sm64coopdx/tools/assemble_sound.py

#!/usr/bin/env python3
from collections import namedtuple, OrderedDict
from json import JSONDecoder
import os
import re
import struct
import subprocess
import sys
import zlib

TYPE_CTL = 1
TYPE_TBL = 2
TYPE_SEQ = 3

STACK_TRACES = False
DUMP_INDIVIDUAL_BINS = False
ENDIAN_MARKER = ">"
WORD_BYTES = 4

orderedJsonDecoder = JSONDecoder(object_pairs_hook=OrderedDict)


class Aifc:
    def __init__(self, name, fname, data, sample_rate, book, loop):
        self.name = name
        self.fname = fname
        self.data = data
        self.sample_rate = sample_rate
        self.book = book
        self.loop = loop
        self.used = False
        self.offset = None


class SampleBank:
    def __init__(self, name, entries):
        self.name = name
        self.uses = []
        self.index = None
        self.entries = entries
        self.name_to_entry = {}
        for e in entries:
            self.name_to_entry[e.name] = e


Book = namedtuple("Book", ["order", "npredictors", "table"])
Loop = namedtuple("Loop", ["start", "end", "count", "state"])
Bank = namedtuple("Bank", ["name", "sample_bank", "json"])


def align(val, al):
    return (val + (al - 1)) & -al


def fail(msg):
    print(msg, file=sys.stderr)
    if STACK_TRACES:
        raise Exception("re-raising exception")
    sys.exit(1)


def validate(cond, msg, forstr=""):
    if not cond:
        if forstr:
            msg += " for " + forstr
        raise Exception(msg)


def strip_comments(string):
    string = re.sub(re.compile(r"/\*.*?\*/", re.DOTALL), "", string)
    return re.sub(re.compile("//.*?\n"), "", string)


def pack(fmt, *args):
    if WORD_BYTES == 4:
        fmt = fmt.replace("P", "I").replace("X", "")
    else:
        fmt = fmt.replace("P", "Q").replace("X", "xxxx")
    return struct.pack(ENDIAN_MARKER + fmt, *args)


def to_bcd(num):
    assert num >= 0
    shift = 0
    ret = 0
    while num:
        ret |= (num % 10) << shift
        shift += 4
        num //= 10
    return ret


def parse_f80(data):
    exp_bits, mantissa_bits = struct.unpack(">HQ", data)
    sign_bit = exp_bits & 2 ** 15
    exp_bits ^= sign_bit
    sign = -1 if sign_bit else 1
    if exp_bits == mantissa_bits == 0:
        return sign * 0.0
    validate(exp_bits != 0, "sample rate is a denormal")
    validate(exp_bits != 0x7FFF, "sample rate is infinity/nan")
    mant = float(mantissa_bits) / 2 ** 63
    return sign * mant * pow(2, exp_bits - 0x3FFF)


def parse_aifc_loop(data):
    validate(len(data) == 48, "loop chunk size should be 48")
    version, nloops, start, end, count = struct.unpack(">HHIIi", data[:16])
    validate(version == 1, "loop version doesn't match")
    validate(nloops == 1, "only one loop is supported")
    state = []
    for i in range(16, len(data), 2):
        state.append(struct.unpack(">h", data[i : i + 2])[0])
    return Loop(start, end, count, state)


def parse_aifc_book(data):
    version, order, npredictors = struct.unpack(">hhh", data[:6])
    validate(version == 1, "codebook version doesn't match")
    validate(
        len(data) == 6 + 16 * order * npredictors,
        "predictor book chunk size doesn't match",
    )
    table = []
    for i in range(6, len(data), 2):
        table.append(struct.unpack(">h", data[i : i + 2])[0])
    return Book(order, npredictors, table)


def parse_aifc(data, name, fname):
    validate(data[:4] == b"FORM", "must start with FORM")
    validate(data[8:12] == b"AIFC", "format must be AIFC")
    i = 12
    sections = []
    while i < len(data):
        tp = data[i : i + 4]
        (le,) = struct.unpack(">I", data[i + 4 : i + 8])
        i += 8
        sections.append((tp, data[i : i + le]))
        i = align(i + le, 2)

    audio_data = None
    vadpcm_codes = None
    vadpcm_loops = None
    sample_rate = None

