mirror of
https://github.com/ElementsProject/lightning.git
synced 2024-11-19 09:54:16 +01:00
232b330284
I verified that the generated files don't change. Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
783 lines
27 KiB
Python
Executable File
783 lines
27 KiB
Python
Executable File
#! /usr/bin/env python3
|
|
# Read from stdin, spit out C header or body.
|
|
|
|
import argparse
|
|
import copy
|
|
import fileinput
|
|
import re
|
|
|
|
from collections import namedtuple
|
|
|
|
Enumtype = namedtuple('Enumtype', ['name', 'value'])
|
|
|
|
type2size = {
|
|
'pad': 1,
|
|
'struct channel_id': 32,
|
|
'struct short_channel_id': 8,
|
|
'struct ipv6': 16,
|
|
'secp256k1_ecdsa_signature': 64,
|
|
'struct preimage': 32,
|
|
'struct pubkey': 33,
|
|
'struct sha256': 32,
|
|
'struct bitcoin_blkid': 32,
|
|
'struct bitcoin_txid': 32,
|
|
'struct secret': 32,
|
|
'u64': 8,
|
|
'u32': 4,
|
|
'u16': 2,
|
|
'u8': 1,
|
|
'bool': 1
|
|
}
|
|
|
|
# These struct array helpers require a context to allocate from.
|
|
varlen_structs = [
|
|
'gossip_getnodes_entry',
|
|
'failed_htlc',
|
|
'utxo',
|
|
'bitcoin_tx',
|
|
'wirestring',
|
|
]
|
|
|
|
|
|
class FieldType(object):
|
|
def __init__(self, name):
|
|
self.name = name
|
|
|
|
def is_assignable(self):
|
|
return self.name in ['u8', 'u16', 'u32', 'u64', 'bool'] or self.name.startswith('enum ')
|
|
|
|
# We only accelerate the u8 case: it's common and trivial.
|
|
def has_array_helper(self):
|
|
return self.name in ['u8']
|
|
|
|
# Returns base size
|
|
@staticmethod
|
|
def _typesize(typename):
|
|
if typename in type2size:
|
|
return type2size[typename]
|
|
elif typename.startswith('struct ') or typename.startswith('enum '):
|
|
# We allow unknown structures/enums, for extensibility (can only happen
|
|
# if explicitly specified in csv)
|
|
return 0
|
|
else:
|
|
raise ValueError('Unknown typename {}'.format(typename))
|
|
|
|
|
|
# Full (message, fieldname)-mappings
|
|
typemap = {
|
|
('update_fail_htlc', 'reason'): FieldType('u8'),
|
|
('node_announcement', 'alias'): FieldType('u8'),
|
|
('update_add_htlc', 'onion_routing_packet'): FieldType('u8'),
|
|
('update_fulfill_htlc', 'payment_preimage'): FieldType('struct preimage'),
|
|
('error', 'data'): FieldType('u8'),
|
|
('shutdown', 'scriptpubkey'): FieldType('u8'),
|
|
('node_announcement', 'rgb_color'): FieldType('u8'),
|
|
('node_announcement', 'addresses'): FieldType('u8'),
|
|
('node_announcement', 'ipv6'): FieldType('struct ipv6'),
|
|
('announcement_signatures', 'short_channel_id'): FieldType('struct short_channel_id'),
|
|
('channel_announcement', 'short_channel_id'): FieldType('struct short_channel_id'),
|
|
('channel_update', 'short_channel_id'): FieldType('struct short_channel_id'),
|
|
('revoke_and_ack', 'per_commitment_secret'): FieldType('struct secret')
|
|
}
|
|
|
|
# Partial names that map to a datatype
|
|
partialtypemap = {
|
|
'signature': FieldType('secp256k1_ecdsa_signature'),
|
|
'features': FieldType('u8'),
|
|
'channel_id': FieldType('struct channel_id'),
|
|
'chain_hash': FieldType('struct bitcoin_blkid'),
|
|
'funding_txid': FieldType('struct bitcoin_txid'),
|
|
'pad': FieldType('pad'),
|
|
}
|
|
|
|
# Size to typename match
|
|
sizetypemap = {
|
|
33: FieldType('struct pubkey'),
|
|
32: FieldType('struct sha256'),
|
|
8: FieldType('u64'),
|
|
4: FieldType('u32'),
|
|
2: FieldType('u16'),
|
|
1: FieldType('u8')
