generate-wire.py

#! /usr/bin/env python3
# Script to parse spec output CSVs and produce C files.
# Released by lisa neigut under CC0:
# https://creativecommons.org/publicdomain/zero/1.0/
#
# Reads from stdin, outputs C header or body file.
#
# Standard message types:
#   msgtype,<msgname>,<value>[,<option>]
#   msgdata,<msgname>,<fieldname>,<typename>,[<count>][,<option>]
#
# TLV types:
#   tlvtype,<tlvstreamname>,<tlvname>,<value>[,<option>]
#   tlvdata,<tlvstreamname>,<tlvname>,<fieldname>,<typename>,[<count>][,<option>]
#
# Subtypes:
#   subtype,<subtypename>
#   subtypedata,<subtypename>,<fieldname>,<typename>,[<count>]
#
# Note: <count> can be a fixed value, a named value read before,
#       or '...' to read until the end of the current structure.

from argparse import ArgumentParser, REMAINDER
from collections import OrderedDict
import copy
import fileinput
from mako.template import Template
import os
import re
import sys


# Generator to give us one line at a time.
def next_line(args, lines):
    if lines is None:
        lines = fileinput.input(args)

    for i, line in enumerate(lines):
        yield i + 1, line.strip()


# Class definitions, to keep things classy
class Field(object):
    def __init__(self, name, type_obj, extensions=[],
                 field_comments=[], optional=False):
        self.name = name
        self.type_obj = type_obj
        self.count = 1
        self.len_field_of = None
        self.len_field = None
        self.implicit_len = False

        self.extension_names = extensions
        self.is_optional = optional
        self.field_comments = field_comments

    def __deepcopy__(self, memo):
        deepcopy_method = self.__deepcopy__
        self.__deepcopy__ = None
        field = copy.deepcopy(self, memo)
        self.__deepcopy__ = deepcopy_method

        field.type_obj = self.type_obj
        return field

    def add_count(self, count):
        self.count = int(count)

    def add_len_field(self, len_field):
        self.count = False
        # we cache our len-field's name
        self.len_field = len_field.name
        # the len-field caches our name
        len_field.len_field_of = self.name

    def add_implicit_len(self):
        self.count = False
        self.implicit_len = True

    def is_array(self):
        return self.count > 1

    def is_varlen(self):
        return not self.count

    def is_implicit_len(self):
        return self.implicit_len

    def is_extension(self):
        return bool(self.extension_names)

    def size(self, implicit_expression=None):
        if self.count:
            return self.count
        if self.len_field:
            return self.len_field
        assert self.is_implicit_len()
        assert implicit_expression
        return implicit_expression

    def needs_context(self):
        """ A field needs a context if it's varsized """
        return self.is_varlen() or self.type_obj.needs_context()

    def arg_desc_to(self):
        if self.len_field_of:
            return ''
        type_name = self.type_obj.type_name()
        if self.is_array():
            return ', const {} {}[{}]'.format(type_name, self.name, self.count)
        if self.type_obj.is_assignable() and not self.is_varlen():
            name = self.name
            if self.is_optional:
                name = '*' + name
            return ', {} {}'.format(type_name, name)
        if self.is_varlen() and self.type_obj.is_varsize():
            return ', const {} **{}'.format(type_name, self.name)
        return ', const {} *{}'.format(type_name, self.name)

    def arg_desc_from(self):
        type_name = self.type_obj.type_name()
        if self.type_obj.is_const_ptr_ptr_type():
            return ', const {} **{}'.format(type_name, self.name)

        if self.len_field_of:
            return ''
        if self.is_array():
            return ', {} {}[{}]'.format(type_name, self.name, self.count)
        ptrs = '*'
        if self.is_varlen() or self.is_optional or self.type_obj.is_varsize():
            ptrs += '*'
        if self.is_varlen() and self.type_obj.is_varsize():
            ptrs += '*'
        return ', {} {}{}'.format(type_name, ptrs, self.name)


class FieldSet(object):
    def __init__(self):
        self.fields = OrderedDict()
        self.len_fields = {}

    def add_data_field(self, field_name, type_obj, count=1,
                       extensions=[], comments=[], optional=False,
                       implicit_len_ok=False):
        field = Field(field_name, type_obj, extensions=extensions,
                      field_comments=comments, optional=optional)
        if bool(count):
            try:
                field.add_count(int(count))
            except ValueError:
                if count in self.fields:
                    len_field = self.find_data_field(count)
                    field.add_len_field(len_field)
                    self.len_fields[len_field.name] = len_field
                else:
                    # '...' means "rest of TLV"
                    assert implicit_len_ok
                    assert count == '...'
                    field.add_implicit_len()

