = o.defs
self.indent += 1
for stmt in block.body:
# Ignore things that look like docstrings
if (isinstance(stmt, ExpressionStmt) and
isinstance(stmt.expr, StrExpr)):
continue
# Constructor is named after class
if isinstance(stmt, FuncDef):
method_name = stmt.name
if method_name == '__init__':
self.always_write_ind('%s(', o.name)
self._WriteFuncParams(stmt.type.arg_types,
stmt.arguments,
write_defaults=True)
self.always_write(');\n')
# Visit for member vars
self.current_method_name = method_name
self.accept(stmt.body)
self.current_method_name = None
continue
if method_name == '__enter__':
continue
if method_name == '__exit__':
# Turn it into a destructor with NO ARGS
self.always_write_ind('~%s();\n', o.name)
continue
if method_name == '__repr__':
# skip during declaration, just like visit_func_def does during definition
continue
# Any other function: Visit for member vars
self.current_method_name = method_name
self.accept(stmt)
self.current_method_name = None
continue
# TODO: Remove this? Everything under a class is a method?
self.accept(stmt)
self._MemberDecl(o, base_class_name)
def _ConstructorImpl(self, o, stmt, base_class_name):
self.def_write('\n')
self.def_write('%s::%s(', o.name, o.name)
self._WriteFuncParams(stmt.type.arg_types, stmt.arguments)
self.def_write(')')
first_index = 0
# Skip docstring
maybe_skip_stmt = stmt.body.body[0]
if (isinstance(maybe_skip_stmt, ExpressionStmt) and
isinstance(maybe_skip_stmt.expr, StrExpr)):
first_index += 1
# Check for Base.__init__(self, ...) and move that to the initializer list.
first_stmt = stmt.body.body[first_index]
if (isinstance(first_stmt, ExpressionStmt) and
isinstance(first_stmt.expr, CallExpr)):
expr = first_stmt.expr
#log('expr %s', expr)
callee = first_stmt.expr.callee
# TextOutput() : ColorOutput(f), ... {
if (isinstance(callee, MemberExpr) and callee.name == '__init__'):
base_constructor_args = expr.args
#log('ARGS %s', base_constructor_args)
self.def_write(' : %s(',
join_name(base_class_name, strip_package=True))
for i, arg in enumerate(base_constructor_args):
if i == 0:
continue # Skip 'this'
if i != 1:
self.def_write(', ')
self.accept(arg)
self.def_write(')')
first_index += 1
self.def_write(' {\n')
# Now visit the rest of the statements
self.indent += 1
if _IsContextManager(self.current_class_name):
# For ctx_* classes only, do gHeap.PushRoot() for all the pointer
# members
member_vars = self.ctx_member_vars[o]
for name in sorted(member_vars):
_, c_type, is_managed = member_vars[name]
if is_managed:
# VALIDATE_ROOTS doesn't complain even if it's not
# initialized? Should be initialized after PushRoot().
#self.def_write_ind('this->%s = nullptr;\n' % name)
self.def_write_ind(
'gHeap.PushRoot(reinterpret_cast<RawObject**>(&(this->%s)));\n'
% name)
for node in stmt.body.body[first_index:]:
self.accept(node)
self.indent -= 1
self.def_write('}\n')
def _DestructorImpl(self, o, stmt, base_class_name):
self.always_write('\n')
self.always_write_ind('%s::~%s()', o.name, o.name)
self.def_write(' {\n')
self.indent += 1
# TODO:
# - Can't throw exception in destructor.
# - Check that you don't return early from destructor. If so, we skip
# PopRoot(), which messes up the invariant!
for node in stmt.body.body:
self.accept(node)
# For ctx_* classes only , gHeap.PopRoot() for all the pointer members
if _IsContextManager(self.current_class_name):
member_vars = self.ctx_member_vars[o]
for name in sorted(member_vars):
_, c_type, is_managed = member_vars[name]
if is_managed:
self.def_write_ind('gHeap.PopRoot();\n')
else:
self.report_error(
o, 'Any class with __exit__ should be named ctx_Foo (%s)' %
(self.current_class_name, ))
return
self.indent -= 1
self.def_write('}\n')
def _ClassDefImpl(self, o, base_class_name):
block = o.defs
for stmt in block.body:
if isinstance(stmt, FuncDef):
if stmt.name == '__init__':
self._ConstructorImpl(o, stmt, base_class_name)
continue
if stmt.name == '__enter__': # We never use these
continue
if stmt.name == '__exit__':
self._DestructorImpl(o, stmt, base_class_name)
continue
self.accept(stmt)
def visit_class_def(self, o: 'mypy.nodes.ClassDef') -> T:
#log(' CLASS %s', o.name)
base_class_name = None # single inheritance only
for b in o.base_type_exprs:
if isinstance(b, NameExpr):