    for (tp, data) in sections:
        if tp == b"APPL" and data[:4] == b"stoc":
            plen = data[4]
            tp = data[5 : 5 + plen]
            data = data[align(5 + plen, 2) :]
            if tp == b"VADPCMCODES":
                vadpcm_codes = data
            elif tp == b"VADPCMLOOPS":
                vadpcm_loops = data
        elif tp == b"SSND":
            audio_data = data[8:]
        elif tp == b"COMM":
            sample_rate = parse_f80(data[8:18])

    validate(sample_rate is not None, "no COMM section")
    validate(audio_data is not None, "no SSND section")
    validate(vadpcm_codes is not None, "no VADPCM table")

    book = parse_aifc_book(vadpcm_codes)
    loop = parse_aifc_loop(vadpcm_loops) if vadpcm_loops is not None else None
    return Aifc(name, fname, audio_data, sample_rate, book, loop)


class ReserveSerializer:
    def __init__(self):
        self.parts = []
        self.sizes = []
        self.size = 0

    def add(self, part):
        assert isinstance(part, (bytes, list))
        self.parts.append(part)
        self.sizes.append(len(part))
        self.size += len(part)

    def reserve(self, space):
        li = []
        self.parts.append(li)
        self.sizes.append(space)
        self.size += space
        return li

    def align(self, alignment):
        new_size = (self.size + alignment - 1) & -alignment
        self.add((new_size - self.size) * b"\0")

    def finish(self):
        flat_parts = []
        for (li, si) in zip(self.parts, self.sizes):
            if isinstance(li, list):
                li = b"".join(li)
            assert (
                len(li) == si
            ), "unfulfilled reservation of size {}, only got {}".format(si, len(li))
            flat_parts.append(li)
        return b"".join(flat_parts)


class GarbageSerializer:
    def __init__(self):
        self.garbage_bufs = [[]]
        self.parts = []
        self.size = 0
        self.garbage_pos = 0

    def reset_garbage_pos(self):
        self.garbage_bufs.append([])
        self.garbage_pos = 0

    def add(self, part):
        assert isinstance(part, bytes)
        self.parts.append(part)
        self.garbage_bufs[-1].append((self.garbage_pos, part))
        self.size += len(part)
        self.garbage_pos += len(part)

    def align(self, alignment):
        new_size = (self.size + alignment - 1) & -alignment
        self.add((new_size - self.size) * b"\0")

    def garbage_at(self, pos):
        # Find the last write to position pos & 0xffff, assuming a cyclic
        # buffer of size 0x10000 where the write position is reset to 0 on
        # each call to reset_garbage_pos.
        pos &= 0xFFFF
        for bufs in self.garbage_bufs[::-1]:
            for (bpos, buf) in bufs[::-1]:
                q = ((bpos + len(buf) - 1 - pos) & ~0xFFFF) + pos
                if q >= bpos:
                    return buf[q - bpos]
        return 0

    def align_garbage(self, alignment):
        while self.size % alignment != 0:
            self.add(bytes([self.garbage_at(self.garbage_pos)]))

    def finish(self):
        return b"".join(self.parts)


def validate_json_format(json, fmt, forstr=""):
    constructor_to_name = {
        str: "a string",
        dict: "an object",
        int: "an integer",
        float: "a floating point number",
        list: "an array",
    }
    for key, tp in fmt.items():
        validate(key in json, 'missing key "' + key + '"', forstr)
        if isinstance(tp, list):
            validate_int_in_range(json[key], tp[0], tp[1], '"' + key + '"', forstr)
        else:
            validate(
                isinstance(json[key], tp)
                or (tp == float and isinstance(json[key], int)),
                '"{}" must be {}'.format(key, constructor_to_name[tp]),
                forstr,
            )


def validate_int_in_range(val, lo, hi, msg, forstr=""):
    validate(isinstance(val, int), "{} must be an integer".format(msg), forstr)
    validate(
        lo <= val <= hi, "{} must be in range {} to {}".format(msg, lo, hi), forstr
    )


def validate_sound(json, sample_bank, forstr=""):
    validate_json_format(json, {"sample": str}, forstr)
    if "tuning" in json:
        validate_json_format(json, {"tuning": float}, forstr)
    validate(
        json["sample"] in sample_bank.name_to_entry,
        "reference to sound {} which isn't found in sample bank {}".format(
            json["sample"], sample_bank.name
        ),
        forstr,
    )


def validate_bank_toplevel(json):
    validate(isinstance(json, dict), "must have a top-level object")
    validate_json_format(
        json,
        {
            "envelopes": dict,
            "sample_bank": str,
            "instruments": dict,
            "instrument_list": list,
        },
    )