|
|
}
|
|
|
|
|
|
# It would be nicer if we had put '*u8' in spec and disallowed bare lenvar.
|
|
# In practice we only recognize lenvar when it's the previous field.
|
|
|
|
# size := baresize | arraysize
|
|
# baresize := simplesize | lenvar
|
|
# simplesize := number | type
|
|
# arraysize := length '*' type
|
|
# length := lenvar | number
|
|
class Field(object):
|
|
def __init__(self, message, name, size, comments, prevname):
|
|
self.message = message
|
|
self.comments = comments
|
|
self.name = name
|
|
self.is_len_var = False
|
|
self.lenvar = None
|
|
self.num_elems = 1
|
|
self.optional = False
|
|
|
|
# ? means optional field (not supported for arrays)
|
|
if size.startswith('?'):
|
|
self.optional = True
|
|
size = size[1:]
|
|
# If it's an arraysize, swallow prefix.
|
|
elif '*' in size:
|
|
number = size.split('*')[0]
|
|
if number == prevname:
|
|
self.lenvar = number
|
|
else:
|
|
self.num_elems = int(number)
|
|
size = size.split('*')[1]
|
|
elif options.bolt and size == prevname:
|
|
# Raw length field, implies u8.
|
|
self.lenvar = size
|
|
size = '1'
|
|
|
|
# Bolts use just a number: Guess type based on size.
|
|
if options.bolt:
|
|
base_size = int(size)
|
|
self.fieldtype = Field._guess_type(message, self.name, base_size)
|
|
# There are some arrays which we have to guess, based on sizes.
|
|
tsize = FieldType._typesize(self.fieldtype.name)
|
|
if base_size % tsize != 0:
|
|
raise ValueError('Invalid size {} for {}.{} not a multiple of {}'
|
|
.format(base_size,
|
|
self.message,
|
|
self.name,
|
|
tsize))
|
|
self.num_elems = int(base_size / tsize)
|
|
else:
|
|
# Real typename.
|
|
self.fieldtype = FieldType(size)
|
|
|
|
def basetype(self):
|
|
base = self.fieldtype.name
|
|
if base.startswith('struct '):
|
|
base = base[7:]
|
|
elif base.startswith('enum '):
|
|
base = base[5:]
|
|
return base
|
|
|
|
def is_padding(self):
|
|
return self.name.startswith('pad')
|
|
|
|
# Padding is always treated as an array.
|
|
def is_array(self):
|
|
return self.num_elems > 1 or self.is_padding()
|
|
|
|
def is_variable_size(self):
|
|
return self.lenvar is not None
|
|
|
|
def needs_ptr_to_ptr(self):
|
|
return self.is_variable_size() or self.optional
|
|
|
|
def is_assignable(self):
|
|
if self.is_array() or self.needs_ptr_to_ptr():
|
|
return False
|
|
return self.fieldtype.is_assignable()
|
|
|
|
def has_array_helper(self):
|
|
return self.fieldtype.has_array_helper()
|
|
|
|
# Returns FieldType
|
|
@staticmethod
|
|
def _guess_type(message, fieldname, base_size):
|
|
# Check for full (message, fieldname)-matches
|
|
if (message, fieldname) in typemap:
|
|
return typemap[(message, fieldname)]
|
|
|
|
# Check for partial field names
|
|
for k, v in partialtypemap.items():
|
|
if k in fieldname:
|
|
return v
|
|
|
|
# Check for size matches
|
|
if base_size in sizetypemap:
|
|
return sizetypemap[base_size]
|
|
|
|
raise ValueError('Unknown size {} for {}'.format(base_size, fieldname))
|
|
|
|
|
|