# TODO: inherit from std::exception?
if b.name != 'object' and b.name != 'Exception':
base_class_name = split_py_name(b.fullname)
elif isinstance(b, MemberExpr): # vm._Executor -> vm::_Executor
assert isinstance(b.expr, NameExpr), b
base_class_name = split_py_name(b.expr.fullname) + (b.name, )
# Forward declare types because they may be used in prototypes
if self.forward_decl:
self.always_write_ind('class %s;\n', o.name)
if base_class_name:
self.virtual.OnSubclass(base_class_name,
split_py_name(o.fullname))
# Visit class body so we get method declarations
self.current_class_name = split_py_name(o.fullname)
self._write_body(o.defs.body)
self.current_class_name = None
return
if self.decl:
self.current_class_name = split_py_name(o.fullname)
self._ClassDefDecl(o, base_class_name)
self.current_class_name = None
return
self.current_class_name = split_py_name(o.fullname)
self._ClassDefImpl(o, base_class_name)
self.current_class_name = None # Stop prefixing functions with class
def visit_global_decl(self, o: 'mypy.nodes.GlobalDecl') -> T:
self.report_error(o, 'global not allowed')
def visit_nonlocal_decl(self, o: 'mypy.nodes.NonlocalDecl') -> T:
pass
def visit_decorator(self, o: 'mypy.nodes.Decorator') -> T:
pass
def visit_var(self, o: 'mypy.nodes.Var') -> T:
pass
# Module structure
def visit_import(self, o: 'mypy.nodes.Import') -> T:
for name, as_name in o.ids:
if as_name is not None:
# import time as time_
self.imported_names.add(as_name)
else:
# import libc
self.imported_names.add(name)
def visit_import_from(self, o: 'mypy.nodes.ImportFrom') -> T:
"""
Write C++ namespace aliases and 'using' for imports.
We need them in the 'decl' phase for default arguments like
runtime_asdl::scope_e -> scope_e
"""
# For MemberExpr . -> module::func() or this->field. Also needed in
# the decl phase for default arg values.
for name, alias in o.names:
if alias:
self.imported_names.add(alias)
else:
self.imported_names.add(name)
if o.id in ('__future__', 'typing'):
return # do nothing
#self.log(' %s ImportFrom id: %s', self.decl, o.id)
for name, alias in o.names:
#self.log('ImportFrom id: %s name: %s alias: %s', o.id, name, alias)
if name == 'log': # varargs translation
continue
if name == 'stderr_line': # TODO: remove this
continue
if o.id == 'mycpp.mylib':
# These mylib functions are translated in a special way
if name in ('switch', 'tagswitch', 'str_switch', 'iteritems',
'NewDict', 'probe'):
continue
# STDIN_FILENO is #included
if name == 'STDIN_FILENO':
continue
# A heuristic that works for the Oil import style.
if '.' in o.id:
# from mycpp.mylib import log => using mylib::log
translate_import = True
else:
# from core import util => NOT translated
# We just rely on 'util' being defined.
translate_import = False
if translate_import:
dotted_parts = o.id.split('.')
last_dotted = dotted_parts[-1]
# Omit these:
# from _gen.ysh import grammar_nt
if last_dotted == 'ysh':
return
# from _devbuild.gen import syntax_asdl
if last_dotted == 'gen':
return
# Problem:
# - The decl stage has to return yaks_asdl::mod_def, so imports should go there
# - But if you change this to decl_write() instead of
# def_write(), you end up 'using error::e_usage' in say
# 'assign_osh', and it hasn't been defined yet.
if alias:
# using runtime_asdl::emit_e = EMIT;
self.def_write_ind('using %s = %s::%s;\n', alias,
last_dotted, name)
else:
# from _devbuild.gen.id_kind_asdl import Id
# -> using id_kind_asdl::Id.
using_str = 'using %s::%s;\n' % (last_dotted, name)
self.def_write_ind(using_str)
