mirror of
https://github.com/pmret/papermario.git
synced 2024-11-09 20:42:41 +01:00
ae66312d8c
* Add Python linter to github actions * wip * Add back splat_ext * Format files * C++ -> C * format 2 files * split workflow into separate file, line length 120, fix excludes * -l 120 in ci * update black locally and apply formatting changes * pyproject.toject --------- Co-authored-by: Ethan Roseman <ethteck@gmail.com>
117 lines
3.8 KiB
Python
Executable File
117 lines
3.8 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
|
|
from sys import argv
|
|
from collections import OrderedDict
|
|
import re
|
|
import msgpack
|
|
import os
|
|
|
|
|
|
class Message:
|
|
def __init__(self, d: dict, header_file_index: int):
|
|
self.section = d.get("section")
|
|
self.index = d.get("index")
|
|
self.name = d.get("name")
|
|
self.bytes = d["bytes"]
|
|
self.header_file_index = header_file_index
|
|
|
|
|
|
if __name__ == "__main__":
|
|
if len(argv) < 3:
|
|
print("usage: combine.py [out.bin] [out.h] [compiled...]")
|
|
exit(1)
|
|
|
|
_, outfile, header_file, *infiles = argv
|
|
|
|
messages = []
|
|
# header_files = []
|
|
|
|
for i, infile in enumerate(infiles):
|
|
# if infile == "--headers":
|
|
# header_files = infiles[i+1:]
|
|
# break
|
|
|
|
with open(infile, "rb") as f:
|
|
messages.extend(Message(msg, i) for msg in msgpack.unpack(f))
|
|
|
|
with open(outfile, "wb") as f:
|
|
# sectioned+indexed, followed by just sectioned, followed by just indexed, followed by named (unsectioned & unindexed)
|
|
# messages.sort(key=lambda msg: bool(msg.section)<<2 + bool(msg.index))
|
|
|
|
names = set()
|
|
|
|
sections = []
|
|
# messages_by_file = {}
|
|
|
|
for message in messages:
|
|
if message.section is None:
|
|
# allocate a section
|
|
for section_idx, section in enumerate(sections):
|
|
if len(section) < 0xFFF:
|
|
break
|
|
message.section = section_idx
|
|
else:
|
|
section_idx = message.section
|
|
while len(sections) <= section_idx:
|
|
sections.append({})
|
|
section = sections[section_idx]
|
|
|
|
if message.index is None:
|
|
message.index = len(section)
|
|
|
|
# if message.name:
|
|
# if message.name in names:
|
|
# print(f"error: multiple messages with name '{message.name}'")
|
|
# exit(1)
|
|
# else:
|
|
# names.add(message.name)
|
|
|
|
# if message.header_file_index in messages_by_file:
|
|
# messages_by_file[message.header_file_index].add(message)
|
|
# else:
|
|
# messages_by_file[message.header_file_index] = set([message])
|
|
|
|
if message.index in section:
|
|
print(f"warning: multiple messages allocated to id {section_idx:02X}:{message.index:03X}")
|
|
|
|
if section[message.index].name and message.name:
|
|
print(f"warning: message '{section[message.index].name}' and '{message.name}' conflict")
|
|
|
|
section[message.index] = message
|
|
|
|
f.seek((len(sections) + 1) * 4) # skip past table of contents
|
|
|
|
section_offsets = []
|
|
for section in sections:
|
|
# convert dict into sorted list
|
|
section = [msg for idx, msg in sorted(section.items(), key=lambda ele: ele[0])]
|
|
|
|
message_offsets = []
|
|
for message in section:
|
|
message_offsets.append(f.tell())
|
|
f.write(message.bytes)
|
|
|
|
section_offset = f.tell()
|
|
section_offsets.append(section_offset)
|
|
for offset in message_offsets:
|
|
f.write(offset.to_bytes(4, byteorder="big"))
|
|
f.write(section_offset.to_bytes(4, byteorder="big"))
|
|
|
|
# padding
|
|
while f.tell() % 0x10 != 0:
|
|
f.write(b"\0\0\0\0")
|
|
|
|
f.seek(0)
|
|
for offset in section_offsets:
|
|
f.write(offset.to_bytes(4, byteorder="big"))
|
|
f.write(b"\0\0\0\0")
|
|
|
|
with open(header_file, "w") as f:
|
|
f.write(f"#ifndef _MESSAGE_IDS_H_\n" f"#define _MESSAGE_IDS_H_\n" "\n" '#include "messages.h"\n' "\n")
|
|
|
|
for message in messages:
|
|
if message.name:
|
|
f.write(f"#define MSG_{message.name} MESSAGE_ID(0x{message.section:02X}, 0x{message.index:03X})\n")
|
|
|
|
f.write("\n#endif\n")
|