fromwire_impl_templ = """bool fromwire_{name}({ctx}const void *p{args})
|
|
{{
|
|
{fields}
|
|
\tconst u8 *cursor = p;
|
|
\tsize_t plen = tal_len(p);
|
|
|
|
\tif (fromwire_u16(&cursor, &plen) != {enum.name})
|
|
\t\treturn false;
|
|
{subcalls}
|
|
\treturn cursor != NULL;
|
|
}}
|
|
"""
|
|
|
|
fromwire_header_templ = """bool fromwire_{name}({ctx}const void *p{args});
|
|
"""
|
|
|
|
towire_header_templ = """u8 *towire_{name}(const tal_t *ctx{args});
|
|
"""
|
|
towire_impl_templ = """u8 *towire_{name}(const tal_t *ctx{args})
|
|
{{
|
|
{field_decls}
|
|
\tu8 *p = tal_arr(ctx, u8, 0);
|
|
\ttowire_u16(&p, {enumname});
|
|
{subcalls}
|
|
|
|
\treturn memcheck(p, tal_count(p));
|
|
}}
|
|
"""
|
|
|
|
printwire_header_templ = """void printwire_{name}(const char *fieldname, const u8 *cursor);
|
|
"""
|
|
printwire_impl_templ = """void printwire_{name}(const char *fieldname, const u8 *cursor)
|
|
{{
|
|
\tsize_t plen = tal_len(cursor);
|
|
|
|
\tif (fromwire_u16(&cursor, &plen) != {enum.name}) {{
|
|
\t\tprintf("WRONG TYPE?!\\n");
|
|
\t\treturn;
|
|
\t}}
|
|
|
|
{subcalls}
|
|
|
|
\tif (plen != 0)
|
|
\t\tprintf("EXTRA: %s\\n", tal_hexstr(NULL, cursor, plen));
|
|
}}
|
|
"""
|
|
|
|
|
|
class CCode(object):
|
|
"""Simple class to create indented C code"""
|
|
def __init__(self):
|
|
self.indent = 1
|
|
self.single_indent = False
|
|
self.code = []
|
|
|
|
def append(self, lines):
|
|
for line in lines.split('\n'):
|
|
# Let us to the indenting please!
|
|
assert '\t' not in line
|
|
|
|
# Special case: } by itself is pre-unindented.
|
|
if line == '}':
|
|
self.indent -= 1
|
|
self.code.append("\t" * self.indent + line)
|
|
continue
|
|
|
|
self.code.append("\t" * self.indent + line)
|
|
if self.single_indent:
|
|
self.indent -= 1
|
|
self.single_indent = False
|
|
|
|
if line.endswith('{'):
|
|
self.indent += 1
|
|
elif line.endswith('}'):
|
|
self.indent -= 1
|
|
elif line.startswith('for') or line.startswith('if'):
|
|
self.indent += 1
|
|
self.single_indent = True
|
|
|
|
def __str__(self):
|
|
assert self.indent == 1
|
|
assert not self.single_indent
|
|
return '\n'.join(self.code)
|
|
|
|
|
|
class Message(object):
|
|
def __init__(self, name, enum, comments):
|
|
self.name = name
|
|
self.enum = enum
|
|
self.comments = comments
|
|
self.fields = []
|
|
self.has_variable_fields = False
|
|
|
|
def checkLenField(self, field):
|
|
# Optional fields don't have a len.
|
|
if field.optional:
|
|
return
|
|
for f in self.fields:
|
|
if f.name == field.lenvar:
|
|
if f.fieldtype.name != 'u16':
|
|
raise ValueError('Field {} has non-u16 length variable {} (type {})'
|
|
.format(field.name, field.lenvar, f.fieldtype.name))
|
|
|
|
if f.is_array() or f.needs_ptr_to_ptr():
|
|
raise ValueError('Field {} has non-simple length variable {}'
|
|
.format(field.name, field.lenvar))
|
|
f.is_len_var = True
|
|
f.lenvar_for = field
|
|
return
|
|
raise ValueError('Field {} unknown length variable {}'
|
|
.format(field.name, field.lenvar))
|
|
|
|
def addField(self, field):
