mirror of
https://github.com/irmen/prog8.git
synced 2024-12-28 12:32:34 +00:00
301 lines
12 KiB
Python
301 lines
12 KiB
Python
import struct
|
|
from typing import Sequence
|
|
|
|
# Chunk types:
|
|
# user types: 0 - 239
|
|
# reserved: 240 - 248
|
|
CHUNK_DUMMY = 249
|
|
CHUNK_BONKRAM = 250
|
|
CHUNK_SYSTEMRAM = 251
|
|
CHUNK_VIDEORAM = 252
|
|
CHUNK_PAUSE = 253
|
|
CHUNK_EOF = 254
|
|
CHUNK_IGNORE = 255
|
|
|
|
ChunksWithData = {CHUNK_DUMMY, CHUNK_SYSTEMRAM, CHUNK_VIDEORAM, CHUNK_BONKRAM}
|
|
|
|
|
|
class LoadList:
|
|
def __init__(self):
|
|
self.data = bytearray([CHUNK_IGNORE] * 256)
|
|
self.data[0:4] = struct.pack("ccBB", b'L', b'L', 1, 0)
|
|
self.index = 4
|
|
|
|
def __eq__(self, other) -> bool:
|
|
return isinstance(other, LoadList) and self.data == other.data
|
|
|
|
def add_chunk(self, chunktype: int, size: int, bank: int = 0, address: int = 0) -> None:
|
|
if chunktype < 0 or chunktype > 255:
|
|
raise ValueError("chunktype must be 0 - 255")
|
|
if size < 0 or size > 65535:
|
|
raise ValueError(f"size must be 0 - 65535 bytes")
|
|
if bank < 0 or bank > 255:
|
|
raise ValueError("bank must be 0 - 255")
|
|
if address < 0 or address > 65535:
|
|
raise ValueError("address must be 0 - 65535")
|
|
data = struct.pack("<BHBH", chunktype, size, bank, address)
|
|
if self.index + len(data) > 255:
|
|
raise IndexError("too many chunks")
|
|
self.data[self.index: self.index + len(data)] = data
|
|
self.index += len(data)
|
|
|
|
def read(self, data: bytes) -> None:
|
|
if len(data) != 256:
|
|
raise ValueError("data must be 256 bytes long")
|
|
self.data[0:256] = data
|
|
self.index = 256
|
|
|
|
def parse(self) -> Sequence[tuple]:
|
|
if self.data[0] != ord('L') or self.data[1] != ord('L') or self.data[2] != 1:
|
|
raise ValueError("invalid loadlist identifier", self.data[:4])
|
|
index = 4
|
|
chunks = []
|
|
while index < 256:
|
|
chunktype = self.data[index]
|
|
if chunktype == CHUNK_IGNORE:
|
|
index += 1 # pad byte
|
|
elif chunktype == CHUNK_EOF:
|
|
chunks.append((CHUNK_EOF, 0, 0, 0))
|
|
return chunks
|
|
else:
|
|
size, bank, address = struct.unpack("<HBH", self.data[index + 1:index + 6])
|
|
chunks.append((chunktype, size, bank, address))
|
|
index += 6
|
|
return chunks
|
|
|
|
def is_empty(self) -> bool:
|
|
return self.index <= 4
|
|
|
|
|
|
class MultiChunkFile:
|
|
def __init__(self):
|
|
self.chunks = []
|
|
self.loadlist = LoadList()
|
|
self.datachunks = []
|
|
|
|
def read(self, filename: str) -> None:
|
|
num_loadlist = 0
|
|
num_data = 0
|
|
data_total = 0
|
|
file_total = 0
|
|
with open(filename, "rb") as inf:
|
|
while True:
|
|
data = inf.read(256)
|
|
if len(data) == 0:
|
|
break
|
|
loadlist = LoadList()
|
|
loadlist.read(data)
|
|
self.chunks.append(loadlist)
|
|
num_loadlist += 1
|
|
file_total += len(data)
|
|
for chunk in loadlist.parse():
|
|
if chunk[0] == CHUNK_EOF:
|
|
break
|
|
elif chunk[0] in ChunksWithData or chunk[0] < 240:
|
|
data = inf.read(chunk[1])
|
|
self.chunks.append(data)
|
|
num_data += 1
|
|
data_total += len(data)
|
|
file_total += len(data)
|
|
self.validate()
|
|
print("Read", filename)
|
|
print(f" {num_loadlist} loadlists, {num_data} data chunks")
|
|
print(f" total data size: {data_total} (${data_total:x})")
|
|
print(f" total file size: {file_total} (${file_total:x})")
|
|
|
|
def write(self, filename: str) -> None:
|
|
self.flush_ll()
|
|
self.validate()
|
|
num_loadlist = 0
|
|
num_data = 0
|
|
data_total = 0
