423 lines
13 KiB
Python
423 lines
13 KiB
Python
from itertools import chain
|
|
from binascii import hexlify
|
|
from collections import namedtuple
|
|
from typing import List
|
|
|
|
from torba.client.bcd_data_stream import BCDataStream
|
|
from torba.client.util import subclass_tuple
|
|
|
|
# bitcoin opcodes
|
|
OP_0 = 0x00
|
|
OP_1 = 0x51
|
|
OP_16 = 0x60
|
|
OP_VERIFY = 0x69
|
|
OP_DUP = 0x76
|
|
OP_HASH160 = 0xa9
|
|
OP_EQUALVERIFY = 0x88
|
|
OP_CHECKSIG = 0xac
|
|
OP_CHECKMULTISIG = 0xae
|
|
OP_EQUAL = 0x87
|
|
OP_PUSHDATA1 = 0x4c
|
|
OP_PUSHDATA2 = 0x4d
|
|
OP_PUSHDATA4 = 0x4e
|
|
OP_RETURN = 0x6a
|
|
OP_2DROP = 0x6d
|
|
OP_DROP = 0x75
|
|
|
|
|
|
# template matching opcodes (not real opcodes)
|
|
# base class for PUSH_DATA related opcodes
|
|
# pylint: disable=invalid-name
|
|
PUSH_DATA_OP = namedtuple('PUSH_DATA_OP', 'name')
|
|
# opcode for variable length strings
|
|
# pylint: disable=invalid-name
|
|
PUSH_SINGLE = subclass_tuple('PUSH_SINGLE', PUSH_DATA_OP)
|
|
# opcode for variable size integers
|
|
# pylint: disable=invalid-name
|
|
PUSH_INTEGER = subclass_tuple('PUSH_INTEGER', PUSH_DATA_OP)
|
|
# opcode for variable number of variable length strings
|
|
# pylint: disable=invalid-name
|
|
PUSH_MANY = subclass_tuple('PUSH_MANY', PUSH_DATA_OP)
|
|
# opcode with embedded subscript parsing
|
|
# pylint: disable=invalid-name
|
|
PUSH_SUBSCRIPT = namedtuple('PUSH_SUBSCRIPT', 'name template')
|
|
|
|
|
|
def is_push_data_opcode(opcode):
|
|
return isinstance(opcode, (PUSH_DATA_OP, PUSH_SUBSCRIPT))
|
|
|
|
|
|
def is_push_data_token(token):
|
|
return 1 <= token <= OP_PUSHDATA4
|
|
|
|
|
|
def push_data(data):
|
|
size = len(data)
|
|
if size < OP_PUSHDATA1:
|
|
yield BCDataStream.uint8.pack(size)
|
|
elif size <= 0xFF:
|
|
yield BCDataStream.uint8.pack(OP_PUSHDATA1)
|
|
yield BCDataStream.uint8.pack(size)
|
|
elif size <= 0xFFFF:
|
|
yield BCDataStream.uint8.pack(OP_PUSHDATA2)
|
|
yield BCDataStream.uint16.pack(size)
|
|
else:
|
|
yield BCDataStream.uint8.pack(OP_PUSHDATA4)
|
|
yield BCDataStream.uint32.pack(size)
|
|
yield data
|
|
|
|
|
|
def read_data(token, stream):
|
|
if token < OP_PUSHDATA1:
|
|
return stream.read(token)
|
|
if token == OP_PUSHDATA1:
|
|
return stream.read(stream.read_uint8())
|
|
if token == OP_PUSHDATA2:
|
|
return stream.read(stream.read_uint16())
|
|
return stream.read(stream.read_uint32())
|
|
|
|
|
|
# opcode for OP_1 - OP_16
|
|
# pylint: disable=invalid-name
|
|
SMALL_INTEGER = namedtuple('SMALL_INTEGER', 'name')
|
|
|
|
|
|
def is_small_integer(token):
|
|
return OP_1 <= token <= OP_16
|
|
|
|
|
|
def push_small_integer(num):
|
|
assert 1 <= num <= 16
|
|
yield BCDataStream.uint8.pack(OP_1 + (num - 1))
|
|
|
|
|
|
def read_small_integer(token):
|
|
return (token - OP_1) + 1
|
|
|
|
|
|
class Token(namedtuple('Token', 'value')):
|
|
__slots__ = ()
|
|
|
|
def __repr__(self):
|
|
name = None
|
|
for var_name, var_value in globals().items():
|
|
if var_name.startswith('OP_') and var_value == self.value:
|
|
name = var_name
|
|
break
|
|
return name or self.value
|
|
|
|
|
|
class DataToken(Token):
|
|
__slots__ = ()
|
|
|
|
def __repr__(self):
|
|
return '"{}"'.format(hexlify(self.value))
|
|
