cpython/Lib/compiler/pycodegen.py

798 lines
23 KiB
Python
Raw Normal View History

"""Python bytecode generator
Currently contains generic ASTVisitor code, a LocalNameFinder, and a
CodeGenerator. Eventually, this will get split into the ASTVisitor as
a generic tool and CodeGenerator as a specific tool.
"""
from p2c import transformer, ast
import dis
import misc
import marshal
import new
import string
import sys
import os
import stat
import struct
def parse(path):
f = open(path)
src = f.read()
f.close()
t = transformer.Transformer()
return t.parsesuite(src)
def walk(tree, visitor):
w = ASTVisitor()
w.preorder(tree, visitor)
return w.visitor
class ASTVisitor:
"""Performs a depth-first walk of the AST
The ASTVisitor will walk the AST, performing either a preorder or
postorder traversal depending on which method is called.
methods:
preorder(tree, visitor)
postorder(tree, visitor)
tree: an instance of ast.Node
visitor: an instance with visitXXX methods
The ASTVisitor is responsible for walking over the tree in the
correct order. For each node, it checks the visitor argument for
a method named 'visitNodeType' where NodeType is the name of the
node's class, e.g. Classdef. If the method exists, it is called
with the node as its sole argument.
The visitor method for a particular node type can control how
child nodes are visited during a preorder walk. (It can't control
the order during a postorder walk, because it is called _after_
the walk has occurred.) The ASTVisitor modifies the visitor
argument by adding a visit method to the visitor; this method can
be used to visit a particular child node. If the visitor method
returns a true value, the ASTVisitor will not traverse the child
nodes.
XXX The interface for controlling the preorder walk needs to be
re-considered. The current interface is convenient for visitors
that mostly let the ASTVisitor do everything. For something like
a code generator, where you want to walk to occur in a specific
order, it's a pain to add "return 1" to the end of each method.
XXX Perhaps I can use a postorder walk for the code generator?
"""
VERBOSE = 1
def __init__(self):
self.node = None
def preorder(self, tree, visitor):
"""Do preorder walk of tree using visitor"""
self.visitor = visitor
visitor.visit = self._preorder
self._preorder(tree)
def _preorder(self, node):
stop = self.dispatch(node)
if stop:
return
for child in node.getChildren():
if isinstance(child, ast.Node):
self._preorder(child)
def postorder(self, tree, visitor):
"""Do preorder walk of tree using visitor"""
self.visitor = visitor
visitor.visit = self._postorder
self._postorder(tree)
def _postorder(self, tree):
for child in node.getChildren():
if isinstance(child, ast.Node):
self._preorder(child)
self.dispatch(node)
def dispatch(self, node):
self.node = node
className = node.__class__.__name__
meth = getattr(self.visitor, 'visit' + className, None)
if self.VERBOSE:
print "dispatch", className, (meth and meth.__name__ or '')
if meth:
return meth(node)
class CodeGenerator:
def __init__(self, filename=None):
self.filename = filename
self.code = PythonVMCode(filename=filename)
self.code.setFlags(0)
self.locals = misc.Stack()
# track the current and max stack size
# XXX does this belong here or in the PythonVMCode?
self.curStack = 0
self.maxStack = 0
def emit(self):
"""Create a Python code object
XXX It is confusing that this method isn't related to the
method named emit in the PythonVMCode.
