pikmin2/tools/calcprogress.py
EpochFlame 3fc7eabe7b Represent data progress in the printout.
Data is represented by treasures.
2021-12-24 13:40:39 -05:00

173 lines
6.2 KiB
Python

#!/usr/bin/env python3
################################################################################
# Description #
################################################################################
# calcprogress: Used to calculate the progress of the Pikmin 2 decomp. #
# Prints to stdout for now, but eventually will have some form of storage, #
# i.e. CSV, so that it can be used for a webpage display. #
# #
# Usage: No arguments needed #
################################################################################
###############################################
# #
# Imports #
# #
###############################################
import os
import sys
import struct
import re
import math
###############################################
# #
# Constants #
# #
###############################################
DOL_PATH = "baserom.dol"
MAP_PATH = "build/pikmin2.usa/pikmin2UP.MAP"
MEM1_HI = 0x81200000
MEM1_LO = 0x80004000
MW_WII_SYMBOL_REGEX = r"^\s*"\
r"(?P<SectOfs>\w{8})\s+"\
r"(?P<Size>\w{6})\s+"\
r"(?P<VirtOfs>\w{8})\s+"\
r"(?P<FileOfs>\w{8})\s+"\
r"(\w{1,2})\s+"\
r"(?P<Symbol>[0-9A-Za-z_<>$@.*]*)\s*"\
r"(?P<Object>\S*)"
MW_GC_SYMBOL_REGEX = r"^\s*"\
r"(?P<SectOfs>\w{8})\s+"\
r"(?P<Size>\w{6})\s+"\
r"(?P<VirtOfs>\w{8})\s+"\
r"(\w{1,2})\s+"\
r"(?P<Symbol>[0-9A-Za-z_<>$@.*]*)\s*"\
r"(?P<Object>\S*)"
REGEX_TO_USE = MW_GC_SYMBOL_REGEX
TEXT_SECTIONS = ["init", "text"]
DATA_SECTIONS = [
"rodata", "data", "bss", "sdata", "sbss", "sdata2", "sbss2", "file",
"ctors", "_ctors", "dtors", "ctors$99", "_ctors$99", "ctors$00", "dtors$99",
"extab_", "extabindex_"
]
# DOL info
TEXT_SECTION_COUNT = 7
DATA_SECTION_COUNT = 11
SECTION_TEXT = 0
SECTION_DATA = 1
###############################################
# #
# Entrypoint #
# #
###############################################
if __name__ == "__main__":
# Sum up DOL section sizes
dol_handle = open(DOL_PATH, "rb")
# Seek to virtual addresses
dol_handle.seek(0x48)
# Read virtual addresses
text_starts = list()
for i in range(TEXT_SECTION_COUNT):
text_starts.append(int.from_bytes(dol_handle.read(4), byteorder='big'))
data_starts = list()
for i in range(DATA_SECTION_COUNT):
data_starts.append(int.from_bytes(dol_handle.read(4), byteorder='big'))
# Read lengths
text_sizes = list()
for i in range(TEXT_SECTION_COUNT):
text_sizes.append(int.from_bytes(dol_handle.read(4), byteorder='big'))
data_sizes = list()
for i in range(DATA_SECTION_COUNT):
data_sizes.append(int.from_bytes(dol_handle.read(4), byteorder='big'))
# BSS address + length
bss_start = int.from_bytes(dol_handle.read(4), byteorder='big')
bss_size = int.from_bytes(dol_handle.read(4), byteorder='big')
bss_end = bss_start + bss_size
dol_code_size = 0
dol_data_size = 0
for i in range(DATA_SECTION_COUNT):
# Ignore sections inside BSS
if (data_starts[i] >= bss_start) and (data_starts[i] + data_sizes[i] <= bss_end): continue
dol_data_size += data_sizes[i]
dol_data_size += bss_size
for i in text_sizes:
dol_code_size += i
# Open map file
mapfile = open(MAP_PATH, "r")
symbols = mapfile.readlines()
decomp_code_size = 0
decomp_data_size = 0
section_type = None
# Find first section
first_section = 0
while (symbols[first_section].startswith(".") == False and "section layout" not in symbols[first_section]): first_section += 1
assert(first_section < len(symbols)), "Map file contains no sections!!!"
for i in range(first_section, len(symbols)):
# New section
if (symbols[i].startswith(".") == True or "section layout" in symbols[i]):
# Grab section name (i.e. ".init section layout" -> "init")
sectionName = re.search(r"\.*(?P<Name>\w+)\s", symbols[i]).group("Name")
# Determine type of section
section_type = SECTION_DATA if (sectionName in DATA_SECTIONS) else SECTION_TEXT
# Parse symbols until we hit the next section declaration
else:
if ("entry of" in symbols[i]) or ("UNUSED" in symbols[i]): continue
assert(section_type != None), f"Symbol found outside of a section!!!\n{symbols[i]}"
match_obj = re.search(REGEX_TO_USE, symbols[i])
# Should be a symbol in ASM (so we discard it)
if (match_obj == None): continue
# Is the symbol a file-wide section?
symb = match_obj.group("Symbol")
if (symb.startswith("*fill*")) or (symb.startswith(".") and symb[1:] in TEXT_SECTIONS or symb[1:] in DATA_SECTIONS): continue
# For sections that don't start with "."
if (symb in DATA_SECTIONS): continue
# If not, we accumulate the file size
if (section_type == SECTION_TEXT):
decomp_code_size += int(match_obj.group("Size"), 16)
else:
decomp_data_size += int(match_obj.group("Size"), 16)
# Calculate percentages
codeCompletionPcnt = (decomp_code_size / dol_code_size)
dataCompletionPcnt = (decomp_data_size / dol_data_size)
bytesPerPoko = dol_code_size / 10000
bytesPerPart = dol_data_size / 201
pokoCount = math.floor(decomp_code_size / bytesPerPoko)
partCount = math.floor(decomp_code_size / bytesPerPart)
print("Progress:")
print(f"\tCode sections: {decomp_code_size} / {dol_code_size} bytes in src ({codeCompletionPcnt:%})")
print(f"\tData sections: {decomp_data_size} / {dol_data_size} bytes in src ({dataCompletionPcnt:%})")
print("\nYou have {} out of 10000 Pokos and collected {} out of 201 treasures.".format(pokoCount, partCount))