Refactor into separate source files
This commit is contained in:
parent
41480ac025
commit
c311171270
|
@ -0,0 +1,163 @@
|
|||
# Singular values may be decimal (no prefix), or any of the prefixes python accepts normally (0x for hex, 0b for binary, 0o for octal)
|
||||
# Additionally, hexadecimals may be prefixed with '#' or '$', or suffixed with 'h', e.g. 0x10 #10 $10 10h are all parsed as 16
|
||||
# For nested IDs, the format is ONLY like IP addresses:
|
||||
# decimal with '.' separator, e.g. 16.127.1
|
||||
# hexadecimal with ':' separator, e.g. 10:7f:1
|
||||
# Nested IDs do not support prefixes.
|
||||
|
||||
def try_int(v):
|
||||
try:
|
||||
if v[0] in '#$': # Maybe % too?
|
||||
return int(v[1:], 16)
|
||||
if v[-1] == 'h':
|
||||
return int(v[:-1], 16)
|
||||
return int(v, 0)
|
||||
except:
|
||||
if v == '':
|
||||
return None
|
||||
return v
|
||||
|
||||
|
||||
def get_max_number_width(container, fmt: str = 'd') -> int:
|
||||
return len(f'{len(container)-1:{fmt}}')
|
||||
|
||||
|
||||
def get_number_zero_pad(container, fmt: str = 'd') -> str:
|
||||
max_digits = len(f'{len(container)-1:{fmt}}') # Could instead call get_max_number_width
|
||||
return f'0{max_digits}{fmt}'
|
||||
|
||||
|
||||
def encode_nested_ids(values: list[int], max_digits: list[int] = None, fmt: str = 'd') -> str:
|
||||
delimiter = ':' if fmt in 'Xx' else '.'
|
||||
if max_digits:
|
||||
return delimiter.join([f'{value:0{digits}{fmt}}' for value, digits in zip(values, max_digits)])
|
||||
else:
|
||||
return delimiter.join([f'{value:{fmt}}' for value in values])
|
||||
|
||||
|
||||
def decode_nested_ids(string: str) -> list[int]:
|
||||
hex = ':' in string
|
||||
delimiter = ':' if hex else '.'
|
||||
return [int(i, 16 if hex else 10) for i in string.split(delimiter)]
|
||||
|
||||
|
||||
def flatten_keys(container: dict | list, prefix: str = '') -> dict:
|
||||
output = {}
|
||||
|
||||
def flatten_item(k: str, v):
|
||||
if isinstance(v, dict) or isinstance(v, list):
|
||||
flat = flatten_keys(v, f'{prefix}{k}.')
|
||||
for k2, v2 in flat.items():
|
||||
output[k2] = v2
|
||||
else:
|
||||
output[f'{prefix}{k}'] = v
|
||||
|
||||
if isinstance(container, list):
|
||||
fmt = get_number_zero_pad(container, 'd') # Zero pad all of the indices to the same decimal string length as the final one
|
||||
for k, v in enumerate(container):
|
||||
flatten_item(f'{k:{fmt}}', v)
|
||||
elif isinstance(container, dict):
|
||||
for k, v in container.items():
|
||||
flatten_item(k, v)
|
||||
else:
|
||||
raise ValueError(f'flatten_keys is undefined for container type "{container}"')
|
||||
|
||||
return output
|
||||
|
||||
|
||||
def flatten_table(table: list, id_fmt: str = 'x') -> list:
|
||||
if len(table) < 1:
|
||||
return table # Empty
|
||||
if isinstance(table[0], dict): # A simple table
|
||||
return [flatten_keys(d) for d in table]
|
||||
if isinstance(table[0], list): # Nested lists are bad when expanded as columns, so we'll expand
|
||||
flattened_table = []
|
||||
|
||||
def flatten_list(data, ids: list[int], id_max_digits: list[int]) -> None:
|
||||
if isinstance(data, list):
|
||||
max_digits = id_max_digits + [get_max_number_width(data, id_fmt)]
|
||||
for id, sub in enumerate(data):
|
||||
flatten_list(sub, ids + [id], max_digits)
|
||||
else:
|
||||
entry = {'ID': encode_nested_ids(ids, id_max_digits, id_fmt)}
|
||||
entry.update(flatten_keys(data))
|
||||
flattened_table.append(entry)
|
||||
|
||||
flatten_list(table, [], [])
|
||||
return flattened_table
|
||||
else:
|
||||
raise NotImplementedError(table[0])
|
||||
|
||||
|
||||
def unflatten_keys(d: dict) -> dict:
|
||||
output = {}
|
||||
for k, v in d.items():
|
||||
keysplit = k.split('.')