"""
return self.code.makeCodeObject(self.maxStack)
def push(self, n):
self.curStack = self.curStack + n
if self.curStack > self.maxStack:
self.maxStack = self.curStack
def pop(self, n):
if n >= self.curStack:
self.curStack = self.curStack - n
else:
self.curStack = 0
def visitDiscard(self, node):
return 1
def visitModule(self, node):
lnf = walk(node.node, LocalNameFinder())
self.locals.push(lnf.getLocals())
self.visit(node.node)
self.code.emit('LOAD_CONST', None)
self.code.emit('RETURN_VALUE')
return 1
def visitFunction(self, node):
codeBody = NestedCodeGenerator(node, filename=self.filename)
walk(node, codeBody)
self.code.setLineNo(node.lineno)
self.code.emit('LOAD_CONST', codeBody)
self.code.emit('MAKE_FUNCTION', 0)
self.code.emit('STORE_NAME', node.name)
return 1
def visitCallFunc(self, node):
self.visit(node.node)
for arg in node.args:
self.visit(arg)
self.code.callFunction(len(node.args))
return 1
def visitIf(self, node):
after = ForwardRef()
for test, suite in node.tests:
self.code.setLineNo(test.lineno)
self.visit(test)
dest = ForwardRef()
self.code.jumpIfFalse(dest)
self.code.popTop()
self.visit(suite)
self.code.jumpForward(after)
dest.bind(self.code.getCurInst())
self.code.popTop()
if node.else_:
self.visit(node.else_)
after.bind(self.code.getCurInst())
return 1
def visitCompare(self, node):
"""Comment from compile.c follows:
The following code is generated for all but the last
comparison in a chain:
label: on stack: opcode: jump to:
a <code to load b>
a, b DUP_TOP
a, b, b ROT_THREE
b, a, b COMPARE_OP
b, 0-or-1 JUMP_IF_FALSE L1
b, 1 POP_TOP
b
We are now ready to repeat this sequence for the next
comparison in the chain.
For the last we generate:
b <code to load c>
b, c COMPARE_OP
0-or-1
If there were any jumps to L1 (i.e., there was more than one
comparison), we generate:
0-or-1 JUMP_FORWARD L2
L1: b, 0 ROT_TWO
0, b POP_TOP
0
L2: 0-or-1
"""
self.visit(node.expr)
# if refs are never emitted, subsequent bind call has no effect
l1 = ForwardRef()
l2 = ForwardRef()
for op, code in node.ops[:-1]:
# emit every comparison except the last
self.visit(code)
self.code.dupTop()
self.code.rotThree()
self.code.compareOp(op)
self.code.jumpIfFalse(l1)
self.code.popTop()
if node.ops:
# emit the last comparison
op, code = node.ops[-1]
self.visit(code)
self.code.compareOp(op)
if len(node.ops) > 1:
self.code.jumpForward(l2)
l1.bind(self.code.getCurInst())
self.code.rotTwo()
self.code.popTop()
l2.bind(self.code.getCurInst())
return 1
def binaryOp(self, node, op):
self.visit(node.left)
self.visit(node.right)
self.code.emit(op)
self.pop(1)
return 1
def visitAdd(self, node):
return self.binaryOp(node, 'BINARY_ADD')
def visitSub(self, node):
return self.binaryOp(node, 'BINARY_SUBTRACT')
def visitMul(self, node):
return self.binaryOp(node, 'BINARY_MULTIPLY')
def visitDiv(self, node):
return self.binaryOp(node, 'BINARY_DIVIDE')
def visitName(self, node):
locals = self.locals.top()
if locals.has_elt(node.name):
self.code.loadFast(node.name)
else:
self.code.loadGlobal(node.name)
self.push(1)
def visitConst(self, node):
self.code.loadConst(node.value)
self.push(1)
def visitReturn(self, node):
self.code.setLineNo(node.lineno)
self.visit(node.value)
self.code.returnValue()
return 1
def visitRaise(self, node):
self.code.setLineNo(node.lineno)
n = 0
if node.expr1:
self.visit(node.expr1)
n = n + 1
if node.expr2:
self.visit(node.expr2)
n = n + 1
if node.expr3:
self.visit(node.expr3)
n = n + 1
self.code.raiseVarargs(n)
return 1
def visitPrint(self, node):
self.code.setLineNo(node.lineno)
for child in node.nodes:
self.visit(child)
self.code.emit('PRINT_ITEM')
self.pop(len(node.nodes))
return 1
def visitPrintnl(self, node):
self.visitPrint(node)
self.code.emit('PRINT_NEWLINE')
return 1
class NestedCodeGenerator(CodeGenerator):
"""Generate code for a function object within another scope
XXX not clear that this subclass is needed
"""
super_init = CodeGenerator.__init__
def __init__(self, func, filename='<?>'):
"""code and args of function or class being walked
XXX need to separately pass to ASTVisitor. the constructor
only uses the code object to find the local names
Copies code form parent __init__ rather than calling it.