def normalize_sound_json(json):
    # Convert {"sound": "str"} into {"sound": {"sample": "str"}}
    fixup = []
    for inst in json["instruments"].values():
        if isinstance(inst, list):
            for drum in inst:
                fixup.append((drum, "sound"))
        else:
            fixup.append((inst, "sound_lo"))
            fixup.append((inst, "sound"))
            fixup.append((inst, "sound_hi"))
    for (obj, key) in fixup:
        if isinstance(obj, dict) and isinstance(obj.get(key), str):
            obj[key] = {"sample": obj[key]}


def validate_bank(json, sample_bank):
    if "date" in json:
        validate(
            isinstance(json["date"], str)
            and re.match(r"[0-9]{4}-[0-9]{2}-[0-9]{2}\Z", json["date"]),
            "date must have format yyyy-mm-dd",
        )

    for key, env in json["envelopes"].items():
        validate(isinstance(env, list), 'envelope "' + key + '" must be an array')
        last_fine = False
        for entry in env:
            if entry in ["stop", "hang", "restart"]:
                last_fine = True
            else:
                validate(
                    isinstance(entry, list) and len(entry) == 2,
                    'envelope entry in "'
                    + key
                    + '" must be a list of length 2, or one of stop/hang/restart',
                )
                if entry[0] == "goto":
                    validate_int_in_range(
                        entry[1], 0, len(env) - 2, "envelope goto target out of range:"
                    )
                    last_fine = True
                else:
                    validate_int_in_range(
                        entry[0], 1, 2 ** 16 - 4, "envelope entry's first part"
                    )
                    validate_int_in_range(
                        entry[1], 0, 2 ** 16 - 1, "envelope entry's second part"
                    )
                    last_fine = False
        validate(
            last_fine, 'envelope "{}" must end with stop/hang/restart/goto'.format(key)
        )

    drums = []
    instruments = []
    instrument_names = set()
    for name, inst in json["instruments"].items():
        if name == "percussion":
            validate(isinstance(inst, list), "drums entry must be a list")
            drums = inst
        else:
            validate(isinstance(inst, dict), "instrument entry must be an object")
            instruments.append((name, inst))
            instrument_names.add(name)

    for drum in drums:
        validate(isinstance(drum, dict), "drum entry must be an object")
        validate_json_format(
            drum,
            {"release_rate": [0, 255], "pan": [0, 128], "envelope": str, "sound": dict},
        )
        validate_sound(drum["sound"], sample_bank)
        validate(
            drum["envelope"] in json["envelopes"],
            "reference to non-existent envelope " + drum["envelope"],
            "drum",
        )

    no_sound = {}

    for name, inst in instruments:
        forstr = "instrument " + name
        for lohi in ["lo", "hi"]:
            nr = "normal_range_" + lohi
            so = "sound_" + lohi
            if nr in inst:
                validate(so in inst, nr + " is specified, but not " + so, forstr)
            if so in inst:
                validate(nr in inst, so + " is specified, but not " + nr, forstr)
            else:
                inst[so] = no_sound
        if "normal_range_lo" not in inst:
            inst["normal_range_lo"] = 0
        if "normal_range_hi" not in inst:
            inst["normal_range_hi"] = 127

        validate_json_format(
            inst,
            {
                "release_rate": [0, 255],
                "envelope": str,
                "normal_range_lo": [0, 127],
                "normal_range_hi": [0, 127],
                "sound_lo": dict,
                "sound": dict,
                "sound_hi": dict,
            },
            forstr,
        )

        if "ifdef" in inst:
            validate(
                isinstance(inst["ifdef"], list)
                and all(isinstance(x, str) for x in inst["ifdef"]),
                '"ifdef" must be an array of strings',
            )

        validate(
            inst["normal_range_lo"] <= inst["normal_range_hi"],
            "normal_range_lo > normal_range_hi",
            forstr,
        )
        validate(
            inst["envelope"] in json["envelopes"],
            "reference to non-existent envelope " + inst["envelope"],
            forstr,
        )
        for key in ["sound_lo", "sound", "sound_hi"]:
            if inst[key] is no_sound:
                del inst[key]
            else:
                validate_sound(inst[key], sample_bank, forstr)