|
|
# We assume field lengths are 16 bit, to avoid overflow issues and
|
|
# massive allocations.
|
|
if field.is_variable_size():
|
|
self.checkLenField(field)
|
|
self.has_variable_fields = True
|
|
elif field.basetype() in varlen_structs or field.optional:
|
|
self.has_variable_fields = True
|
|
self.fields.append(field)
|
|
|
|
def print_fromwire_array(self, subcalls, basetype, f, name, num_elems):
|
|
if f.has_array_helper():
|
|
subcalls.append('fromwire_{}_array(&cursor, &plen, {}, {});'
|
|
.format(basetype, name, num_elems))
|
|
else:
|
|
subcalls.append('for (size_t i = 0; i < {}; i++)'
|
|
.format(num_elems))
|
|
if f.fieldtype.is_assignable():
|
|
subcalls.append('({})[i] = fromwire_{}(&cursor, &plen);'
|
|
.format(name, basetype))
|
|
elif basetype in varlen_structs:
|
|
subcalls.append('({})[i] = fromwire_{}(ctx, &cursor, &plen);'
|
|
.format(name, basetype))
|
|
else:
|
|
subcalls.append('fromwire_{}(&cursor, &plen, {} + i);'
|
|
.format(basetype, name))
|
|
|
|
def print_fromwire(self, is_header):
|
|
ctx_arg = 'const tal_t *ctx, ' if self.has_variable_fields else ''
|
|
|
|
args = []
|
|
|
|
for f in self.fields:
|
|
if f.is_len_var or f.is_padding():
|
|
continue
|
|
elif f.is_array():
|
|
args.append(', {} {}[{}]'.format(f.fieldtype.name, f.name, f.num_elems))
|
|
else:
|
|
ptrs = '*'
|
|
# If we're handing a variable array, we need a ptr-to-ptr.
|
|
if f.needs_ptr_to_ptr():
|
|
ptrs += '*'
|
|
# If each type is a variable length, we need a ptr to that.
|
|
if f.basetype() in varlen_structs:
|
|
ptrs += '*'
|
|
|
|
args.append(', {} {}{}'.format(f.fieldtype.name, ptrs, f.name))
|
|
|
|
template = fromwire_header_templ if is_header else fromwire_impl_templ
|
|
fields = ['\t{} {};\n'.format(f.fieldtype.name, f.name) for f in self.fields if f.is_len_var]
|
|
|
|
subcalls = CCode()
|
|
for f in self.fields:
|
|
basetype = f.basetype()
|
|
|
|
for c in f.comments:
|
|
subcalls.append('/*{} */'.format(c))
|
|
|
|
if f.is_padding():
|
|
subcalls.append('fromwire_pad(&cursor, &plen, {});'
|
|
.format(f.num_elems))
|
|
elif f.is_array():
|
|
self.print_fromwire_array(subcalls, basetype, f, f.name,
|
|
f.num_elems)
|
|
elif f.is_variable_size():
|
|
subcalls.append("//2nd case {name}".format(name=f.name))
