mirror of
https://github.com/zeldaret/oot.git
synced 2024-11-15 06:06:04 +00:00
741c140aef
* Split files * Format rodata * Some more code rodata migrated * Some more actor rodata migrated * Migrate rodata for ovl_Boss_Ganon * Migrate rodata for code_800EC960 * Remove unused rodata * x1b occurences all lowercase b
297 lines
11 KiB
Python
297 lines
11 KiB
Python
"""
|
|
Given a code file name (excluding the file extension, e.g. z_view) this script
|
|
will attempt to build rodata and late_rodata sections for all of its functions.
|
|
|
|
Supports overlays and other files as long as the rodata file is formatted
|
|
correctly(see files in code for examples of properly formatted rodata sections)
|
|
|
|
This supports automatically commenting or deleting the rodata file from the
|
|
spec and automatic deletion of the rodata file itself (only use if you are sure
|
|
it will build afterwards)
|
|
|
|
Has trouble with rodata that goes unused, and .incbin placement
|
|
"""
|
|
|
|
import re
|
|
import os
|
|
from os import system, listdir, remove, walk
|
|
from os.path import exists, isdir, sep
|
|
|
|
# FUNCTIONS -----------------------------------------------------------------
|
|
|
|
"""
|
|
Extracts rodata symbols from asm
|
|
"""
|
|
def asm_syms(asm):
|
|
split = re.split("jtbl_|D_",asm)
|
|
return [s.partition(")")[0] for s in split[1:len(split)]]
|
|
|
|
"""
|
|
Extracts rodata symbols from rodata
|
|
"""
|
|
def rodata_syms(rodata):
|
|
split = re.split("glabel jtbl_|glabel D_",rodata)
|
|
return [s.partition("\n")[0] for s in split[1:len(split)]]
|
|
|
|
"""
|
|
Extracts the symbol from a single rodata block
|
|
"""
|
|
def rodata_sym(rodata_block):
|
|
return (re.split("glabel jtbl_|glabel D_",rodata_block)[1]).split("\n")[0]
|
|
|
|
"""
|
|
Splits rodata into blocks
|
|
"""
|
|
def rodata_blocks(rodata):
|
|
split = rodata.split("glabel")
|
|
return ["glabel" + b for b in split[1:len(split)]]
|
|
|
|
"""
|
|
Gets the block size
|
|
"""
|
|
def rodata_block_size(rodata):
|
|
split = rodata.split(" .")
|
|
elements = split[1:len(split)]
|
|
size = 0
|
|
for e in elements:
|
|
element_type = e.split(" ")[0]
|
|
if element_type == "double":
|
|
size += 8
|
|
if element_type == "float":
|
|
size += 4
|
|
if element_type == "word":
|
|
size += 4
|
|
if element_type == "incbin":
|
|
size += int(e.rpartition(", ")[2].split("\n")[0],16)
|
|
if element_type == "ascii":
|
|
size += len(e.split("\"")[1].split("\"")[0])
|
|
if element_type == "asciz":
|
|
size += len(e.split("\"")[1].split("\"")[0])
|
|
if element_type == "balign":
|
|
size += size % int(e.split(" ")[1])
|
|
return size
|
|
|
|
"""
|
|
Gets the text size
|
|
"""
|
|
def text_size(asm):
|
|
instructions = [l for l in asm.split("\n") if l.startswith("/* ")]
|
|
return 4*(len(instructions)-1)
|
|
|
|
"""
|
|
Gets the rodata-to-text ratio
|
|
"""
|
|
def late_rodata_ratio(asm,late_rodata_blocks):
|
|
total_rodata_size = 0
|
|
for b in late_rodata_blocks:
|
|
total_rodata_size += rodata_block_size(b)
|
|
return total_rodata_size/text_size(asm)
|
|
|
|
"""
|
|
Accepts a single block of rodata and extracts the type
|
|
"""
|
|
def rodata_type(rodata_block):
|
|
first_split = re.split("\s+\.", rodata_block)
|
|
return first_split[1].split(" ")[0]
|
|
|
|
"""
|
|
Checks if a block should go in .rdata or .late_rodata
|
|
"""
|
|
def is_rodata(r):
|
|
return (rodata_type(r)=="asciz" or rodata_type=="ascii")
|
|
|
|
"""
|
|
For given asm and rodata files, build a rodata section for the asm file
|
|
"""
|
|
def build_rodata(asm, rodata):
|
|
contained_syms = [s for s in asm_syms(asm) if s in rodata_syms(rodata)]
|
|
contained_blocks = [b for b in rodata_blocks(rodata) if rodata_sym(b) in contained_syms]
|
|
# TODO include arrays in rodata_list
|
|
rodata_list = [r for r in contained_blocks if is_rodata(r)]
|
|
return_str = ""
|
|
if (len(rodata_list)!=0):
|
|
return_str += ".rdata\n"