|
|
|
|
class SmallIntegerToken(Token):
|
|
__slots__ = ()
|
|
|
|
def __repr__(self):
|
|
return 'SmallIntegerToken({})'.format(self.value)
|
|
|
|
|
|
def token_producer(source):
|
|
token = source.read_uint8()
|
|
while token is not None:
|
|
if is_push_data_token(token):
|
|
yield DataToken(read_data(token, source))
|
|
elif is_small_integer(token):
|
|
yield SmallIntegerToken(read_small_integer(token))
|
|
else:
|
|
yield Token(token)
|
|
token = source.read_uint8()
|
|
|
|
|
|
def tokenize(source):
|
|
return list(token_producer(source))
|
|
|
|
|
|
class ScriptError(Exception):
|
|
""" General script handling error. """
|
|
|
|
|
|
class ParseError(ScriptError):
|
|
""" Script parsing error. """
|
|
|
|
|
|
class Parser:
|
|
|
|
def __init__(self, opcodes, tokens):
|
|
self.opcodes = opcodes
|
|
self.tokens = tokens
|
|
self.values = {}
|
|
self.token_index = 0
|
|
self.opcode_index = 0
|
|
|
|
def parse(self):
|
|
while self.token_index < len(self.tokens) and self.opcode_index < len(self.opcodes):
|
|
token = self.tokens[self.token_index]
|
|
opcode = self.opcodes[self.opcode_index]
|
|
if isinstance(token, DataToken):
|
|
if isinstance(opcode, (PUSH_SINGLE, PUSH_INTEGER, PUSH_SUBSCRIPT)):
|
|
self.push_single(opcode, token.value)
|
|
elif isinstance(opcode, PUSH_MANY):
|
|
self.consume_many_non_greedy()
|
|
else:
|
|
raise ParseError("DataToken found but opcode was '{}'.".format(opcode))
|
|
elif isinstance(token, SmallIntegerToken):
|
|
if isinstance(opcode, SMALL_INTEGER):
|
|
self.values[opcode.name] = token.value
|
|
else:
|
|
raise ParseError("SmallIntegerToken found but opcode was '{}'.".format(opcode))
|
|
elif token.value == opcode:
|
|
pass
|
|
else:
|
|
raise ParseError("Token is '{}' and opcode is '{}'.".format(token.value, opcode))
|
|
self.token_index += 1
|
|
self.opcode_index += 1
|
|
|
|
if self.token_index < len(self.tokens):
|
|
raise ParseError("Parse completed without all tokens being consumed.")
|
|
|
|
if self.opcode_index < len(self.opcodes):
|
|
raise ParseError("Parse completed without all opcodes being consumed.")
|
|
|
|
return self
|
|
|
|
def consume_many_non_greedy(self):
|
|
""" Allows PUSH_MANY to consume data without being greedy
|
|
in cases when one or more PUSH_SINGLEs follow a PUSH_MANY. This will
|
|
prioritize giving all PUSH_SINGLEs some data and only after that
|
|
subsume the rest into PUSH_MANY.
|
|
"""
|
|
|
|
token_values = []
|
|
while self.token_index < len(self.tokens):
|
|
token = self.tokens[self.token_index]
|
|
if not isinstance(token, DataToken):
|
|
self.token_index -= 1
|
|
break
|
|
token_values.append(token.value)
|
|
self.token_index += 1
|
|
|
|
push_opcodes = []
|
|
push_many_count = 0
|
|
while self.opcode_index < len(self.opcodes):
|
|
opcode = self.opcodes[self.opcode_index]
|
|
if not is_push_data_opcode(opcode):
|
|
self.opcode_index -= 1
|
|
break
|
|
if isinstance(opcode, PUSH_MANY):
|
|
push_many_count += 1
|
|
push_opcodes.append(opcode)
|
|
self.opcode_index += 1
|
|
|
|
if push_many_count > 1:
|
|
raise ParseError(
|
|
"Cannot have more than one consecutive PUSH_MANY, as there is no way to tell which"
|
|
" token value should go into which PUSH_MANY."