    seen_instruments = set()
    for inst in json["instrument_list"]:
        if inst is None:
            continue
        validate(
            isinstance(inst, str),
            "instrument list should contain only strings and nulls",
        )
        validate(
            inst in instrument_names, "reference to non-existent instrument " + inst
        )
        validate(
            inst not in seen_instruments, inst + " occurs twice in the instrument list"
        )
        seen_instruments.add(inst)

    for inst in instrument_names:
        validate(inst in seen_instruments, "unreferenced instrument " + inst)


def apply_ifs(json, defines):
    if isinstance(json, dict) and "ifdef" in json and "then" in json and "else" in json:
        validate_json_format(json, {"ifdef": list})
        true = any(d in defines for d in json["ifdef"])
        return apply_ifs(json["then"] if true else json["else"], defines)
    elif isinstance(json, list):
        for i in range(len(json)):
            json[i] = apply_ifs(json[i], defines)
    elif isinstance(json, dict):
        for key in json:
            json[key] = apply_ifs(json[key], defines)
    return json


def apply_version_diffs(json, defines):
    date_str = json.get("date")
    if "VERSION_EU" in defines and isinstance(date_str, str):
        json["date"] = date_str.replace("1996-03-19", "1996-06-24")

    ifdef_removed = set()
    for key, inst in json["instruments"].items():
        if (
            isinstance(inst, dict)
            and isinstance(inst.get("ifdef"), list)
            and all(d not in defines for d in inst["ifdef"])
        ):
            ifdef_removed.add(key)
    for key in ifdef_removed:
        del json["instruments"][key]
        json["instrument_list"].remove(key)


def mark_sample_bank_uses(bank):
    bank.sample_bank.uses.append(bank)

    def mark_used(name):
        bank.sample_bank.name_to_entry[name].used = True

    for inst in bank.json["instruments"].values():
        if isinstance(inst, list):
            for drum in inst:
                mark_used(drum["sound"]["sample"])
        else:
            if "sound_lo" in inst:
                mark_used(inst["sound_lo"]["sample"])
            mark_used(inst["sound"]["sample"])
            if "sound_hi" in inst:
                mark_used(inst["sound_hi"]["sample"])


def serialize_ctl(bank, base_ser, is_shindou, asset_offsets):
    json = bank.json

    drums = []
    instruments = []
    for inst in json["instruments"].values():
        if isinstance(inst, list):
            drums = inst
        else:
            instruments.append(inst)

    if not is_shindou:
        y, m, d = map(int, json.get("date", "0000-00-00").split("-"))
        date = y * 10000 + m * 100 + d
        base_ser.add(
            pack(
                "IIII",
                len(json["instrument_list"]),
                len(drums),
                1 if len(bank.sample_bank.uses) > 1 else 0,
                to_bcd(date),
            )
        )

    ser = ReserveSerializer()
    if drums:
        drum_pos_buf = ser.reserve(WORD_BYTES)
    else:
        ser.add(b"\0" * WORD_BYTES)
        drum_pos_buf = None

    inst_pos_buf = ser.reserve(WORD_BYTES * len(json["instrument_list"]))
    ser.align(16)

    used_samples = []
    for inst in json["instruments"].values():
        if isinstance(inst, list):
            for drum in inst:
                used_samples.append(drum["sound"]["sample"])
        else:
            if "sound_lo" in inst:
                used_samples.append(inst["sound_lo"]["sample"])
            used_samples.append(inst["sound"]["sample"])
            if "sound_hi" in inst:
                used_samples.append(inst["sound_hi"]["sample"])

    sample_name_to_addr = {}
    for name in used_samples:
        if name in sample_name_to_addr:
            continue
        sample_name_to_addr[name] = ser.size
        aifc = bank.sample_bank.name_to_entry[name]
        sample_len = len(aifc.data)

        # Sample
        ser.add(pack("IX", align(sample_len, 2) if is_shindou else 0))
        ser.add(pack("P", aifc.offset))
        loop_addr_buf = ser.reserve(WORD_BYTES)
        book_addr_buf = ser.reserve(WORD_BYTES)
        if not is_shindou:
            ser.add(pack("I", align(sample_len, 2)))
        ser.align(16)

        # Book
        book_addr_buf.append(pack("P", ser.size))
        ser.add(pack("ii", aifc.book.order, aifc.book.npredictors))
        for x in aifc.book.table:
            ser.add(pack("h", x))
        ser.align(16)