"""
self.name = func.name
self.super_init(filename)
args = func.argnames
self.code = PythonVMCode(len(args), name=func.name,
filename=filename)
if func.varargs:
self.code.setVarArgs()
if func.kwargs:
self.code.setKWArgs()
lnf = walk(func.code, LocalNameFinder(args))
self.locals.push(lnf.getLocals())
def __repr__(self):
return "<NestedCodeGenerator: %s>" % self.name
def visitFunction(self, node):
lnf = walk(node.code, LocalNameFinder(node.argnames))
self.locals.push(lnf.getLocals())
# XXX need to handle def foo((a, b)):
self.code.setLineNo(node.lineno)
self.visit(node.code)
self.code.emit('LOAD_CONST', None)
self.code.emit('RETURN_VALUE')
return 1
class LocalNameFinder:
def __init__(self, names=()):
self.names = misc.Set()
for name in names:
self.names.add(name)
def getLocals(self):
return self.names
def visitFunction(self, node):
self.names.add(node.name)
return 1
def visitImport(self, node):
for name in node.names:
self.names.add(name)
def visitFrom(self, node):
for name in node.names:
self.names.add(name)
def visitClassdef(self, node):
self.names.add(node.name)
return 1
def visitAssName(self, node):
self.names.add(node.name)
class Label:
def __init__(self, num):
self.num = num
def __repr__(self):
return "Label(%d)" % self.num
class ForwardRef:
count = 0
def __init__(self, id=None, val=None):
if id is None:
id = ForwardRef.count
ForwardRef.count = ForwardRef.count + 1
self.id = id
self.val = val
def __repr__(self):
if self.val:
return "ForwardRef(val=%d)" % self.val
else:
return "ForwardRef(id=%d)" % self.id
def bind(self, inst):
self.val = inst
def resolve(self):
return self.val
def add_hook(hooks, type, meth):
"""Helper function for PythonVMCode _emit_hooks"""
l = hooks.get(type, [])
l.append(meth)
hooks[type] = l
class PythonVMCode:
"""Creates Python code objects
The new module is used to create the code object. The following
attribute definitions are included from the reference manual:
co_name gives the function name
co_argcount is the number of positional arguments (including
arguments with default values)
co_nlocals is the number of local variables used by the function
(including arguments)
co_varnames is a tuple containing the names of the local variables
(starting with the argument names)
co_code is a string representing the sequence of bytecode instructions
co_consts is a tuple containing the literals used by the bytecode
co_names is a tuple containing the names used by the bytecode
co_filename is the filename from which the code was compiled
co_firstlineno is the first line number of the function
co_lnotab is a string encoding the mapping from byte code offsets
to line numbers (for detais see the source code of the
interpreter)
see code com_set_lineno and com_add_lnotab
it's a string with 2bytes per set_lineno
co_stacksize is the required stack size (including local variables)
co_flags is an integer encoding a number of flags for the
interpreter.
The following flag bits are defined for co_flags: bit 2 is set if
the function uses the "*arguments" syntax to accept an arbitrary
number of positional arguments; bit 3 is set if the function uses
the "**keywords" syntax to accept arbitrary keyword arguments;
other bits are used internally or reserved for future use.
If a code object represents a function, the first item in
co_consts is the documentation string of the function, or None if
undefined.
"""
# XXX flag bits
VARARGS = 0x04
KWARGS = 0x08
def __init__(self, argcount=0, name='?', filename='<?>',
docstring=None):
# XXX why is the default value for flags 3?
self.insts = []
# used by makeCodeObject
self.argcount = argcount
self.code = ''
self.consts = [docstring]
self.filename = filename
self.flags = 3
self.name = name
self.names = []
self.varnames = []
# lnotab support
self.firstlineno = 0
self.lastlineno = 0
self.last_addr = 0
self.lnotab = ''
def __repr__(self):
return "<bytecode: %d instrs>" % len(self.insts)
def setFlags(self, val):
"""XXX for module's function"""
self.flags = 0
def setVarArgs(self):
self.flags = self.flags | self.VARARGS
def setKWArgs(self):
self.flags = self.flags | self.KWARGS
def getCurInst(self):
return len(self.insts)
def getNextInst(self):
return len(self.insts) + 1
def dump(self, io=sys.stdout):
i = 0
for inst in self.insts:
if inst[0] == 'SET_LINENO':
io.write("\n")
io.write(" %3d " % i)
if len(inst) == 1:
io.write("%s\n" % inst)
else:
io.write("%-15.15s\t%s\n" % inst)
i = i + 1
def makeCodeObject(self, stacksize):
"""Make a Python code object
This creates a Python code object using the new module. This
seems simpler than reverse-engineering the way marshal dumps
code objects into .pyc files. One of the key difficulties is
figuring out how to layout references to code objects that
appear on the VM stack; e.g.