|
|
for r in rodata_list:
|
|
return_str += r
|
|
late_rodata_list = [r for r in contained_blocks if r not in rodata_list]
|
|
if (len(late_rodata_list)!=0):
|
|
if late_rodata_ratio(asm,late_rodata_list) > (1/3):
|
|
top_sym = rodata_sym(late_rodata_list[0])
|
|
if top_sym.endswith("0") or top_sym.endswith("8"):
|
|
return_str += ".late_rodata\n.late_rodata_alignment 8\n"
|
|
elif top_sym.endswith("4") or top_sym.endswith("C"):
|
|
return_str += ".late_rodata\n.late_rodata_alignment 4\n"
|
|
else:
|
|
return_str += ".late_rodata\n"
|
|
for r in late_rodata_list:
|
|
return_str += r
|
|
return None if return_str=="" else return_str + ("" if return_str.endswith("\n\n") else "\n") + ".text\n"
|
|
|
|
"""
|
|
Gets all file paths contained in a given folder, does not enter subfolders
|
|
"""
|
|
def get_file_paths(folder_path):
|
|
return next(walk(folder_path),(None,None,[]))[2]
|
|
|
|
"""
|
|
Given a path, reads the asm .s file into a single string
|
|
"""
|
|
def file_read(path):
|
|
with open(path, 'r', encoding="utf-8") as infile:
|
|
return infile.read()
|
|
|
|
"""
|
|
Writes the rodata section to the start of the file specified by path
|
|
"""
|
|
def rodata_write(path, section):
|
|
with open(path, 'r+', encoding="utf-8", newline="\n") as outfile:
|
|
original = outfile.read()
|
|
outfile.seek(0,0)
|
|
outfile.write(str(section) + original)
|
|
|
|
"""
|
|
Comments out the line in spec associated with the given filenames
|
|
"""
|
|
def modify_spec(filenames, identifier, delete):
|
|
lines = ""
|
|
with open("spec", 'r+', encoding="utf-8", newline="\n") as spec:
|
|
lines = spec.read().split("\n")
|
|
changed = False
|
|
files = filenames.split(",")
|
|
for filename in files:
|
|
eff_filename = filename.lower().replace("effect_", "eff_")
|
|
if identifier == "code" and " include \"build/data/" + filename + ".rodata.o\"" in lines:
|
|
e = lines.index(" include \"build/data/" + filename + ".rodata.o\"")
|
|
if delete:
|
|
del lines[e]
|
|
else:
|
|
lines[e] = " //include \"build/data/" + filename + ".rodata.o\""
|
|
changed = True
|
|
elif " include \"build/data/overlays/" + identifier + "/z_" + eff_filename + ".rodata.o\"" in lines:
|
|
e = lines.index(" include \"build/data/overlays/" + identifier + "/z_" + eff_filename + ".rodata.o\"")
|
|
if delete:
|
|
del lines[e]
|
|
else:
|
|
lines[e] = " //include \"build/data/overlays/" + identifier + "/z_" + eff_filename + ".rodata.o\""
|
|
changed = True
|
|
if changed:
|
|
modified = "\n".join(lines)
|
|
with open("spec", 'w', encoding="utf-8", newline="\n") as spec:
|
|
#spec.seek(0,0)
|
|
spec.write(modified)
|
|
|
|
"""
|
|
Processes each individual file
|
|
asm\non_matchings\overlays\<identifier>
|
|
data\overlays\<identifier>
|
|
asm\non_matchings\code\
|
|
data\code\
|
|
"""
|
|
def process_file(filename, identifier, delete_rodata):
|
|
folder_path = "asm" + sep + "non_matchings" + sep + ("code" + sep if identifier=="code" else "overlays" + sep + identifier.lower() + sep + "ovl_") + filename + sep
|
|
rodata_path = "data" + sep + (sep if identifier=="code" else "overlays" + sep + identifier.lower() + sep + "z_") + filename.lower() + ".rodata.s"
|
|
if filename == "player":
|
|
folder_path = "asm" + sep + "non_matchings" + sep + "overlays" + sep + "actors" + sep + "ovl_player_actor" + sep
|
|
rodata_path = rodata_path.replace("effect_", "eff_")
|
|
print("ASM at: " + folder_path)
|
|
print("Data at: " + rodata_path)
|
|
if not exists(folder_path):
|
|
print("Aborting: ASM does not exist")
|
|
return
|
|
if not exists(rodata_path):
|
|
print("Aborting: Data does not exist")
|
|
return
|
|
files = [folder_path + f for f in get_file_paths(folder_path)]
|
|
for asm_file in files:
|
|
asm = file_read(asm_file)
|
|
print("Found asm file " + asm_file)
|
|
if ".rdata" in asm:
|
|
print("Skipping: it already has a rodata section")
|
|
continue
|
|
print("Processing asm file " + asm_file)