|
|
file_total = 0
|
|
with open(filename, "wb") as out:
|
|
for chunk in self.chunks:
|
|
if isinstance(chunk, LoadList):
|
|
num_loadlist += 1
|
|
out.write(chunk.data)
|
|
file_total += len(chunk.data)
|
|
elif isinstance(chunk, bytes):
|
|
num_data += 1
|
|
out.write(chunk)
|
|
file_total += len(chunk)
|
|
data_total += len(chunk)
|
|
else:
|
|
raise TypeError("invalid chunk type")
|
|
print("Written", filename)
|
|
print(f" {num_loadlist} loadlists, {num_data} data chunks")
|
|
print(f" total data size: {data_total} (${data_total:x})")
|
|
print(f" total file size: {file_total} (${file_total:x})")
|
|
|
|
def validate(self) -> None:
|
|
if len(self.chunks) < 2:
|
|
raise ValueError("must be at least 2 chunks", self.chunks)
|
|
chunk_iter = iter(self.chunks)
|
|
eof_found = False
|
|
while not eof_found:
|
|
try:
|
|
chunk = next(chunk_iter)
|
|
except StopIteration:
|
|
raise ValueError("missing EOF chunk")
|
|
else:
|
|
if isinstance(chunk, LoadList):
|
|
for lc in chunk.parse():
|
|
if lc[0] == CHUNK_EOF:
|
|
eof_found = True
|
|
break
|
|
elif lc[0] in ChunksWithData or lc[0] < 240:
|
|
size, bank, address = lc[1:]
|
|
data = next(chunk_iter)
|
|
if isinstance(data, bytes):
|
|
if len(data) != size:
|
|
raise ValueError("data chunk size mismatch")
|
|
else:
|
|
raise TypeError("expected data chunk")
|
|
else:
|
|
raise TypeError("expected LoadList chunk")
|
|
try:
|
|
next(chunk_iter)
|
|
except StopIteration:
|
|
pass
|
|
else:
|
|
raise ValueError("trailing chunks")
|
|
|
|
def add_Dummy(self, size: int) -> None:
|
|
self.add_chunk(CHUNK_DUMMY, data=bytearray(size))
|
|
|
|
def add_SystemRam(self, bank: int, address: int, data: bytes, chunksize: int = 0xfe00) -> None:
|
|
if address >= 0xa000 and address < 0xc000:
|
|
raise ValueError("use add_BankedRam instead to load chunks into banked ram $a000-$c000")
|
|
if bank < 0 or bank > 31:
|
|
raise ValueError("bank must be 0 - 31")
|
|
while data:
|
|
if address >= 65536:
|
|
raise ValueError("data too large for system ram")
|
|
self.add_chunk(CHUNK_SYSTEMRAM, bank, address, data[:chunksize])
|
|
data = data[chunksize:]
|
|
address += chunksize
|
|
|
|
def add_BankedRam(self, bank: int, address: int, data: bytes, chunksize: int = 0x2000) -> None:
|
|
if address < 0xa000 or address >= 0xc000:
|
|
raise ValueError("use add_SystemRam instead to load chunks into normal system ram")
|
|
if chunksize > 0x2000:
|
|
raise ValueError("chunksize too large for banked ram, max 8K")
|
|
while data:
|
|
if address >= 0xc000:
|
|
address -= 0x2000
|
|
bank += 1
|
|
if bank >= 32:
|
|
raise ValueError("data too large for banked ram")
|
|
self.add_chunk(CHUNK_SYSTEMRAM, bank, address, data[:chunksize])
|
|
data = data[chunksize:]
|
|
address += chunksize
|
|
|
|
def add_BonkRam(self, bonk: int, address: int, data: bytes, chunksize: int = 0x4000) -> None:
|
|
if bonk < 32 or bonk > 255:
|
|
raise ValueError("bank for bonk ram (cartridge ram) must be 32 - 255")
|
|
if chunksize > 0x4000:
|
|
raise ValueError("chunksize too large for bonk ram (cartridge ram), max 16K")
|
|
if address < 0xc000 or address > 0xffff:
|
|
raise ValueError("use add_SystemRam instead to load chunks into normal system ram")
|
|
while data:
|
|
if address > 0xffff:
|
|
address -= 0x4000
|
|
bonk += 1
|
|
if bonk > 255:
|
|
raise ValueError("data too large for bonk ram (cartridge ram)")
|
|
self.add_chunk(CHUNK_BONKRAM, bonk, address, data[:chunksize])
|