3 SET_LINENO 1
6 LOAD_CONST 0 (<code object fact at 8115878 [...]
9 MAKE_FUNCTION 0
12 STORE_NAME 0 (fact)
"""
self._findOffsets()
lnotab = LineAddrTable()
for t in self.insts:
opname = t[0]
if len(t) == 1:
lnotab.addCode(chr(self.opnum[opname]))
elif len(t) == 2:
oparg = self._convertArg(opname, t[1])
if opname == 'SET_LINENO':
lnotab.nextLine(oparg)
hi, lo = divmod(oparg, 256)
lnotab.addCode(chr(self.opnum[opname]) + chr(lo) +
chr(hi))
# why is a module a special case?
if self.flags == 0:
nlocals = 0
else:
nlocals = len(self.varnames)
co = new.code(self.argcount, nlocals, stacksize,
self.flags, lnotab.getCode(), self._getConsts(),
tuple(self.names), tuple(self.varnames),
self.filename, self.name, self.firstlineno,
lnotab.getTable())
return co
def _getConsts(self):
"""Return a tuple for the const slot of a code object
Converts PythonVMCode objects to code objects
"""
l = []
for elt in self.consts:
if isinstance(elt, CodeGenerator):
l.append(elt.emit())
else:
l.append(elt)
return tuple(l)
def _findOffsets(self):
"""Find offsets for use in resolving ForwardRefs"""
self.offsets = []
cur = 0
for t in self.insts:
self.offsets.append(cur)
l = len(t)
if l == 1:
cur = cur + 1
elif l == 2:
arg = t[1]
if isinstance(arg, ForwardRef):
arg.__offset = cur
cur = cur + 3
def _convertArg(self, op, arg):
"""Convert the string representation of an arg to a number
The specific handling depends on the opcode.
XXX This first implementation isn't going to be very
efficient.
"""
if op == 'SET_LINENO':
return arg
if op == 'LOAD_CONST':
return self._lookupName(arg, self.consts)
if op == 'LOAD_FAST':
if arg in self.names:
return self._lookupName(arg, self.varnames)
else:
return self._lookupName(arg, self.varnames, self.names)
if op == 'LOAD_GLOBAL':
return self._lookupName(arg, self.names)
if op == 'STORE_NAME':
return self._lookupName(arg, self.names)
if op == 'COMPARE_OP':
return self.cmp_op.index(arg)
if self.hasjrel.has_elt(op):
return self.offsets[arg.resolve()]
if self.hasjabs.has_elt(op):
return self.offsets[arg.resolve()] - arg.__offset
return arg
def _lookupName(self, name, list, list2=None):
"""Return index of name in list, appending if necessary
Yicky hack: Second list can be used for lookup of local names
where the name needs to be added to varnames and names.
"""
if name in list:
return list.index(name)
else:
end = len(list)
list.append(name)
if list2 is not None:
list2.append(name)
return end
# Convert some stuff from the dis module for local use
cmp_op = list(dis.cmp_op)
hasjrel = misc.Set()
for i in dis.hasjrel:
hasjrel.add(dis.opname[i])
hasjabs = misc.Set()
for i in dis.hasjabs:
hasjabs.add(dis.opname[i])
opnum = {}
for num in range(len(dis.opname)):