|
|
section = build_rodata(asm, file_read(rodata_path))
|
|
if section is not None:
|
|
print("Writing asm file " + asm_file)
|
|
rodata_write(asm_file, section)
|
|
else: print("Skipping: no rodata to write")
|
|
print("Built rodata sections for " + identifier + " file " + filename)
|
|
if delete_rodata:
|
|
remove(rodata_path)
|
|
print("Deleted rodata at " + rodata_path)
|
|
|
|
"""
|
|
Allows files to be provided as comma-separated filenames for batch migration
|
|
"""
|
|
def process_files(filenames, identifier, spechandle, delete_rodata):
|
|
if "," in filenames:
|
|
files = filenames.split(",")
|
|
for f in files:
|
|
process_file(f, identifier, delete_rodata)
|
|
else:
|
|
process_file(filenames, identifier, delete_rodata)
|
|
if spechandle.lower() == "delete":
|
|
modify_spec(filenames, identifier, True)
|
|
print("Deleted rodata for files in spec")
|
|
elif spechandle.lower() == "comment":
|
|
result = modify_spec(filenames, identifier, False)
|
|
if result:
|
|
print("Commented out rodata for files in spec")
|
|
|
|
"""
|
|
Asks what to do about spec and rodata, converts 'all' to comma-separated filenames
|
|
"""
|
|
def check_spec_rodata(filenames, identifier):
|
|
spechandle = input("Delete, Comment or Leave spec? ")
|
|
delete_rodata = input("Delete rodata file(s)? (Y/N) ")
|
|
if filenames == "all" or "all|" in filenames:
|
|
to_remove_list = []
|
|
if "all|" in filenames:
|
|
print("all| in filenames")
|
|
to_remove_list = filenames.split("|")[1]
|
|
basedir = "asm" + sep + "non_matchings" + sep + ("code" if identifier=="code" else "overlays" + sep + identifier.lower()) + sep
|
|
folder_names = [name.replace("ovl_","") for name in listdir(basedir) if isdir(basedir + name) and name not in to_remove_list]
|
|
filenames = ",".join(map(str, folder_names))
|
|
print(filenames)
|
|
process_files(filenames, identifier, spechandle, delete_rodata == "Y")
|
|
|
|
"""
|
|
Main execution
|
|
"""
|
|
def run(show_help):
|
|
if(show_help):
|
|
print("""Usage: Enter 'Code' or 'Overlay' and follow the instructions.
|
|
Batch migrate files by entering comma-separated filenames instead of a single filename.
|
|
Migrate all files by entering 'all' for filenames. Exclude files from all with all| followed by comma-separated filenames Use at your own risk.
|
|
Enter 'q' to the code or overlay question to quit.""")
|
|
code_or_overlay = input("Code or Overlay? ")
|
|
if(code_or_overlay == "q"):
|
|
return
|
|
if(code_or_overlay == "Code"):
|
|
filename = input("Enter the code file name(s) (excluding .c) or all: ")
|
|
check_spec_rodata(filename, "code")
|
|
elif(code_or_overlay == "Overlay"):
|
|
overlay_type = input("Actor, Effect or Gamestate? ")
|
|
if(overlay_type == "Actor"):
|
|
filename = input("Enter the actor name(s) (excluding ovl_ or z_, ex. obj_bombiwa) or all: ")
|
|
check_spec_rodata(filename, "actors")
|
|
elif(overlay_type == "Effect"):
|
|
filename = input("Enter the effect name(s) (excluding ovl_ or z_, ex. effect_ss_bomb) or all: ")
|
|
check_spec_rodata(filename, "effects")
|
|
elif(overlay_type == "Gamestate"):
|
|
filename = input("Enter the gamestate name(s) (excluding ovl_ or z_, ex. kaleido_scope) or all: ")
|
|
check_spec_rodata(filename, "gamestates")
|
|
run(True)
|
|
|
|
# PROGRAM -------------------------------------------------------------------
|
|
|
|
run(False)
|
|
#bigs = ["Boss_Ganon", "Boss_Ganondrof","En_Wf", "Door_Warp1",]
|
|
#ovls = ["En_Elf"]
|
|
#effects = [x[0] for x in os.walk("src/overlays/effects")][1:]
|
|
|
|
#for i, ovl in enumerate(effects):
|
|
# process_files(ovl.split("/")[-1][4:], "effects", "Delete", True)
|
|
# command = "echo >> src/overlays/effects/ovl_" + effects[i] + "/z_" + effects[i].lower() + ".c"
|
|
# os.system(command) # purpose of this is to "modify" each C file in order to prevent undefined symbol errors.
|
|
# # the new line will be removed by format.sh
|
|
|