|
||||
target_dict = output
|
||||
for prefix in keysplit[:-1]:
|
||||
if prefix not in target_dict:
|
||||
target_dict[prefix] = {}
|
||||
target_dict = target_dict[prefix]
|
||||
target_dict[k] = v
|
||||
return output
|
||||
|
||||
|
||||
def unflatten_table(headers: list[str], entries: list):
|
||||
if 'ID' not in headers:
|
||||
return entries
|
||||
# This could be an array of an array of an array of an...
|
||||
id0 = entries[0]['ID']
|
||||
if '.' not in id0 and ':' not in id0:
|
||||
return entries
|
||||
# Treat this as a nested array
|
||||
table = {tuple(decode_nested_ids(entry['ID'])): entry for entry in entries}
|
||||
|
||||
output = []
|
||||
def unflatten_arrays(id_split: tuple[int], cur_array: list, value):
|
||||
i, *remainder = id_split
|
||||
if len(remainder) > 0:
|
||||
while len(cur_array) <= i: # Make sure our array has the index we're about to jump into
|
||||
cur_array.append([])
|
||||
unflatten_arrays(remainder, cur_array[i], value)
|
||||
else:
|
||||
while len(cur_array) <= i: # Make sure our array has the index we're about to overwrite
|
||||
cur_array.append(None)
|
||||
cur_array[i] = value
|
||||
|
||||
for id_split in sorted(table.keys()):
|
||||
unflatten_arrays(id_split, output, table[id_split])
|
||||
|
||||
return output
|
||||
|
||||
|
||||
def dump_tsv(filename, table, id_column=True) -> None:
|
||||
table_flat = flatten_table(table)
|
||||
|
||||
with open(filename, 'w') as file:
|
||||
headers = list(table_flat[0].keys())
|
||||
if id_column and 'ID' not in headers: # Some flattened tables build their own ID column!
|
||||
# See how long the hex representation of the last number will be, so we can zero-pad the rest to match.
|
||||
fmt = get_number_zero_pad(table_flat, 'X')
|
||||
file.write('\t'.join(['ID'] + headers) + '\n')
|
||||
for i, entry in enumerate(table_flat):
|
||||
file.write('\t'.join([f'0x{i:{fmt}}'] + [str(entry[key]) for key in headers]) + '\n')
|
||||
else:
|
||||
file.write('\t'.join(headers) + '\n')
|
||||
for i, entry in enumerate(table_flat):
|
||||
file.write('\t'.join([str(entry[key]) for key in headers]) + '\n')
|
||||
|
||||
|
||||
def load_tsv(filename) -> list:
|
||||
with open(filename, 'r') as file:
|
||||
lines = file.read().rstrip().split('\n')
|
||||
if len(lines) < 2:
|
||||
return []
|
||||
headers = lines[0].split('\t')
|
||||
|
||||
# Simple line-by-line unflatten
|
||||
entries = []
|
||||
for line in lines[1:]:
|
||||
entry = {key: try_int(value) for key, value in zip(headers, line.split('\t'))}
|
||||
entries.append(unflatten_keys(entry))
|
||||
|
||||
return unflatten_table(headers, entries)
|
|
@ -0,0 +1,47 @@
|
|||
from ChocolateBirdData.reference_implementation import get_base_structarraytypes, parse_struct_definitions_from_tsv_filename, get_structarraytype, LeftoverBits, ReadBuffer, WriteBuffer
|
||||
from includes.helpers import load_tsv
|
||||
|
||||
class ROMHandler:
|
||||
offset_key: str
|
||||
struct_definitions: dict
|
||||
|
||||
def extract(self, table: str, in_buffer) -> list[dict]:
|
||||
# Deserialize a table
|
||||
leftover_bits = LeftoverBits()
|
||||
entry = self.addresses[table] # Remember to try/catch
|
||||
offset = entry[self.offset_key]
|
||||
buf = ReadBuffer(in_buffer, offset)
|
||||
return get_structarraytype(entry['format'], self.struct_definitions).get_value(buf, leftover_bits)
|
||||
|
||||
def build(self, table: str, new_data: list[dict], out_buffer):