        # You can't have any fields after an implicit-length field.
        if len(self.fields) != 0:
            assert not self.fields[next(reversed(self.fields))].is_implicit_len()
        self.fields[field_name] = field

    def find_data_field(self, field_name):
        return self.fields[field_name]

    def get_len_fields(self):
        return list(self.len_fields.values())

    def has_len_fields(self):
        return bool(self.len_fields)

    def needs_context(self):
        return any([field.needs_context() or field.is_optional for field in self.fields.values()])

    def singleton(self):
        """Return the single message, if there's only one, otherwise None"""
        if len(self.fields) == 1:
            return next(iter(self.fields.values()))
        return None


class Type(FieldSet):
    assignables = [
        'u8',
        'u16',
        'u32',
        'u64',
        'tu16',
        'tu32',
        'tu64',
        'bool',
        'amount_sat',
        'amount_msat',
        'errcode_t',
        'bigsize',
        'varint'
    ]

    typedefs = [
        'u8',
        'u16',
        'u32',
        'u64',
        'bool',
        'secp256k1_ecdsa_signature',
        'secp256k1_ecdsa_recoverable_signature',
        'utf8',
        'wirestring',
        'errcode_t',
        'bigsize',
        'varint',
    ]

    truncated_typedefs = [
        'tu16',
        'tu32',
        'tu64',
    ]

    # Externally defined variable size types (require a context)
    varsize_types = [
        'peer_features',
        'channel_type',
        'gossip_getnodes_entry',
        'gossip_getchannels_entry',
        'failed_htlc',
        'existing_htlc',
        'utxo',
        'bitcoin_tx',
        'wirestring',
        'per_peer_state',
        'bitcoin_tx_output',
        'exclude_entry',
        'fee_states',
        'height_states',
        'onionreply',
        'feature_set',
        'onionmsg_path',
        'route_hop',
        'tx_parts',
        'wally_psbt',
        'wally_tx',
        'channel_type',
    ]

    # Some BOLT types are re-typed based on their field name
    # ('fieldname partial', 'original type', 'outer type'): ('true type', 'collapse array?')
    name_field_map = {
        ('txid', 'sha256'): ('bitcoin_txid', False),
        ('amt', 'u64'): ('amount_msat', False),
        ('msat', 'u64'): ('amount_msat', False),
        ('satoshis', 'u64'): ('amount_sat', False),
        ('node_id', 'pubkey', 'channel_announcement'): ('node_id', False),
        ('node_id', 'pubkey', 'node_announcement'): ('node_id', False),
        ('temporary_channel_id', 'u8'): ('channel_id', True),
        ('secret', 'u8'): ('secret', True),
        ('preimage', 'u8'): ('preimage', True),
    }

    # For BOLT specified types, a few type names need to be simply 'remapped'
    # 'original type': 'true type'
    name_remap = {
        'byte': 'u8',
        'signature': 'secp256k1_ecdsa_signature',
        'chain_hash': 'bitcoin_blkid',
        'point': 'pubkey',
        # FIXME: omits 'pad'
    }

    # Types that are const pointer-to-pointers, such as chainparams, i.e.,
    # they set a reference to some const entry.
    const_ptr_ptr_types = [
        'chainparams'
    ]

    @staticmethod
    def true_type(type_name, field_name=None, outer_name=None):
        """ Returns 'true' type of a given type and a flag if
            we've remapped a variable size/array type to a single struct
            (an example of this is 'temporary_channel_id' which is specified
            as a 32*byte, but we re-map it to a channel_id
        """
        if type_name in Type.name_remap:
            type_name = Type.name_remap[type_name]

        if field_name:
            for t, true_type in Type.name_field_map.items():
                if t[0] in field_name and t[1] == type_name:
                    if len(t) == 2 or outer_name == t[2]:
                        return true_type
        return (type_name, False)

    def __init__(self, name):
        FieldSet.__init__(self)
        self.name, self.is_enum = self.parse_name(name)
        self.depends_on = {}
        self.type_comments = []
        self.tlv = False

    def parse_name(self, name):
        if name.startswith('enum '):
            return name[5:], True
        return name, False

    def add_data_field(self, field_name, type_obj, count=1,
                       extensions=[], comments=[], optional=False):
        FieldSet.add_data_field(self, field_name, type_obj, count,
                                extensions=extensions,
                                comments=comments, optional=optional)
        if type_obj.name not in self.depends_on:
            self.depends_on[type_obj.name] = type_obj

    def type_name(self):
        if self.name in self.typedefs:
            return self.name
        if self.name in self.truncated_typedefs:
            return self.name[1:]
        if self.is_enum:
            prefix = 'enum '
        else:
            prefix = 'struct '

        return prefix + self.struct_name()