|
|
)
|
|
|
|
if len(push_opcodes) > len(token_values):
|
|
raise ParseError(
|
|
"Not enough token values to match all of the PUSH_MANY and PUSH_SINGLE opcodes."
|
|
)
|
|
|
|
many_opcode = push_opcodes.pop(0)
|
|
|
|
# consume data into PUSH_SINGLE opcodes, working backwards
|
|
for opcode in reversed(push_opcodes):
|
|
self.push_single(opcode, token_values.pop())
|
|
|
|
# finally PUSH_MANY gets everything that's left
|
|
self.values[many_opcode.name] = token_values
|
|
|
|
def push_single(self, opcode, value):
|
|
if isinstance(opcode, PUSH_SINGLE):
|
|
self.values[opcode.name] = value
|
|
elif isinstance(opcode, PUSH_INTEGER):
|
|
self.values[opcode.name] = int.from_bytes(value, 'little')
|
|
elif isinstance(opcode, PUSH_SUBSCRIPT):
|
|
self.values[opcode.name] = Script.from_source_with_template(value, opcode.template)
|
|
else:
|
|
raise ParseError("Not a push single or subscript: {}".format(opcode))
|
|
|
|
|
|
class Template:
|
|
|
|
__slots__ = 'name', 'opcodes'
|
|
|
|
def __init__(self, name, opcodes):
|
|
self.name = name
|
|
self.opcodes = opcodes
|
|
|
|
def parse(self, tokens):
|
|
return Parser(self.opcodes, tokens).parse().values
|
|
|
|
def generate(self, values):
|
|
source = BCDataStream()
|
|
for opcode in self.opcodes:
|
|
if isinstance(opcode, PUSH_SINGLE):
|
|
data = values[opcode.name]
|
|
source.write_many(push_data(data))
|
|
elif isinstance(opcode, PUSH_INTEGER):
|
|
data = values[opcode.name]
|
|
source.write_many(push_data(
|
|
data.to_bytes((data.bit_length() + 7) // 8, byteorder='little')
|
|
))
|
|
elif isinstance(opcode, PUSH_SUBSCRIPT):
|
|
data = values[opcode.name]
|
|
source.write_many(push_data(data.source))
|
|
elif isinstance(opcode, PUSH_MANY):
|
|
for data in values[opcode.name]:
|
|
source.write_many(push_data(data))
|
|
elif isinstance(opcode, SMALL_INTEGER):
|
|
data = values[opcode.name]
|
|
source.write_many(push_small_integer(data))
|
|
else:
|
|
source.write_uint8(opcode)
|
|
return source.get_bytes()
|
|
|
|
|
|
class Script:
|
|
|
|
__slots__ = 'source', 'template', 'values'
|
|
|
|
templates: List[Template] = []
|
|
|
|
def __init__(self, source=None, template=None, values=None, template_hint=None):
|
|
self.source = source
|
|
self.template = template
|
|
self.values = values
|
|
if source:
|
|
self.parse(template_hint)
|
|
elif template and values:
|
|
self.generate()
|
|
|
|
@property
|
|
def tokens(self):
|
|
return tokenize(BCDataStream(self.source))
|
|
|
|
@classmethod
|
|
def from_source_with_template(cls, source, template):
|
|
return cls(source, template_hint=template)
|
|
|
|
def parse(self, template_hint=None):
|
|
tokens = self.tokens
|
|
for template in chain((template_hint,), self.templates):
|
|
if not template:
|
|
continue
|
|
try:
|
|
self.values = template.parse(tokens)
|
|
self.template = template
|
|
return
|
|
except ParseError:
|
|
continue
|
|