|
||||
# Serialize complete data. This WILL fail if the input data is incomplete.
|
||||
leftover_bits = LeftoverBits()
|
||||
entry = self.addresses[table] # Remember to try/catch
|
||||
offset = entry[self.offset_key]
|
||||
buf = WriteBuffer(out_buffer, offset)
|
||||
get_structarraytype(entry['format'], self.struct_definitions).put_value(buf, new_data, leftover_bits)
|
||||
|
||||
def build_partial(self, table: str, new_data: list[dict], in_buffer, out_buffer):
|
||||
# Safely merge partial data over the existing data, then serialize it.
|
||||
existing_data = self.extract(table, in_buffer)
|
||||
for i, new in enumerate(new_data):
|
||||
id = new.get('ID', i)
|
||||
for k, v in new.items():
|
||||
if k != 'ID' and v is not None: # Allow holes in the table for values we don't care about overwriting
|
||||
existing_data[id][k] = v
|
||||
self.build(table, existing_data, out_buffer)
|
||||
|
||||
|
||||
def load_ff5_snes_struct_definitions() -> dict:
|
||||
existing_structs = get_base_structarraytypes()
|
||||
parse_struct_definitions_from_tsv_filename('ChocolateBirdData/structs_SNES_stubs.tsv', existing_structs)
|
||||
parse_struct_definitions_from_tsv_filename('ChocolateBirdData/5/structs/SNES_stubs.tsv', existing_structs)
|
||||
parse_struct_definitions_from_tsv_filename('ChocolateBirdData/5/structs/SNES.tsv', existing_structs)
|
||||
parse_struct_definitions_from_tsv_filename('ChocolateBirdData/5/structs/SNES_save.tsv', existing_structs)
|
||||
return existing_structs
|
||||
|
||||
|
||||
class FF5SNESHandler(ROMHandler):
|
||||
offset_key: str = 'SNES'
|
||||
struct_definitions: dict = load_ff5_snes_struct_definitions()
|
||||
addresses: dict = {entry['Label']: entry for entry in load_tsv('ChocolateBirdData/5/addresses_SNES_PSX.tsv')}
|
208
tabcomp.py
208
tabcomp.py
|
@ -1,209 +1,5 @@
|
|||
from ChocolateBirdData.reference_implementation import get_base_structarraytypes, parse_struct_definitions_from_tsv_filename, get_structarraytype, LeftoverBits, ReadBuffer, WriteBuffer
|
||||
|
||||
# Singular values may be decimal (no prefix), or any of the prefixes python accepts normally (0x for hex, 0b for binary, 0o for octal)
|
||||
# Additionally, hexadecimals may be prefixed with '#' or '$', or suffixed with 'h', e.g. 0x10 #10 $10 10h are all parsed as 16
|
||||
# For nested IDs, the format is ONLY like IP addresses:
|
||||
# decimal with '.' separator, e.g. 16.127.1
|
||||
# hexadecimal with ':' separator, e.g. 10:7f:1
|
||||
# Nested IDs do not support prefixes.
|
||||
|
||||
def try_int(v):
|
||||
try:
|
||||
if v[0] in '#$': # Maybe % too?
|
||||
return int(v[1:], 16)
|
||||
if v[-1] == 'h':
|
||||
return int(v[:-1], 16)
|
||||
return int(v, 0)
|
||||
except:
|
||||
if v == '':
|
||||
return None
|
||||
return v
|
||||
|
||||
|
||||
def get_max_number_width(container, fmt: str = 'd') -> int:
|
||||
return len(f'{len(container)-1:{fmt}}')
|
||||
|
||||
|
||||
def get_number_zero_pad(container, fmt: str = 'd') -> str:
|
||||
max_digits = len(f'{len(container)-1:{fmt}}') # Could instead call get_max_number_width
|
||||
return f'0{max_digits}{fmt}'
|
||||
|
||||
|
||||
def encode_nested_ids(values: list[int], max_digits: list[int] = None, fmt: str = 'd') -> str:
|
||||
delimiter = ':' if fmt in 'Xx' else '.'
|
||||
if max_digits:
|
||||
return delimiter.join([f'{value:0{digits}{fmt}}' for value, digits in zip(values, max_digits)])
|
||||
else:
|
||||
return delimiter.join([f'{value:{fmt}}' for value in values])
|
||||
|
||||
|
||||
def decode_nested_ids(string: str) -> list[int]:
|
||||
hex = ':' in string
|
||||
delimiter = ':' if hex else '.'