|
|
typename = f.fieldtype.name
|
|
# If structs are varlen, need array of ptrs to them.
|
|
if basetype in varlen_structs:
|
|
typename += ' *'
|
|
subcalls.append('*{} = {} ? tal_arr(ctx, {}, {}) : NULL;'
|
|
.format(f.name, f.lenvar, typename, f.lenvar))
|
|
|
|
self.print_fromwire_array(subcalls, basetype, f, '*' + f.name,
|
|
f.lenvar)
|
|
else:
|
|
if f.optional:
|
|
subcalls.append("if (!fromwire_bool(&cursor, &plen))\n"
|
|
"*{} = NULL;\n"
|
|
"else {{\n"
|
|
"*{} = tal(ctx, {});\n"
|
|
"fromwire_{}(&cursor, &plen, *{});\n"
|
|
"}}"
|
|
.format(f.name, f.name, f.fieldtype.name,
|
|
basetype, f.name))
|
|
elif f.is_assignable():
|
|
subcalls.append("//3th case {name}".format(name=f.name))
|
|
if f.is_len_var:
|
|
subcalls.append('{} = fromwire_{}(&cursor, &plen);'
|
|
.format(f.name, basetype))
|
|
else:
|
|
subcalls.append('*{} = fromwire_{}(&cursor, &plen);'
|
|
.format(f.name, basetype))
|
|
elif basetype in varlen_structs:
|
|
subcalls.append('*{} = fromwire_{}(ctx, &cursor, &plen);'
|
|
.format(f.name, basetype))
|
|
else:
|
|
subcalls.append('fromwire_{}(&cursor, &plen, {});'
|
|
.format(basetype, f.name))
|
|
|
|
return template.format(
|
|
name=self.name,
|
|
ctx=ctx_arg,
|
|
args=''.join(args),
|
|
fields=''.join(fields),
|
|
enum=self.enum,
|
|
subcalls=str(subcalls)
|
|
)
|
|
|
|
def print_towire_array(self, subcalls, basetype, f, num_elems):
|
|
if f.has_array_helper():
|
|
subcalls.append('towire_{}_array(&p, {}, {});'
|
|
.format(basetype, f.name, num_elems))
|
|
else:
|
|
subcalls.append('for (size_t i = 0; i < {}; i++)'
|
|
.format(num_elems))
|
|
if f.fieldtype.is_assignable() or basetype in varlen_structs:
|
|
subcalls.append('towire_{}(&p, {}[i]);'
|
|
.format(basetype, f.name))
|
|
else:
|
|
subcalls.append('towire_{}(&p, {} + i);'
|
|
.format(basetype, f.name))
|
|
|
|
def print_towire(self, is_header):
|
|
template = towire_header_templ if is_header else towire_impl_templ
|
|
args = []
|
|
for f in self.fields:
|
|
if f.is_padding() or f.is_len_var:
|
|
continue
|
|
if f.is_array():
|
|
args.append(', const {} {}[{}]'.format(f.fieldtype.name, f.name, f.num_elems))
|
|
elif f.is_assignable():
|
|
args.append(', {} {}'.format(f.fieldtype.name, f.name))
|
|
elif f.is_variable_size() and f.basetype() in varlen_structs:
|
|
args.append(', const {} **{}'.format(f.fieldtype.name, f.name))
|
|
else:
|
|
args.append(', const {} *{}'.format(f.fieldtype.name, f.name))
|
|
|
|
field_decls = []
|
|
for f in self.fields:
|
|
if f.is_len_var:
|
|
field_decls.append('\t{0} {1} = tal_count({2});'.format(
|
|
f.fieldtype.name, f.name, f.lenvar_for.name
|
|
))
|
|
|
|
subcalls = CCode()
|
|
for f in self.fields:
|
|
basetype = f.fieldtype.name
|
|
if basetype.startswith('struct '):
|
|
basetype = basetype[7:]
|
|
elif basetype.startswith('enum '):
|
|
basetype = basetype[5:]
|
|
|
|
for c in f.comments:
|
|
subcalls.append('/*{} */'.format(c))
|
|
|
|
if f.is_padding():
|
|
subcalls.append('towire_pad(&p, {});'
|
|
.format(f.num_elems))
|
|
elif f.is_array():
|
|
self.print_towire_array(subcalls, basetype, f, f.num_elems)
|
|
elif f.is_variable_size():
|
|
self.print_towire_array(subcalls, basetype, f, f.lenvar)
|
|
else:
|
|
if f.optional:
|
|
subcalls.append("if (!{})\n"
|
|
"towire_bool(&p, false);\n"
|
|
"else {{\n"
|
|
"towire_bool(&p, true);\n"
|
|
"towire_{}(&p, {});\n"
|
|
"}}".format(f.name, basetype, f.name))
|
|
else:
|
|
subcalls.append('towire_{}(&p, {});'
|
|
.format(basetype, f.name))
|
|
|
|
return template.format(
|
|
name=self.name,
|
|
args=''.join(args),
|
|
enumname=self.enum.name,
|
|
field_decls='\n'.join(field_decls),
|
|
subcalls=str(subcalls),
|
|
)
|
|
|
|
def add_truncate_check(self, subcalls):