raise ValueError('No matching templates for source: {}'.format(hexlify(self.source)))
|
|
|
|
def generate(self):
|
|
self.source = self.template.generate(self.values)
|
|
|
|
|
|
class BaseInputScript(Script):
|
|
""" Input / redeem script templates (aka scriptSig) """
|
|
|
|
__slots__ = ()
|
|
|
|
REDEEM_PUBKEY = Template('pubkey', (
|
|
PUSH_SINGLE('signature'),
|
|
))
|
|
REDEEM_PUBKEY_HASH = Template('pubkey_hash', (
|
|
PUSH_SINGLE('signature'), PUSH_SINGLE('pubkey')
|
|
))
|
|
REDEEM_SCRIPT = Template('script', (
|
|
SMALL_INTEGER('signatures_count'), PUSH_MANY('pubkeys'), SMALL_INTEGER('pubkeys_count'),
|
|
OP_CHECKMULTISIG
|
|
))
|
|
REDEEM_SCRIPT_HASH = Template('script_hash', (
|
|
OP_0, PUSH_MANY('signatures'), PUSH_SUBSCRIPT('script', REDEEM_SCRIPT)
|
|
))
|
|
|
|
templates = [
|
|
REDEEM_PUBKEY,
|
|
REDEEM_PUBKEY_HASH,
|
|
REDEEM_SCRIPT_HASH,
|
|
REDEEM_SCRIPT
|
|
]
|
|
|
|
@classmethod
|
|
def redeem_pubkey_hash(cls, signature, pubkey):
|
|
return cls(template=cls.REDEEM_PUBKEY_HASH, values={
|
|
'signature': signature,
|
|
'pubkey': pubkey
|
|
})
|
|
|
|
@classmethod
|
|
def redeem_script_hash(cls, signatures, pubkeys):
|
|
return cls(template=cls.REDEEM_SCRIPT_HASH, values={
|
|
'signatures': signatures,
|
|
'script': cls.redeem_script(signatures, pubkeys)
|
|
})
|
|
|
|
@classmethod
|
|
def redeem_script(cls, signatures, pubkeys):
|
|
return cls(template=cls.REDEEM_SCRIPT, values={
|
|
'signatures_count': len(signatures),
|
|
'pubkeys': pubkeys,
|
|
'pubkeys_count': len(pubkeys)
|
|
})
|
|
|
|
|
|
class BaseOutputScript(Script):
|
|
|
|
__slots__ = ()
|
|
|
|
# output / payment script templates (aka scriptPubKey)
|
|
PAY_PUBKEY_FULL = Template('pay_pubkey_full', (
|
|
PUSH_SINGLE('pubkey'), OP_CHECKSIG
|
|
))
|
|
PAY_PUBKEY_HASH = Template('pay_pubkey_hash', (
|
|
OP_DUP, OP_HASH160, PUSH_SINGLE('pubkey_hash'), OP_EQUALVERIFY, OP_CHECKSIG
|
|
))
|
|
PAY_SCRIPT_HASH = Template('pay_script_hash', (
|
|
OP_HASH160, PUSH_SINGLE('script_hash'), OP_EQUAL
|
|
))
|
|
RETURN_DATA = Template('return_data', (
|
|
OP_RETURN, PUSH_SINGLE('data')
|
|
))
|
|
|
|
templates = [
|
|
PAY_PUBKEY_FULL,
|
|
PAY_PUBKEY_HASH,
|
|
PAY_SCRIPT_HASH,
|
|
RETURN_DATA
|
|
]
|
|
|
|
@classmethod
|
|
def pay_pubkey_hash(cls, pubkey_hash):
|
|
return cls(template=cls.PAY_PUBKEY_HASH, values={
|
|
'pubkey_hash': pubkey_hash
|
|
})
|
|
|
|
@classmethod
|
|
def pay_script_hash(cls, script_hash):
|
|
return cls(template=cls.PAY_SCRIPT_HASH, values={
|
|
'script_hash': script_hash
|
|
})
|
|
|
|
@property
|
|
def is_pay_pubkey(self):
|
|
return self.template.name.endswith('pay_pubkey_full')
|
|
|
|
@property
|
|
def is_pay_pubkey_hash(self):
|
|
return self.template.name.endswith('pay_pubkey_hash')
|
|
|
|
@property
|
|
def is_pay_script_hash(self):
|
|
return self.template.name.endswith('pay_script_hash')
|
|
|
|
@property
|
|
def is_return_data(self):
|
|
return self.template.name.endswith('return_data')
|