    # We accelerate the u8 case: it's common and trivial.
    # We handle the utf8 case so we can be sure it's actually a UTF-8 string.
    def has_array_helper(self):
        return self.name in ['u8', 'utf8']

    def struct_name(self):
        if self.is_tlv():
            return self.tlv.struct_name()
        return self.name

    def subtype_deps(self):
        return [dep for dep in self.depends_on.values() if dep.is_subtype()]

    def is_subtype(self):
        return bool(self.fields)

    def is_const_ptr_ptr_type(self):
        return self.name in self.const_ptr_ptr_types

    def is_truncated(self):
        return self.name in self.truncated_typedefs

    def needs_context(self):
        return self.is_varsize()

    def is_assignable(self):
        """ Generally typedef's and enums """
        return self.name in self.assignables or self.is_enum

    def is_varsize(self):
        """ A type is variably sized if it's marked as such (in varsize_types)
            or it contains a field of variable length """
        return self.name in self.varsize_types or self.has_len_fields()

    def add_comments(self, comments):
        self.type_comments = comments

    def mark_tlv(self, tlv):
        self.tlv = tlv

    def is_tlv(self):
        return bool(self.tlv)


class Message(FieldSet):
    def __init__(self, name, number, option=[], enum_prefix='wire',
                 struct_prefix=None, comments=[]):
        FieldSet.__init__(self)
        self.name = name
        self.number = number
        self.enum_prefix = enum_prefix
        self.option = option[0] if len(option) else None
        self.struct_prefix = struct_prefix
        self.enumname = None
        self.msg_comments = comments
        self.if_token = None

    def has_option(self):
        return self.option is not None

    def enum_name(self):
        name = self.enumname if self.enumname else self.name
        return "{}_{}".format(self.enum_prefix, name).upper()

    def struct_name(self):
        if self.struct_prefix:
            return self.struct_prefix + "_" + self.name
        return self.name

    def add_if(self, if_token):
        self.if_token = if_token


class Tlv(object):
    def __init__(self, name):
        self.name = name
        self.messages = {}

    def add_message(self, tokens, comments=[]):
        """ tokens -> (name, value[, option]) """
        self.messages[tokens[0]] = Message(tokens[0], tokens[1], option=tokens[2:],
                                           enum_prefix=self.name,
                                           struct_prefix=self.struct_name(),
                                           comments=comments)

    def type_name(self):
        return 'struct ' + self.struct_name()

    def struct_name(self):
        return "tlv_{}".format(self.name)

    def find_message(self, name):
        return self.messages[name]

    def ordered_msgs(self):
        return sorted(self.messages.values(), key=lambda item: int(item.number))


class Master(object):
    types = {}
    tlvs = {}
    messages = {}
    extension_msgs = {}
    inclusions = []
    top_comments = []

    def add_comments(self, comments):
        self.top_comments += comments

    def add_include(self, inclusion):
        self.inclusions.append(inclusion)

    def add_tlv(self, tlv_name):
        if tlv_name not in self.tlvs:
            self.tlvs[tlv_name] = Tlv(tlv_name)

        if tlv_name not in self.types:
            self.types[tlv_name] = Type(tlv_name)

        return self.tlvs[tlv_name]

    def add_message(self, tokens, comments=[]):
        """ tokens -> (name, value[, option])"""
        self.messages[tokens[0]] = Message(tokens[0], tokens[1], option=tokens[2:],
                                           comments=comments)

    def add_extension_msg(self, name, msg):
        self.extension_msgs[name] = msg

    def add_type(self, type_name, field_name=None, outer_name=None):
        optional = False
        if type_name.startswith('?'):
            type_name = type_name[1:]
            optional = True
        # Check for special type name re-mapping
        type_name, collapse_original = Type.true_type(type_name, field_name,
                                                      outer_name)

        if type_name not in self.types:
            self.types[type_name] = Type(type_name)
        return self.types[type_name], collapse_original, optional

    def find_type(self, type_name):
        return self.types[type_name]

    def find_message(self, msg_name):
        if msg_name in self.messages:
            return self.messages[msg_name]
        if msg_name in self.extension_msgs:
            return self.extension_msgs[msg_name]
        return None

    def find_tlv(self, tlv_name):
        return self.tlvs[tlv_name]

    def get_ordered_subtypes(self):
        """ We want to order subtypes such that the 'no dependency'
        types are printed first """
        subtypes = [s for s in self.types.values() if s.is_subtype()]