# Fully qualified:
# self.def_write_ind('using %s::%s;\n', '::'.join(dotted_parts), name)
# Hack for default args. Without this limitation, we write
# 'using' of names that aren't declared yet.
# suffix_op is needed for string_ops.py, for some reason
if (name in ('Id', 'scope_e', 'lex_mode_e', 'suffix_op',
'sh_lvalue', 'part_value', 'loc', 'word',
'word_part', 'cmd_value', 'hnode')):
self.decl_write(using_str)
else:
# If we're importing a module without an alias, we don't need to do
# anything. 'namespace cmd_eval' is already defined.
if not alias:
return
# from asdl import format as fmt
# -> namespace fmt = format;
self.def_write_ind('namespace %s = %s;\n', alias, name)
def visit_import_all(self, o: 'mypy.nodes.ImportAll') -> T:
pass
# Statements
def visit_block(self, block: 'mypy.nodes.Block') -> T:
self.def_write('{\n') # not indented to use same line as while/if
self.indent += 1
if self.prepend_to_block:
# TODO: put the pointers first, and then register a single
# StackRoots record.
done = set()
for lval_name, c_type, is_param in self.prepend_to_block:
if not is_param and lval_name not in done:
if util.SMALL_STR and c_type == 'Str':
self.def_write_ind('%s %s(nullptr);\n', c_type,
lval_name)
else:
rhs = ' = nullptr' if CTypeIsManaged(c_type) else ''
self.def_write_ind('%s %s%s;\n', c_type, lval_name,
rhs)
done.add(lval_name)
# Figure out if we have any roots to write with StackRoots
roots = [] # keep it sorted
for lval_name, c_type, is_param in self.prepend_to_block:
#self.log('%s %s %s', lval_name, c_type, is_param)
if lval_name not in roots and CTypeIsManaged(c_type):
roots.append(lval_name)
#self.log('roots %s', roots)
if len(roots):
if (self.stack_roots_warn and
len(roots) > self.stack_roots_warn):
log('WARNING: %s() has %d stack roots. Consider refactoring this function.'
% (self.current_func_node.fullname, len(roots)))
for i, r in enumerate(roots):
self.def_write_ind('StackRoot _root%d(&%s);\n' % (i, r))
self.def_write('\n')
self.prepend_to_block = None
self._write_body(block.body)
self.indent -= 1
self.def_write_ind('}\n')
def visit_expression_stmt(self, o: 'mypy.nodes.ExpressionStmt') -> T:
# TODO: Avoid writing docstrings.
# If it's just a string, then we don't need it.
self.def_write_ind('')
self.accept(o.expr)
self.def_write(';\n')
def visit_operator_assignment_stmt(
self, o: 'mypy.nodes.OperatorAssignmentStmt') -> T:
self.def_write_ind('')
self.accept(o.lvalue)
self.def_write(' %s= ', o.op) # + to +=
self.accept(o.rvalue)
self.def_write(';\n')
def visit_while_stmt(self, o: 'mypy.nodes.WhileStmt') -> T:
self.def_write_ind('while (')
self.accept(o.expr)
self.def_write(') ')
self.accept(o.body)
def visit_return_stmt(self, o: 'mypy.nodes.ReturnStmt') -> T:
# Examples:
# return
# return None
# return my_int + 3;
self.def_write_ind('return ')
if o.expr:
if not (isinstance(o.expr, NameExpr) and o.expr.name == 'None'):
# Note: the type of the return expression (self.types[o.expr])
# and the return type of the FUNCTION are different. Use the
# latter.
ret_type = self.current_func_node.type.ret_type
c_ret_type, returning_tuple, _ = GetCReturnType(ret_type)
# return '', None # tuple literal
# but NOT
# return tuple_func()
if returning_tuple and isinstance(o.expr, TupleExpr):
self.def_write('%s(' % c_ret_type)
for i, item in enumerate(o.expr.items):
if i != 0:
self.def_write(', ')
self.accept(item)
self.def_write(');\n')
return
# Not returning tuple
self.accept(o.expr)
self.def_write(';\n')
def visit_assert_stmt(self, o: 'mypy.nodes.AssertStmt') -> T:
pass
def visit_if_stmt(self, o: 'mypy.nodes.IfStmt') -> T:
# Not sure why this wouldn't be true
assert len(o.expr) == 1, o.expr
cond = o.expr[0]
if not _CheckCondition(cond, self.types):
self.report_error(
o,
"Use explicit len(obj) or 'obj is not None' for mystr, mylist, mydict"
)