|
|
# Report if truncated, otherwise print.
|
|
subcalls.append('if (!cursor) {\n'
|
|
'printf("**TRUNCATED**\\n");\n'
|
|
'return;\n'
|
|
'}')
|
|
|
|
def print_printwire_array(self, subcalls, basetype, f, num_elems):
|
|
if f.has_array_helper():
|
|
subcalls.append('printwire_{}_array(tal_fmt(NULL, "%s.{}", fieldname), &cursor, &plen, {});'
|
|
.format(basetype, f.name, num_elems))
|
|
else:
|
|
subcalls.append('printf("[");')
|
|
subcalls.append('for (size_t i = 0; i < {}; i++) {{'
|
|
.format(num_elems))
|
|
subcalls.append('{} v;'.format(f.fieldtype.name))
|
|
if f.fieldtype.is_assignable():
|
|
subcalls.append('v = fromwire_{}(&cursor, plen);'
|
|
.format(f.fieldtype.name, basetype))
|
|
else:
|
|
# We don't handle this yet!
|
|
assert(basetype not in varlen_structs)
|
|
|
|
subcalls.append('fromwire_{}(&cursor, &plen, &v);'
|
|
.format(basetype))
|
|
|
|
self.add_truncate_check(subcalls)
|
|
|
|
subcalls.append('printwire_{}(tal_fmt(NULL, "%s.{}", fieldname), &v);'
|
|
.format(basetype, f.name))
|
|
subcalls.append('}')
|
|
subcalls.append('printf("]");')
|
|
|
|
def print_printwire(self, is_header):
|
|
template = printwire_header_templ if is_header else printwire_impl_templ
|
|
fields = ['\t{} {};\n'.format(f.fieldtype.name, f.name) for f in self.fields if f.is_len_var]
|
|
|
|
subcalls = CCode()
|
|
for f in self.fields:
|
|
basetype = f.basetype()
|
|
|
|
for c in f.comments:
|
|
subcalls.append('/*{} */'.format(c))
|
|
|
|
if f.is_len_var:
|
|
subcalls.append('{} {} = fromwire_{}(&cursor, &plen);'
|
|
.format(f.fieldtype.name, f.name, basetype))
|
|
self.add_truncate_check(subcalls)
|
|
continue
|
|
|
|
subcalls.append('printf("{}=");'.format(f.name))
|
|
if f.is_padding():
|
|
subcalls.append('printwire_pad(tal_fmt(NULL, "%s.{}", fieldname), &cursor, &plen, {});'
|
|
.format(f.name, f.num_elems))
|
|
self.add_truncate_check(subcalls)
|
|
elif f.is_array():
|
|
self.print_printwire_array(subcalls, basetype, f, f.num_elems)
|
|
self.add_truncate_check(subcalls)
|
|
elif f.is_variable_size():
|
|
self.print_printwire_array(subcalls, basetype, f, f.lenvar)
|
|
self.add_truncate_check(subcalls)
|
|
else:
|
|
if f.optional:
|
|
subcalls.append("if (fromwire_bool(&cursor, &plen)) {")
|
|
|
|
if f.is_assignable():
|
|
subcalls.append('{} {} = fromwire_{}(&cursor, &plen);'
|
|
.format(f.fieldtype.name, f.name, basetype))
|
|
else:
|
|
# Don't handle these yet.
|
|
assert(basetype not in varlen_structs)
|
|
subcalls.append('{} {};'.