        # Start with subtypes without subtype dependencies
        sorted_types = [s for s in subtypes if not len(s.subtype_deps())]
        unsorted = [s for s in subtypes if len(s.subtype_deps())]
        while len(unsorted):
            names = [s.name for s in sorted_types]
            for s in list(unsorted):
                if all([dependency.name in names for dependency in s.subtype_deps()]):
                    sorted_types.append(s)
                    unsorted.remove(s)
        return sorted_types

    def tlv_structs(self):
        ret = []
        for tlv in self.tlvs.values():
            for v in tlv.messages.values():
                if not v.singleton():
                    ret.append(v)

        return ret

    def find_template(self, options):
        dirpath = os.path.dirname(os.path.abspath(__file__))
        filename = dirpath + '/gen/{}{}_template'.format(
            'print_' if options.print_wire else '', options.page)

        return Template(filename=filename)

    def post_process(self):
        """ method to handle any 'post processing' that needs to be done.
            for now, we just need match up types to TLVs """
        for tlv_name, tlv in self.tlvs.items():
            if tlv_name in self.types:
                self.types[tlv_name].mark_tlv(tlv)

    def write(self, options, output):
        template = self.find_template(options)
        enum_sets = []
        if len(self.messages.values()) != 0:
            enum_sets.append({
                'name': options.enum_name,
                'set': self.messages.values(),
            })
        stuff = {}
        stuff['top_comments'] = self.top_comments
        stuff['options'] = options
        stuff['idem'] = re.sub(r'[^A-Z0-9]+', '_', options.header_filename.upper())
        stuff['header_filename'] = options.header_filename
        stuff['includes'] = self.inclusions
        stuff['enum_sets'] = enum_sets
        subtypes = self.get_ordered_subtypes()
        stuff['structs'] = subtypes + self.tlv_structs()
        stuff['tlvs'] = self.tlvs

        # We leave out extension messages in the printing pages. Any extension
        # fields will get printed under the 'original' message, if present
        if options.print_wire:
            stuff['messages'] = list(self.messages.values())
        else:
            stuff['messages'] = list(self.messages.values()) + list(self.extension_msgs.values())
        stuff['subtypes'] = subtypes

        for line in template.render(**stuff).splitlines():
            print(line.rstrip(), file=output)


def main(options, args=None, output=sys.stdout, lines=None):
    genline = next_line(args, lines)

    comment_set = []
    token_name = None

    # Create a new 'master' that serves as the coordinator for the file generation
    master = Master()
    for i in options.include:
        master.add_include('#include <{}>'.format(i))

    try:
        while True:
            ln, line = next(genline)
            tokens = line.split(',')
            token_type = tokens[0]

            if not bool(line):
                master.add_comments(comment_set)
                comment_set = []
                token_name = None
                continue

            if len(tokens) > 2:
                token_name = tokens[1]

            if token_type == 'subtype':
                subtype, _, _ = master.add_type(tokens[1])

                subtype.add_comments(list(comment_set))
                comment_set = []
            elif token_type == 'subtypedata':
                subtype = master.find_type(tokens[1])
                if not subtype:
                    raise ValueError('Unknown subtype {} for data.\nat {}:{}'
                                     .format(tokens[1], ln, line))
                type_obj, collapse, optional = master.add_type(tokens[3], tokens[2], tokens[1])
                if optional:
                    raise ValueError('Subtypes cannot have optional fields {}.{}\n at {}:{}'
                                     .format(subtype.name, tokens[2], ln, line))
                if collapse:
                    count = 1
                else:
                    count = tokens[4]

                subtype.add_data_field(tokens[2], type_obj, count, comments=list(comment_set),
                                       optional=optional)
                comment_set = []
            elif token_type == 'tlvtype':
                tlv = master.add_tlv(tokens[1])
                tlv.add_message(tokens[2:], comments=list(comment_set))