|
||||
return [int(i, 16 if hex else 10) for i in string.split(delimiter)]
|
||||
|
||||
|
||||
def flatten_keys(container: dict | list, prefix: str = '') -> dict:
|
||||
output = {}
|
||||
|
||||
def flatten_item(k: str, v):
|
||||
if isinstance(v, dict) or isinstance(v, list):
|
||||
flat = flatten_keys(v, f'{prefix}{k}.')
|
||||
for k2, v2 in flat.items():
|
||||
output[k2] = v2
|
||||
else:
|
||||
output[f'{prefix}{k}'] = v
|
||||
|
||||
if isinstance(container, list):
|
||||
fmt = get_number_zero_pad(container, 'd') # Zero pad all of the indices to the same decimal string length as the final one
|
||||
for k, v in enumerate(container):
|
||||
flatten_item(f'{k:{fmt}}', v)
|
||||
elif isinstance(container, dict):
|
||||
for k, v in container.items():
|
||||
flatten_item(k, v)
|
||||
else:
|
||||
raise ValueError(f'flatten_keys is undefined for container type "{container}"')
|
||||
|
||||
return output
|
||||
|
||||
|
||||
def flatten_table(table: list, id_fmt: str = 'x') -> list:
|
||||
if len(table) < 1:
|
||||
return table # Empty
|
||||
if isinstance(table[0], dict): # A simple table
|
||||
return [flatten_keys(d) for d in table]
|
||||
if isinstance(table[0], list): # Nested lists are bad when expanded as columns, so we'll expand
|
||||
print(table[0])
|
||||
flattened_table = []
|
||||
|
||||
def flatten_list(data, ids: list[int], id_max_digits: list[int]) -> None:
|
||||
if isinstance(data, list):
|
||||
max_digits = id_max_digits + [get_max_number_width(data, id_fmt)]
|
||||
for id, sub in enumerate(data):
|
||||
flatten_list(sub, ids + [id], max_digits)
|
||||
else:
|
||||
entry = {'ID': encode_nested_ids(ids, id_max_digits, id_fmt)}
|
||||
entry.update(flatten_keys(data))
|
||||
flattened_table.append(entry)
|
||||
|
||||
flatten_list(table, [], [])
|
||||
return flattened_table
|
||||
else:
|
||||
raise NotImplementedError(table[0])
|
||||
|
||||
|
||||
def unflatten_keys(d: dict) -> dict:
|
||||
output = {}
|
||||
for k, v in d.items():
|
||||
keysplit = k.split('.')
|
||||
target_dict = output
|
||||
for prefix in keysplit[:-1]:
|
||||
if prefix not in target_dict:
|
||||
target_dict[prefix] = {}
|
||||
target_dict = target_dict[prefix]
|
||||
target_dict[k] = v
|
||||
return output
|
||||
|
||||
|
||||
def unflatten_table(headers: list[str], entries: list):
|
||||
if 'ID' not in headers:
|
||||
return entries
|
||||
# This could be an array of an array of an array of an...
|
||||
id0 = entries[0]['ID']
|
||||
if '.' not in id0 and ':' not in id0:
|
||||
return entries
|
||||
# Treat this as a nested array
|
||||
table = {tuple(decode_nested_ids(entry['ID'])): entry for entry in entries}
|
||||
|
||||
output = []
|
||||
def unflatten_arrays(id_split: tuple[int], cur_array: list, value):
|
||||
i, *remainder = id_split
|
||||
if len(remainder) > 0:
|
||||
while len(cur_array) <= i: # Make sure our array has the index we're about to jump into
|
||||
cur_array.append([])
|
||||
unflatten_arrays(remainder, cur_array[i], value)
|
||||
else:
|
||||
while len(cur_array) <= i: # Make sure our array has the index we're about to overwrite
|
||||
cur_array.append(None)
|
||||
cur_array[i] = value
|
||||
|
||||
for id_split in sorted(table.keys()):
|
||||
unflatten_arrays(id_split, output, table[id_split])
|
||||
|
||||
return output
|
||||
|
||||
|
||||
def dump_tsv(filename, table, id_column=True) -> None:
|
||||
table_flat = flatten_table(table)
|
||||
|
||||
with open(filename, 'w') as file:
|
||||
headers = list(table_flat[0].keys())
|
||||
if id_column and 'ID' not in headers: # Some flattened tables build their own ID column!