        # Loop
        loop_addr_buf.append(pack("P", ser.size))
        if aifc.loop is None:
            assert sample_len % 9 in [0, 1]
            end = sample_len // 9 * 16 + (sample_len % 2) + (sample_len % 9)
            ser.add(pack("IIiI", 0, end, 0, 0))
        else:
            ser.add(pack("IIiI", aifc.loop.start, aifc.loop.end, aifc.loop.count, 0))
            assert aifc.loop.count != 0
            for x in aifc.loop.state:
                ser.add(pack("h", x))
        ser.align(16)

    env_name_to_addr = {}
    for name, env in json["envelopes"].items():
        env_name_to_addr[name] = ser.size
        for entry in env:
            if entry == "stop":
                entry = [0, 0]
            elif entry == "hang":
                entry = [2 ** 16 - 1, 0]
            elif entry == "restart":
                entry = [2 ** 16 - 3, 0]
            elif entry[0] == "goto":
                entry[0] = 2 ** 16 - 2
            # Envelopes are always written as big endian, to match sequence files
            # which are byte blobs and can embed envelopes.
            ser.add(struct.pack(">HH", *entry))
        ser.align(16)

    def ser_sound(sound):
        sample_addr = (
            0 if sound["sample"] is None else sample_name_to_addr[sound["sample"]]
        )
        if "tuning" in sound:
            tuning = sound["tuning"]
        else:
            aifc = bank.sample_bank.name_to_entry[sound["sample"]]
            tuning = aifc.sample_rate / 32000
        ser.add(pack("PfX", sample_addr, tuning))

    no_sound = {"sample": None, "tuning": 0.0}

    inst_name_to_pos = {}
    for name, inst in json["instruments"].items():
        if isinstance(inst, list):
            continue
        inst_name_to_pos[name] = ser.size
        env_addr = env_name_to_addr[inst["envelope"]]
        ser.add(
            pack(
                "BBBBXP",
                0,
                inst.get("normal_range_lo", 0),
                inst.get("normal_range_hi", 127),
                inst["release_rate"],
                env_addr,
            )
        )
        ser_sound(inst.get("sound_lo", no_sound))
        ser_sound(inst["sound"])
        ser_sound(inst.get("sound_hi", no_sound))
    ser.align(16)

    for name in json["instrument_list"]:
        if name is None:
            inst_pos_buf.append(pack("P", 0))
            continue
        inst_pos_buf.append(pack("P", inst_name_to_pos[name]))

    if drums:
        drum_poses = []
        for drum in drums:
            drum_poses.append(ser.size)
            ser.add(pack("BBBBX", drum["release_rate"], drum["pan"], 0, 0))
            ser_sound(drum["sound"])
            env_addr = env_name_to_addr[drum["envelope"]]
            ser.add(pack("P", env_addr))
        ser.align(16)

        drum_pos_buf.append(pack("P", ser.size))
        for pos in drum_poses:
            ser.add(pack("P", pos))
        ser.align(16)

    base_ser.add(ser.finish())

    return pack(
        "BBBB", bank.sample_bank.index, 0xFF, len(json["instrument_list"]), len(drums)
    )


def serialize_tbl(sample_bank, ser, is_shindou, asset_offsets):
    ser.reset_garbage_pos()
    base_addr = ser.size
    for aifc in sample_bank.entries:
        if not aifc.used:
            continue
        ser.align(16)
        aifc.offset = ser.size - base_addr

        if 'custom' in aifc.fname:
            ser.add(aifc.data)
        else:
            asset_offsets[aifc.fname] = ser.size
            fake_data = list(aifc.fname)
            fake_data = [ord(char) for char in fake_data]
            padding = [0] * (len(aifc.data) - len(fake_data))
            fake_data = fake_data + padding
            ser.add(bytes(fake_data))
    ser.align(2)
    if is_shindou and sample_bank.index not in [4, 10]:
        ser.align(16)
    else:
        ser.align_garbage(16)


def serialize_seqfile(
    out_filename,
    out_header_filename,
    entries,
    serialize_entry,
    entry_list,
    magic,
    is_shindou,
    extra_padding=True,
):
    data_ser = GarbageSerializer()
    entry_offsets = []
    entry_lens = []
    entry_meta = []
    asset_offsets = {}
    for entry in entries:
        entry_offsets.append(data_ser.size)
        ret = serialize_entry(entry, data_ser, is_shindou, asset_offsets)
        entry_meta.append(ret)
        entry_lens.append(data_ser.size - entry_offsets[-1])
    data = data_ser.finish()

    if is_shindou:
        ser = ReserveSerializer()
        ser.add(pack("H", len(entries)))
        ser.align(16)
        sh_magic = 0x0204 if magic == TYPE_TBL else 0x0203