|
|
format(f.fieldtype.name, f.name))
|
|
subcalls.append('fromwire_{}(&cursor, &plen, &{});'
|
|
.format(basetype, f.name))
|
|
|
|
self.add_truncate_check(subcalls)
|
|
subcalls.append('printwire_{}(tal_fmt(NULL, "%s.{}", fieldname), &{});'
|
|
.format(basetype, f.name, f.name))
|
|
if f.optional:
|
|
subcalls.append("} else {")
|
|
self.add_truncate_check(subcalls)
|
|
subcalls.append("}")
|
|
|
|
return template.format(
|
|
name=self.name,
|
|
fields=''.join(fields),
|
|
enum=self.enum,
|
|
subcalls=str(subcalls)
|
|
)
|
|
|
|
|
|
def find_message(messages, name):
|
|
for m in messages:
|
|
if m.name == name:
|
|
return m
|
|
|
|
return None
|
|
|
|
|
|
def find_message_with_option(messages, optional_messages, name, option):
|
|
fullname = name + "_" + option.replace('-', '_')
|
|
|
|
base = find_message(messages, name)
|
|
if not base:
|
|
raise ValueError('Unknown message {}'.format(name))
|
|
|
|
m = find_message(optional_messages, fullname)
|
|
if not m:
|
|
# Add a new option.
|
|
m = copy.deepcopy(base)
|
|
m.name = fullname
|
|
optional_messages.append(m)
|
|
return m
|
|
|
|
|
|
parser = argparse.ArgumentParser(description='Generate C from CSV')
|
|
parser.add_argument('--header', action='store_true', help="Create wire header")
|
|
parser.add_argument('--bolt', action='store_true', help="Generate wire-format for BOLT")
|
|
parser.add_argument('--printwire', action='store_true', help="Create print routines")
|
|
parser.add_argument('headerfilename', help='The filename of the header')
|
|
parser.add_argument('enumname', help='The name of the enum to produce')
|
|
parser.add_argument('files', nargs='*', help='Files to read in (or stdin)')
|
|
options = parser.parse_args()
|
|
|
|
# Maps message names to messages
|
|
messages = []
|
|
messages_with_option = []
|
|
comments = []
|
|
includes = []
|
|
prevfield = None
|
|
|
|
# Read csv lines. Single comma is the message values, more is offset/len.
|
|
for line in fileinput.input(options.files):
|
|
# #include gets inserted into header
|
|
if line.startswith('#include '):
|
|
includes.append(line)
|
|
continue
|
|
|
|
by_comments = line.rstrip().split('#')
|
|
|
|
# Emit a comment if they included one
|
|
if by_comments[1:]:
|
|
comments.append(' '.join(by_comments[1:]))
|
|
|
|
parts = by_comments[0].split(',')
|
|
if parts == ['']:
|
|
continue
|
|
|
|
if len(parts) == 2:
|
|
# eg commit_sig,132
|
|
messages.append(Message(parts[0], Enumtype("WIRE_" + parts[0].upper(), parts[1]), comments))
|
|
comments = []
|
|
prevfield = None
|
|
else:
|
|
if len(parts) == 4:
|
|
# eg commit_sig,0,channel-id,8 OR
|
|
# commit_sig,0,channel-id,u64
|
|
m = find_message(messages, parts[0])
|
|
if m is None:
|
|
raise ValueError('Unknown message {}'.format(parts[0]))
|
|
elif len(parts) == 5:
|
|
# eg.
|
|
# channel_reestablish,48,your_last_per_commitment_secret,32,option209
|
|
m = find_message_with_option(messages, messages_with_option, parts[0], parts[4])
|
|
else:
|
|
raise ValueError('Line {} malformed'.format(line.rstrip()))
|
|
|
|
f = Field(m.name, parts[2], parts[3], comments, prevfield)
|
|
m.addField(f)