return
# Omit anything that looks like if __name__ == ...
if (isinstance(cond, ComparisonExpr) and
isinstance(cond.operands[0], NameExpr) and
cond.operands[0].name == '__name__'):
return
# Omit if 0:
if isinstance(cond, IntExpr) and cond.value == 0:
# But write else: body
# Note: this would be invalid at the top level!
if o.else_body:
self.accept(o.else_body)
return
# Omit if TYPE_CHECKING blocks. They contain type expressions that
# don't type check!
if isinstance(cond, NameExpr) and cond.name == 'TYPE_CHECKING':
return
# mylib.CPP
if isinstance(cond, MemberExpr) and cond.name == 'CPP':
# just take the if block
self.def_write_ind('// if MYCPP\n')
self.def_write_ind('')
for node in o.body:
self.accept(node)
self.def_write_ind('// endif MYCPP\n')
return
# mylib.PYTHON
if isinstance(cond, MemberExpr) and cond.name == 'PYTHON':
if o.else_body:
self.def_write_ind('// if not PYTHON\n')
self.def_write_ind('')
self.accept(o.else_body)
self.def_write_ind('// endif MYCPP\n')
return
self.def_write_ind('if (')
for e in o.expr:
self.accept(e)
self.def_write(') ')
for node in o.body:
self.accept(node)
if o.else_body:
self.def_write_ind('else ')
self.accept(o.else_body)
def visit_break_stmt(self, o: 'mypy.nodes.BreakStmt') -> T:
self.def_write_ind('break;\n')
def visit_continue_stmt(self, o: 'mypy.nodes.ContinueStmt') -> T:
self.def_write_ind('continue;\n')
def visit_pass_stmt(self, o: 'mypy.nodes.PassStmt') -> T:
self.def_write_ind('; // pass\n')
def visit_raise_stmt(self, o: 'mypy.nodes.RaiseStmt') -> T:
# C++ compiler is aware of assert(0) for unreachable code
if o.expr and isinstance(o.expr, CallExpr):
if o.expr.callee.name == 'AssertionError':
self.def_write_ind('assert(0); // AssertionError\n')
return
if o.expr.callee.name == 'NotImplementedError':
self.def_write_ind(
'FAIL(kNotImplemented); // Python NotImplementedError\n')
return
self.def_write_ind('throw ')
# it could be raise -> throw ; . OSH uses that.
if o.expr:
self.accept(o.expr)
self.def_write(';\n')
def visit_try_stmt(self, o: 'mypy.nodes.TryStmt') -> T:
self.def_write_ind('try ')
self.accept(o.body)
caught = False
for t, v, handler in zip(o.types, o.vars, o.handlers):
c_type = None
if isinstance(t, NameExpr):
if t.name in ('IOError', 'OSError'):
self.report_error(
handler,
'Use except (IOError, OSError) rather than catching just one'
)
c_type = '%s*' % t.name
elif isinstance(t, MemberExpr):
# Heuristic
c_type = '%s::%s*' % (t.expr.name, t.name)
elif isinstance(t, TupleExpr):
if len(t.items) == 2:
e1 = t.items[0]
e2 = t.items[1]
if isinstance(e1, NameExpr) and isinstance(e2, NameExpr):
names = [e1.name, e2.name]
names.sort()
if names == ['IOError', 'OSError']:
c_type = 'IOError_OSError*' # Base class in mylib
else:
raise AssertionError()
if c_type is None:
c_type = 'INVALID_TRY_EXCEPT' # Causes compile error
if v:
self.def_write_ind('catch (%s %s) ', c_type, v.name)
else:
self.def_write_ind('catch (%s) ', c_type)
self.accept(handler)
caught = True
# DUMMY to prevent compile errors
# TODO: Remove this
if not caught:
self.def_write_ind('catch (std::exception const&) { }\n')
if o.else_body:
self.report_error(o, 'try/else not supported')
if o.finally_body:
self.report_error(o, 'try/finally not supported')
def visit_print_stmt(self, o: 'mypy.nodes.PrintStmt') -> T:
self.report_error(
o,
'File should start with "from __future__ import print_function"')
def visit_exec_stmt(self, o: 'mypy.nodes.ExecStmt') -> T:
self.report_error(o, 'exec not allowed')