        # Ignore entry_list and loop over all entries instead. This makes a
        # difference for sample banks, where US/JP/EU doesn't use a normal
        # header for sample banks but instead has a mapping from sound bank to
        # sample bank offset/length. Shindou uses a normal header and makes the
        # mapping part of the sound bank header instead (part of entry_meta).
        for i in range(len(entries)):
            ser.add(pack("PIH", entry_offsets[i], entry_lens[i], sh_magic))
            ser.add(entry_meta[i] or b"\0\0\0\0")
            ser.align(16)

        if out_header_filename:
            with open(out_header_filename, "wb") as f:
                f.write(ser.finish())
        with open(out_filename, "wb") as f:
            f.write(data)

    else:
        ser = ReserveSerializer()
        ser.add(pack("HHX", magic, len(entry_list)))
        table = ser.reserve(len(entry_list) * 2 * WORD_BYTES)
        ser.align(16)
        data_start = ser.size

        ser.add(data)
        if extra_padding:
            ser.add(b"\0")
        ser.align(64)

        for index in entry_list:
            table.append(pack("P", entry_offsets[index] + data_start))
            table.append(pack("IX", entry_lens[index]))

        data = ser.finish()
        compress = False

        if out_filename.endswith('sound_data.tbl'):
            data = data[:entry_offsets[0] + data_start] # remove the fake data
            out_filename = 'sound/sound_data_compressed.tbl'
            compress = True

        if out_filename.endswith('sequences.bin'):
            data = data[:entry_offsets[1] + data_start] # remove the fake data
            out_filename = 'sound/sequences_compressed.bin'
            compress = True

        if out_filename.endswith('sound_data.ctl'):
            out_filename = 'sound/sound_data_compressed.ctl'
            compress = True

        with open(out_filename, "wb") as f:
            if compress:
                f.write(zlib.compress(data))
            else:
                f.write(data)

def validate_and_normalize_sequence_json(json, bank_names, defines):
    validate(isinstance(json, dict), "must have a top-level object")
    if "comment" in json:
        del json["comment"]
    for key, seq in json.items():
        if isinstance(seq, dict):
            validate_json_format(seq, {"ifdef": list, "banks": list}, key)
            validate(
                all(isinstance(x, str) for x in seq["ifdef"]),
                '"ifdef" must be an array of strings',
                key,
            )
            if all(d not in defines for d in seq["ifdef"]):
                seq = None
            else:
                seq = seq["banks"]
            json[key] = seq
        if isinstance(seq, list):
            for x in seq:
                validate(
                    isinstance(x, str), "bank list must be an array of strings", key
                )
                validate(
                    x in bank_names, "reference to non-existing sound bank " + x, key
                )
        else:
            validate(seq is None, "bad JSON type, expected null, array or object", key)


def write_sequences(
    inputs,
    out_filename,
    out_header_filename,
    out_bank_sets,
    sound_bank_dir,
    seq_json,
    defines,
    is_shindou,
):
    bank_names = sorted(
        [os.path.splitext(os.path.basename(x))[0] for x in os.listdir(sound_bank_dir)]
    )

    try:
        with open(seq_json, "r") as inf:
            data = inf.read()
            data = strip_comments(data)
            json = orderedJsonDecoder.decode(data)
            validate_and_normalize_sequence_json(json, bank_names, defines)

    except Exception as e:
        fail("failed to parse " + str(seq_json) + ": " + str(e))

    inputs.sort(key=lambda f: os.path.basename(f))
    name_to_fname = {}
    for fname in inputs:
        name = os.path.splitext(os.path.basename(fname))[0]
        if name in name_to_fname:
            fail(
                "Files "
                + fname
                + " and "
                + name_to_fname[name]
                + " conflict. Remove one of them."
            )
        name_to_fname[name] = fname
        if name not in json:
            fail(
                "Sequence file " + fname + " is not mentioned in sequences.json. "
                "Either assign it a list of sound banks, or set it to null to "
                "explicitly leave it out from the build."
            )

    for key, seq in json.items():
        if key not in name_to_fname and seq is not None:
            fail(
                "sequences.json assigns sound banks to "
                + key
                + ", but there is no such sequence file. Either remove the entry (or "
                "set it to null), or create sound/sequences/" + key + ".m64."
            )