|
data = data[chunksize:]
|
|
address += chunksize
|
|
|
|
def add_VideoRam(self, bank: int, address: int, data: bytes, chunksize: int = 0xfe00) -> None:
|
|
if bank < 0 or bank > 1:
|
|
raise ValueError("bank for videoram must be 0 or 1")
|
|
while data:
|
|
if address >= 65536:
|
|
address -= 65536
|
|
bank += 1
|
|
if bank >= 2:
|
|
raise ValueError("data too large for video ram")
|
|
self.add_chunk(CHUNK_VIDEORAM, bank, address, data[:chunksize])
|
|
data = data[chunksize:]
|
|
address += chunksize
|
|
|
|
def add_User(self, chunktype: int, data: bytes) -> None:
|
|
if chunktype < 0 or chunktype > 239:
|
|
raise ValueError("user chunk type must be 0 - 239")
|
|
if len(data) > 65535:
|
|
raise ValueError("data too large to fit in a single chunk")
|
|
self.add_chunk(chunktype, data=data)
|
|
|
|
def add_Pause(self, code: int) -> None:
|
|
self.add_chunk_nodata(CHUNK_PAUSE, code)
|
|
|
|
def add_EOF(self) -> None:
|
|
self.add_chunk_nodata(CHUNK_EOF, 0)
|
|
|
|
def add_chunk(self, chunktype: int, bank: int = 0, address: int = 0, data: bytes = b"") -> None:
|
|
try:
|
|
self.loadlist.add_chunk(chunktype, len(data), bank, address)
|
|
except IndexError:
|
|
# loadlist is full, flush everything out and create a new one and put it in there.
|
|
self.flush_ll()
|
|
self.loadlist.add_chunk(chunktype, len(data), bank, address)
|
|
if data:
|
|
self.datachunks.append(bytes(data))
|
|
|
|
def add_chunk_nodata(self, chunktype: int, code: int = 0) -> None:
|
|
try:
|
|
self.loadlist.add_chunk(chunktype, code, 0, 0)
|
|
except IndexError:
|
|
# loadlist is full, flush everything out and create a new one and put it in there.
|
|
self.flush_ll()
|
|
self.loadlist.add_chunk(chunktype, code, 0, 0)
|
|
|
|
def flush_ll(self) -> None:
|
|
if not self.loadlist.is_empty():
|
|
self.chunks.append(self.loadlist)
|
|
self.chunks.extend(self.datachunks)
|
|
self.loadlist = LoadList()
|
|
self.datachunks.clear()
|
|
|
|
|
|
if __name__ == "__main__":
|
|
try:
|
|
bitmap1 = open("testdata/ME-TITLESCREEN.BIN", "rb").read()
|
|
bitmap2 = open("testdata/DS-TITLESCREEN.BIN", "rb").read()
|
|
palette1 = open("testdata/ME-TITLESCREEN.PAL", "rb").read()
|
|
palette2 = open("testdata/DS-TITLESCREEN.PAL", "rb").read()
|
|
except IOError:
|
|
print("""ERROR: cannot load the demo data files.
|
|
You'll need to put the titlescreen data files from the 'musicdemo' project into the testdata directory.
|
|
The musicdemo is on github: https://github.com/irmen/cx16musicdemo
|
|
The four files are the two ME- and the two DS- TITLESCREEN.BIN and .PAL files, and are generated by running the makefile in that project.""")
|
|
raise SystemExit
|
|
|
|
program = bytearray(3333)
|
|
textdata = b"hello this is a demo text.... \x00"
|
|
|
|
mcf = MultiChunkFile()
|
|
for _ in range(4):
|
|
mcf.add_User(42, bytearray(999))
|
|
mcf.add_SystemRam(0, 0x4000, program)
|
|
mcf.add_BankedRam(10, 0xa000, textdata)
|
|
mcf.add_Pause(444)
|
|
for _ in range(4):
|
|
mcf.add_User(42, bytearray(999))
|
|
mcf.add_VideoRam(1, 0xfa00, palette1)
|
|
mcf.add_VideoRam(0, 0, bitmap1)
|
|
mcf.add_Pause(333)
|
|
mcf.add_VideoRam(1, 0xfa00, palette2)
|
|
mcf.add_VideoRam(0, 0, bitmap2)
|
|
mcf.add_Pause(222)
|
|
mcf.add_BonkRam(32, 0xc000, bytearray(32768))
|
|
mcf.add_Pause(111)
|
|
mcf.add_EOF()
|
|
mcf.write("demo.mcf")
|
|
print("Verifying file...")
|
|
mcf2 = MultiChunkFile()
|
|
mcf2.read("demo.mcf")
|
|
assert mcf2.chunks == mcf.chunks
|