|
|
# If it used prevfield as lenvar, keep that for next
|
|
# time (multiple fields can use the same lenvar).
|
|
if not f.lenvar:
|
|
prevfield = parts[2]
|
|
comments = []
|
|
|
|
header_template = """/* This file was generated by generate-wire.py */
|
|
/* Do not modify this file! Modify the _csv file it was generated from. */
|
|
#ifndef LIGHTNING_{idem}
|
|
#define LIGHTNING_{idem}
|
|
#include <ccan/tal/tal.h>
|
|
#include <wire/wire.h>
|
|
{includes}
|
|
enum {enumname} {{
|
|
{enums}}};
|
|
const char *{enumname}_name(int e);
|
|
|
|
{func_decls}
|
|
#endif /* LIGHTNING_{idem} */
|
|
"""
|
|
|
|
impl_template = """/* This file was generated by generate-wire.py */
|
|
/* Do not modify this file! Modify the _csv file it was generated from. */
|
|
#include <{headerfilename}>
|
|
#include <ccan/mem/mem.h>
|
|
#include <ccan/tal/str/str.h>
|
|
#include <stdio.h>
|
|
|
|
const char *{enumname}_name(int e)
|
|
{{
|
|
\tstatic char invalidbuf[sizeof("INVALID ") + STR_MAX_CHARS(e)];
|
|
|
|
\tswitch ((enum {enumname})e) {{
|
|
\t{cases}
|
|
\t}}
|
|
|
|
\tsprintf(invalidbuf, "INVALID %i", e);
|
|
\treturn invalidbuf;
|
|
}}
|
|
|
|
{func_decls}
|
|
"""
|
|
|
|
print_header_template = """/* This file was generated by generate-wire.py */
|
|
/* Do not modify this file! Modify the _csv file it was generated from. */
|
|
#ifndef LIGHTNING_{idem}
|
|
#define LIGHTNING_{idem}
|
|
#include <ccan/tal/tal.h>
|
|
#include <devtools/print_wire.h>
|
|
{includes}
|
|
|
|
void print_message(const u8 *msg);
|
|
|
|
{func_decls}
|
|
#endif /* LIGHTNING_{idem} */
|
|
"""
|
|
|
|
print_template = """/* This file was generated by generate-wire.py */
|
|
/* Do not modify this file! Modify the _csv file it was generated from. */
|
|
#include "{headerfilename}"
|
|
#include <ccan/mem/mem.h>
|
|
#include <ccan/tal/str/str.h>
|
|
#include <common/utils.h>
|
|
#include <stdio.h>
|
|
|
|
void print_message(const u8 *msg)
|
|
{{
|
|
\tswitch ((enum {enumname})fromwire_peektype(msg)) {{
|
|
\t{printcases}
|
|
\t}}
|
|
|
|
\tprintf("UNKNOWN: %s\\n", tal_hex(msg, msg));
|
|
}}
|
|
|
|
{func_decls}
|
|
"""
|
|
|
|
idem = re.sub(r'[^A-Z]+', '_', options.headerfilename.upper())
|
|
if options.printwire:
|
|
if options.header:
|
|
template = print_header_template
|
|
else:
|
|
template = print_template
|
|
elif options.header:
|
|
template = header_template
|
|
else:
|
|
template = impl_template
|
|
|
|
# Dump out enum, sorted by value order.
|
|
enums = ""
|
|
for m in messages:
|
|
for c in m.comments:
|
|
enums += '\t/*{} */\n'.format(c)
|
|
enums += '\t{} = {},\n'.format(m.enum.name, m.enum.value)
|
|
includes = '\n'.join(includes)
|
|
cases = ['case {enum.name}: return "{enum.name}";'.format(enum=m.enum) for m in messages]
|
|
printcases = ['case {enum.name}: printf("{enum.name}:\\n"); printwire_{name}("{name}", msg); return;'.format(enum=m.enum, name=m.name) for m in messages]
|
|
|
|
if options.printwire:
|
|
decls = [m.print_printwire(options.header) for m in messages + messages_with_option]
|
|
else:
|
|
fromwire_decls = [m.print_fromwire(options.header) for m in messages + messages_with_option]
|
|
towire_decls = towire_decls = [m.print_towire(options.header) for m in messages + messages_with_option]
|
|
decls = fromwire_decls + towire_decls
|
|
|
|
print(template.format(
|
|
headerfilename=options.headerfilename,
|
|
cases='\n\t'.join(cases),
|
|
printcases='\n\t'.join(printcases),
|
|
idem=idem,
|
|
includes=includes,
|
|
enumname=options.enumname,
|
|
enums=enums,
|
|
func_decls='\n'.join(decls)))
|