    ind_to_name = []
    for key in json:
        ind = int(key.split("_")[0], 16)
        while len(ind_to_name) <= ind:
            ind_to_name.append(None)
        if ind_to_name[ind] is not None:
            fail(
                "Sequence files "
                + key
                + " and "
                + ind_to_name[ind]
                + " have the same index. Renumber or delete one of them."
            )
        ind_to_name[ind] = key

    while ind_to_name and json.get(ind_to_name[-1]) is None:
        ind_to_name.pop()

    def serialize_file(name, ser, is_shindou, asset_offsets):
        if json.get(name) is None:
            return
        ser.reset_garbage_pos()
        fname = name_to_fname[name]
        if name == '00_sound_player':
            with open(fname, "rb") as f:
                ser.add(f.read())
        else:
            length = 0
            with open(fname, "rb") as f:
                length = len(f.read())
            asset_offsets[fname] = ser.size
            fake_data = list(name)
            fake_data = [ord(char) for char in fake_data]
            padding = [0] * (length - len(fake_data))
            fake_data = fake_data + padding
            ser.add(bytes(fake_data))
        if is_shindou and name.startswith("17"):
            ser.align(16)
        else:
            ser.align_garbage(16)

    serialize_seqfile(
        out_filename,
        out_header_filename,
        ind_to_name,
        serialize_file,
        range(len(ind_to_name)),
        TYPE_SEQ,
        is_shindou,
        extra_padding=False,
    )

    compress = True
    out_bank_sets = 'sound/bank_sets_compressed'
    with open(out_bank_sets, "wb") as f:
        ser = ReserveSerializer()
        table = ser.reserve(len(ind_to_name) * 2)
        for name in ind_to_name:
            bank_set = json.get(name) or []
            table.append(pack("H", ser.size))
            ser.add(bytes([len(bank_set)]))
            for bank in bank_set[::-1]:
                ser.add(bytes([bank_names.index(bank)]))
        ser.align(16)
        data = ser.finish()
        if compress:
            f.write(zlib.compress(data))
        else:
            f.write(data)

def main():
    global STACK_TRACES
    global DUMP_INDIVIDUAL_BINS
    global ENDIAN_MARKER
    global WORD_BYTES
    need_help = False
    skip_next = 0
    cpp_command = None
    print_samples = False
    sequences_out_file = None
    sequences_header_out_file = None
    defines = []
    args = []
    for i, a in enumerate(sys.argv[1:], 1):
        if skip_next > 0:
            skip_next -= 1
            continue
        if a == "--help" or a == "-h":
            need_help = True
        elif a == "--cpp":
            cpp_command = sys.argv[i + 1]
            skip_next = 1
        elif a == "-D":
            defines.append(sys.argv[i + 1])
            skip_next = 1
        elif a == "--endian":
            endian = sys.argv[i + 1]
            if endian == "big":
                ENDIAN_MARKER = ">"
            elif endian == "little":
                ENDIAN_MARKER = "<"
            elif endian == "native":
                ENDIAN_MARKER = "="
            else:
                fail("--endian takes argument big, little or native")
            skip_next = 1
        elif a == "--bitwidth":
            bitwidth = sys.argv[i + 1]
            if bitwidth == "native":
                WORD_BYTES = struct.calcsize("P")
            else:
                if bitwidth not in ["32", "64"]:
                    fail("--bitwidth takes argument 32, 64 or native")
                WORD_BYTES = int(bitwidth) // 8
            skip_next = 1
        elif a.startswith("-D"):
            defines.append(a[2:])
        elif a == "--stack-trace":
            STACK_TRACES = True
        elif a == "--dump-individual-bins":
            DUMP_INDIVIDUAL_BINS = True
        elif a == "--print-samples":
            print_samples = True
        elif a == "--sequences":
            sequences_out_file = sys.argv[i + 1]
            sequences_header_out_file = sys.argv[i + 2]
            bank_sets_out_file = sys.argv[i + 3]
            sound_bank_dir = sys.argv[i + 4]
            sequence_json = sys.argv[i + 5]
            skip_next = 5
        elif a.startswith("-"):
            print("Unrecognized option " + a)
            sys.exit(1)
        else:
            args.append(a)

    defines_set = {d.split("=")[0] for d in defines}
    is_shindou = "VERSION_SH" in defines_set