|
||||
# See how long the hex representation of the last number will be, so we can zero-pad the rest to match.
|
||||
fmt = get_number_zero_pad(table_flat, 'X')
|
||||
file.write('\t'.join(['ID'] + headers) + '\n')
|
||||
for i, entry in enumerate(table_flat):
|
||||
file.write('\t'.join([f'0x{i:{fmt}}'] + [str(entry[key]) for key in headers]) + '\n')
|
||||
else:
|
||||
file.write('\t'.join(headers) + '\n')
|
||||
for i, entry in enumerate(table_flat):
|
||||
file.write('\t'.join([str(entry[key]) for key in headers]) + '\n')
|
||||
|
||||
|
||||
def load_tsv(filename) -> list:
|
||||
with open(filename, 'r') as file:
|
||||
lines = file.read().rstrip().split('\n')
|
||||
if len(lines) < 2:
|
||||
return []
|
||||
headers = lines[0].split('\t')
|
||||
|
||||
# Simple line-by-line unflatten
|
||||
entries = []
|
||||
for line in lines[1:]:
|
||||
entry = {key: try_int(value) for key, value in zip(headers, line.split('\t'))}
|
||||
entries.append(unflatten_keys(entry))
|
||||
|
||||
return unflatten_table(headers, entries)
|
||||
|
||||
|
||||
def load_ff5_snes_struct_definitions() -> dict:
|
||||
existing_structs = get_base_structarraytypes()
|
||||
parse_struct_definitions_from_tsv_filename('ChocolateBirdData/structs_SNES_stubs.tsv', existing_structs)
|
||||
parse_struct_definitions_from_tsv_filename('ChocolateBirdData/5/structs/SNES_stubs.tsv', existing_structs)
|
||||
parse_struct_definitions_from_tsv_filename('ChocolateBirdData/5/structs/SNES.tsv', existing_structs)
|
||||
parse_struct_definitions_from_tsv_filename('ChocolateBirdData/5/structs/SNES_save.tsv', existing_structs)
|
||||
return existing_structs
|
||||
|
||||
class FF5SNESHandler:
|
||||
struct_definitions: dict = load_ff5_snes_struct_definitions()
|
||||
addresses: dict = {entry['Label']: entry for entry in load_tsv('ChocolateBirdData/5/addresses_SNES_PSX.tsv')}
|
||||
|
||||
def extract(self, table: str, in_buffer) -> list[dict]:
|
||||
# Deserialize a table
|
||||
leftover_bits = LeftoverBits()
|
||||
entry = self.addresses[table] # Remember to try/catch
|
||||
offset = entry['SNES']
|
||||
buf = ReadBuffer(in_buffer, offset)
|
||||
return get_structarraytype(entry['format'], self.struct_definitions).get_value(buf, leftover_bits)
|
||||
|
||||
def build(self, table: str, new_data: list[dict], out_buffer):
|
||||
# Serialize complete data. This WILL fail if the input data is incomplete.
|
||||
leftover_bits = LeftoverBits()
|
||||
entry = self.addresses[table] # Remember to try/catch
|
||||
offset = entry['SNES']
|
||||
buf = WriteBuffer(out_buffer, offset)
|
||||
get_structarraytype(entry['format'], self.struct_definitions).put_value(buf, new_data, leftover_bits)
|
||||
|
||||
def build_partial(self, table: str, new_data: list[dict], in_buffer, out_buffer):
|
||||
# Safely merge partial data over the existing data, then serialize it.
|
||||
existing_data = self.extract(table, in_buffer)
|
||||
for i, new in enumerate(new_data):
|
||||
id = new.get('ID', i)
|
||||
for k, v in new.items():
|
||||
if k != 'ID' and v is not None: # Allow holes in the table for values we don't care about overwriting
|
||||
existing_data[id][k] = v
|
||||
self.build(table, existing_data, out_buffer)
|
||||
|
||||
from includes.helpers import load_tsv, dump_tsv
|
||||
from includes.rom_serde import FF5SNESHandler
|
||||
|
||||
if __name__ == '__main__':
|
||||
from argparse import ArgumentParser
|
||||
|
|
Loading…
Reference in New Issue