opnum[dis.opname[num]] = num
# the interface below here seemed good at first. upon real use,
# it seems redundant to add a function for each opcode,
# particularly because the method and opcode basically have the
# same name.
# on the other hand, we need to track things like stack depth in
# order to generator code objects. if we wrap instructions in a
# method, we get an easy way to track these. a simpler
# approach, however, would be to define hooks that can be called
# by emit.
def setLineNo(self, num):
self.emit('SET_LINENO', num)
def popTop(self):
self.emit('POP_TOP')
def dupTop(self):
self.emit('DUP_TOP')
def rotTwo(self):
self.emit('ROT_TWO')
def rotThree(self):
self.emit('ROT_THREE')
def jumpIfFalse(self, dest):
self.emit('JUMP_IF_FALSE', dest)
def loadFast(self, name):
self.emit('LOAD_FAST', name)
def loadGlobal(self, name):
self.emit('LOAD_GLOBAL', name)
def binaryAdd(self):
self.emit('BINARY_ADD')
def compareOp(self, op):
self.emit('COMPARE_OP', op)
def loadConst(self, val):
self.emit('LOAD_CONST', val)
def returnValue(self):
self.emit('RETURN_VALUE')
def jumpForward(self, dest):
self.emit('JUMP_FORWARD', dest)
def raiseVarargs(self, num):
self.emit('RAISE_VARARGS', num)
def callFunction(self, num):
self.emit('CALL_FUNCTION', num)
# this version of emit + arbitrary hooks might work, but it's damn
# messy.
def emit(self, *args):
self._emitDispatch(args[0], args[1:])
self.insts.append(args)
def _emitDispatch(self, type, args):
for func in self._emit_hooks.get(type, []):
func(self, args)
_emit_hooks = {}
class LineAddrTable:
"""lnotab
This class builds the lnotab, which is undocumented but described
by com_set_lineno in compile.c. Here's an attempt at explanation:
For each SET_LINENO instruction after the first one, two bytes are
added to lnotab. (In some cases, multiple two-byte entries are
added.) The first byte is the distance in bytes between the
instruction for the last SET_LINENO and the current SET_LINENO.
The second byte is offset in line numbers. If either offset is
greater than 255, multiple two-byte entries are added -- one entry
for each factor of 255.
"""
def __init__(self):
self.code = []
self.codeOffset = 0
self.firstline = 0
self.lastline = 0
self.lastoff = 0
self.lnotab = []
def addCode(self, code):
self.code.append(code)
self.codeOffset = self.codeOffset + len(code)
def nextLine(self, lineno):
if self.firstline == 0:
self.firstline = lineno
self.lastline = lineno
else:
# compute deltas
addr = self.codeOffset - self.lastoff
line = lineno - self.lastline
while addr > 0 or line > 0:
# write the values in 1-byte chunks that sum
# to desired value
trunc_addr = addr
trunc_line = line
if trunc_addr > 255:
trunc_addr = 255
if trunc_line > 255:
trunc_line = 255
self.lnotab.append(trunc_addr)
self.lnotab.append(trunc_line)
addr = addr - trunc_addr
line = line - trunc_line
self.lastline = lineno
self.lastoff = self.codeOffset
def getCode(self):
return string.join(self.code, '')
def getTable(self):
return string.join(map(chr, self.lnotab), '')
class CompiledModule:
"""Store the code object for a compiled module
XXX Not clear how the code objects will be stored. Seems possible
that a single code attribute is sufficient, because it will
contains references to all the need code objects. That might be
messy, though.
"""
MAGIC = (20121 | (ord('\r')<<16) | (ord('\n')<<24))
def __init__(self, source, filename):
self.source = source
self.filename = filename
def compile(self):
t = transformer.Transformer()
self.ast = t.parsesuite(self.source)
cg = CodeGenerator(self.filename)
walk(self.ast, cg)
self.code = cg.emit()
def dump(self, path):
"""create a .pyc file"""
f = open(path, 'wb')
f.write(self._pyc_header())
marshal.dump(self.code, f)
f.close()
def _pyc_header(self):
# compile.c uses marshal to write a long directly, with
# calling the interface that would also generate a 1-byte code
# to indicate the type of the value. simplest way to get the
# same effect is to call marshal and then skip the code.
magic = marshal.dumps(self.MAGIC)[1:]
mtime = os.stat(self.filename)[stat.ST_MTIME]
mtime = struct.pack('i', mtime)
return magic + mtime
if __name__ == "__main__":
if len(sys.argv) > 1:
filename = sys.argv[1]
else:
filename = 'test.py'
buf = open(filename).read()
mod = CompiledModule(buf, filename)
mod.compile()
mod.dump(filename + 'c')