    if sequences_out_file is not None and not need_help:
        write_sequences(
            args,
            sequences_out_file,
            sequences_header_out_file,
            bank_sets_out_file,
            sound_bank_dir,
            sequence_json,
            defines_set,
            is_shindou,
        )
        sys.exit(0)

    if need_help or len(args) != 6:
        print(
            "Usage: {} <samples dir> <sound bank dir>"
            " <out .ctl file> <out .ctl Shindou header file>"
            " <out .tbl file> <out .tbl Shindou header file>"
            " [--cpp <preprocessor>]"
            " [-D <symbol>]"
            " [--stack-trace]"
            " | --sequences <out sequence .bin> <out Shindou sequence header .bin> "
            "<out bank sets .bin> <sound bank dir> <sequences.json> <inputs...>".format(
                sys.argv[0]
            )
        )
        sys.exit(0 if need_help else 1)

    sample_bank_dir = args[0]
    sound_bank_dir = args[1]
    ctl_data_out = args[2]
    ctl_data_header_out = args[3]
    tbl_data_out = args[4]
    tbl_data_header_out = args[5]

    banks = []
    sample_banks = []
    name_to_sample_bank = {}

    sample_bank_names = sorted(os.listdir(sample_bank_dir))
    for name in sample_bank_names:
        dir = os.path.join(sample_bank_dir, name)
        if not os.path.isdir(dir):
            continue
        entries = []
        for f in sorted(os.listdir(dir)):
            fname = os.path.join(dir, f)
            if not f.endswith(".aifc"):
                continue
            try:
                with open(fname, "rb") as inf:
                    data = inf.read()
                    entries.append(parse_aifc(data, f[:-5], fname))
            except Exception as e:
                fail("malformed AIFC file " + fname + ": " + str(e))
        if entries:
            sample_bank = SampleBank(name, entries)
            sample_banks.append(sample_bank)
            name_to_sample_bank[name] = sample_bank

    bank_names = sorted(os.listdir(sound_bank_dir))
    for f in bank_names:
        fname = os.path.join(sound_bank_dir, f)
        if not f.endswith(".json"):
            continue

        try:
            if cpp_command:
                data = subprocess.run(
                    [cpp_command, fname] + ["-D" + x for x in defines],
                    stdout=subprocess.PIPE,
                    check=True,
                ).stdout.decode()
            else:
                with open(fname, "r") as inf:
                    data = inf.read()
                data = strip_comments(data)
            bank_json = orderedJsonDecoder.decode(data)

            bank_json = apply_ifs(bank_json, defines_set)
            validate_bank_toplevel(bank_json)
            apply_version_diffs(bank_json, defines_set)
            normalize_sound_json(bank_json)

            sample_bank_name = bank_json["sample_bank"]
            validate(
                sample_bank_name in name_to_sample_bank,
                "sample bank " + sample_bank_name + " not found",
            )
            sample_bank = name_to_sample_bank[sample_bank_name]

            validate_bank(bank_json, sample_bank)

            bank = Bank(f[:-5], sample_bank, bank_json)
            mark_sample_bank_uses(bank)
            banks.append(bank)

        except Exception as e:
            fail("failed to parse bank " + fname + ": " + str(e))

    sample_banks = [b for b in sample_banks if b.uses]
    sample_banks.sort(key=lambda b: b.uses[0].name)
    sample_bank_index = 0
    for sample_bank in sample_banks:
        sample_bank.index = sample_bank_index
        sample_bank_index += 1

    serialize_seqfile(
        tbl_data_out,
        tbl_data_header_out,
        sample_banks,
        serialize_tbl,
        [x.sample_bank.index for x in banks],
        TYPE_TBL,
        is_shindou,
    )

    if DUMP_INDIVIDUAL_BINS:
        # Debug logic, may simplify diffing
        os.makedirs("ctl/", exist_ok=True)
        for b in banks:
            with open("ctl/" + b.name + ".bin", "wb") as f:
                ser = GarbageSerializer()
                serialize_ctl(b, ser, is_shindou)
                f.write(ser.finish())
        print("wrote to ctl/")

    serialize_seqfile(
        ctl_data_out,
        ctl_data_header_out,
        banks,
        serialize_ctl,
        list(range(len(banks))),
        TYPE_CTL,
        is_shindou,
    )

    if print_samples:
        for sample_bank in sample_banks:
            for entry in sample_bank.entries:
                if entry.used:
                    print(entry.fname)


if __name__ == "__main__":
    main()