                comment_set = []
            elif token_type == 'tlvdata':
                type_obj, collapse, optional = master.add_type(tokens[4], tokens[3], tokens[1])
                if optional:
                    raise ValueError('TLV messages cannot have optional fields {}.{}\n at {}:{}'
                                     .format(tokens[2], tokens[3], ln, line))

                tlv = master.find_tlv(tokens[1])
                if not tlv:
                    raise ValueError('tlvdata for unknown tlv {}.\nat {}:{}'
                                     .format(tokens[1], ln, line))
                msg = tlv.find_message(tokens[2])
                if not msg:
                    raise ValueError('tlvdata for unknown tlv-message {}.\nat {}:{}'
                                     .format(tokens[2], ln, line))
                if collapse:
                    count = 1
                else:
                    count = tokens[5]

                msg.add_data_field(tokens[3], type_obj, count, comments=list(comment_set),
                                   optional=optional, implicit_len_ok=True)
                comment_set = []
            elif token_type == 'msgtype':
                master.add_message(tokens[1:], comments=list(comment_set))
                comment_set = []
            elif token_type == 'msgdata':
                msg = master.find_message(tokens[1])
                if not msg:
                    raise ValueError('Unknown message type {}. {}:{}'.format(tokens[1], ln, line))
                type_obj, collapse, optional = master.add_type(tokens[3], tokens[2], tokens[1])

                if collapse:
                    count = 1
                elif len(tokens) < 5:
                    raise ValueError('problem with parsing {}:{}'.format(ln, line))
                else:
                    count = tokens[4]

                # if this is an 'extension' field*, we want to add a new 'message' type
                # in the future, extensions will be handled as TLV's
                #
                # *(in the spec they're called 'optional', but that term is overloaded
                #   in that internal wire messages have 'optional' fields that are treated
                #   differently. for the sake of clarity here, for bolt-wire messages,
                #   we'll refer to 'optional' message fields as 'extensions')
                #
                if tokens[5:] == []:
                    msg.add_data_field(tokens[2], type_obj, count, comments=list(comment_set),
                                       optional=optional)
                else:  # is one or more extension fields
                    if optional:
                        raise ValueError("Extension fields cannot be optional. {}:{}"
                                         .format(ln, line))
                    orig_msg = msg
                    for extension in tokens[5:]:
                        extension_name = "{}_{}".format(tokens[1], extension)
                        msg = master.find_message(extension_name)
                        if not msg:
                            msg = copy.deepcopy(orig_msg)
                            msg.enumname = msg.name
                            msg.name = extension_name
                            master.add_extension_msg(msg.name, msg)
                        msg.add_data_field(tokens[2], type_obj, count, comments=list(comment_set), optional=optional)

                    # If this is a print_wire page, add the extension fields to the
                    # original message, so we can print them if present.
                    if options.print_wire:
                        orig_msg.add_data_field(tokens[2], type_obj, count=count,
                                                extensions=tokens[5:],
                                                comments=list(comment_set),
                                                optional=optional)

                comment_set = []
            elif token_type.startswith('#include'):
                master.add_include(token_type)
            elif token_type.startswith('#if'):
                msg = master.find_message(token_name)
                if (msg):
                    if_token = token_type[token_type.index(' ') + 1:]
                    msg.add_if(if_token)
            elif token_type.startswith('#'):
                comment_set.append(token_type[1:])
            else:
                raise ValueError("Unknown token type {} on line {}:{}".format(token_type, ln, line))

    except StopIteration:
        pass

    master.post_process()
    master.write(options, output)


if __name__ == "__main__":
    parser = ArgumentParser()
    parser.add_argument("-s", "--expose-subtypes", help="print subtypes in header",
                        action="store_true", default=False)
    parser.add_argument("-P", "--print_wire", help="generate wire printing source files",
                        action="store_true", default=False)
    parser.add_argument("--page", choices=['header', 'impl'], help="page to print")
    parser.add_argument('--expose-tlv-type', action='append', default=[])
    parser.add_argument('--include', action='append', default=[])
    parser.add_argument('header_filename', help='The filename of the header')
    parser.add_argument('enum_name', help='The name of the enum to produce')
    parser.add_argument("files", help='Files to read in (or stdin)', nargs=REMAINDER)
    parsed_args = parser.parse_args()

    main(parsed_args, parsed_args.files)