mirror of
https://github.com/zeldaret/oot.git
synced 2024-11-25 01:34:18 +00:00
e833011ccd
* Pass all paths to tools rather than tools constructing them * fix: make --baserom-segments required * sync with mm reviews * --version everywhere
416 lines
13 KiB
Python
Executable file
416 lines
13 KiB
Python
Executable file
#!/usr/bin/env python3
|
|
|
|
# SPDX-FileCopyrightText: © 2024 ZeldaRET
|
|
# SPDX-License-Identifier: CC0-1.0
|
|
|
|
import argparse
|
|
import collections
|
|
from colorama import Fore, Style
|
|
from dataclasses import dataclass
|
|
import difflib
|
|
import itertools
|
|
import math
|
|
from pathlib import Path
|
|
import re
|
|
import subprocess
|
|
import sys
|
|
import multiprocessing
|
|
import multiprocessing.pool
|
|
from typing import Dict, Iterator, List, Optional, Tuple
|
|
|
|
|
|
def green(s: str) -> str:
|
|
return f"{Fore.GREEN}{s}{Style.RESET_ALL}"
|
|
|
|
|
|
def red(s: str) -> str:
|
|
return f"{Fore.RED}{s}{Style.RESET_ALL}"
|
|
|
|
|
|
# Make interrupting with ^C less jank
|
|
# https://stackoverflow.com/questions/72967793/keyboardinterrupt-with-python-multiprocessing-pool
|
|
def set_sigint_ignored():
|
|
import signal
|
|
|
|
signal.signal(signal.SIGINT, signal.SIG_IGN)
|
|
|
|
|
|
@dataclass
|
|
class Inst:
|
|
func_name: str
|
|
mnemonic: str
|
|
regs: List[str]
|
|
imm: Optional[int]
|
|
reloc_type: Optional[str]
|
|
reloc_symbol: Optional[str]
|
|
|
|
|
|
FUNC_RE = re.compile(r"([0-9a-f]+) <(.*)>:")
|
|
|
|
|
|
def parse_func_name(line: str) -> str:
|
|
match = FUNC_RE.match(line)
|
|
if not match:
|
|
raise Exception(f"could not parse function name from '{line}'")
|
|
return match.group(2)
|
|
|
|
|
|
def is_branch(mnemonic: str) -> bool:
|
|
return mnemonic.startswith("b") and mnemonic != "break"
|
|
|
|
|
|
def parse_inst(func_name: str, line: str) -> Inst:
|
|
parts = line.split()
|
|
addr = int(parts[0][:-1], 16)
|
|
mnemonic = parts[2]
|
|
regs = []
|
|
imm = None
|
|
if len(parts) > 3:
|
|
for part in parts[3].split(","):
|
|
if "(" in part: # load/store
|
|
offset_str, rest = part.split("(")
|
|
regs.append(rest[:-1])
|
|
imm = int(offset_str, 10)
|
|
elif is_branch(mnemonic):
|
|
try:
|
|
# convert branch targets to relative offsets
|
|
offset = int(part, 16)
|
|
imm = offset - addr - 4
|
|
except ValueError:
|
|
regs.append(part)
|
|
else:
|
|
try:
|
|
imm = int(part, 0)
|
|
except ValueError:
|
|
regs.append(part)
|
|
return Inst(func_name, mnemonic, regs, imm, None, None)
|
|
|
|
|
|
def run_objdump(path: Path, args: List[str]) -> str:
|
|
if not path.exists():
|
|
raise Exception(f"file {path} does not exist")
|
|
|
|
command = ["mips-linux-gnu-objdump"] + args + [str(path)]
|
|
try:
|
|
return subprocess.run(
|
|
command,
|
|
stdout=subprocess.PIPE,
|
|
stderr=subprocess.PIPE,
|
|
check=True,
|
|
encoding="utf-8",
|
|
).stdout
|
|
except subprocess.CalledProcessError as e:
|
|
return ""
|
|
|
|
|
|
def disassemble(path: Path) -> List[Inst]:
|
|
lines = run_objdump(path, ["-drz", "-m", "mips:4300", "-j", ".text"]).splitlines()
|
|
result = []
|
|
|
|
func_name = None
|
|
i = 6 # skip preamble
|
|
while i < len(lines):
|
|
row = lines[i]
|
|
i += 1
|
|
|
|
if not row:
|
|
continue
|
|
|
|
if not row.startswith(" "):
|
|
func_name = parse_func_name(row)
|
|
continue
|
|
|
|
if not func_name:
|
|
raise Exception(f"no function name for line '{row}'")
|
|
|
|
inst = parse_inst(func_name, row)
|
|
|
|
if i < len(lines) and lines[i].startswith("\t"):
|
|
reloc = lines[i]
|
|
i += 1
|
|
_, inst.reloc_type, inst.reloc_symbol = reloc.split()
|
|
|
|
result.append(inst)
|
|
|
|
# trim trailing nops
|
|
while result and result[-1].mnemonic == "nop":
|
|
result.pop()
|
|
return result
|
|
|
|
|
|
def pair_instructions(
|
|
insts1: List[Inst], insts2: List[Inst]
|
|
) -> Iterator[Tuple[Optional[Inst], Optional[Inst]]]:
|
|
differ = difflib.SequenceMatcher(
|
|
a=[(inst.func_name, inst.mnemonic) for inst in insts1],
|
|
b=[(inst.func_name, inst.mnemonic) for inst in insts2],
|
|
autojunk=False,
|
|
)
|
|
for tag, i1, i2, j1, j2 in differ.get_opcodes():
|
|
for inst1, inst2 in itertools.zip_longest(insts1[i1:i2], insts2[j1:j2]):
|
|
yield (inst1, inst2)
|
|
|
|
|
|
def has_diff(inst1: Inst, inst2: Inst) -> bool:
|
|
if (
|
|
inst1.func_name != inst2.func_name
|
|
or inst1.mnemonic != inst2.mnemonic
|
|
or inst1.regs != inst2.regs
|
|
):
|
|
return True
|
|
|
|
if inst1.reloc_type == inst2.reloc_type and inst1.reloc_type in (
|
|
"R_MIPS_HI16",
|
|
"R_MIPS_LO16",
|
|
):
|
|
# ignore symbol differences
|
|
return False
|
|
|
|
return inst1 != inst2
|
|
|
|
|
|
def get_section_sizes(path: Path) -> Dict[str, int]:
|
|
lines = run_objdump(path, ["-h"]).splitlines()
|
|
if len(lines) < 5:
|
|
return {}
|
|
|
|
result = {}
|
|
for i in range(5, len(lines), 2):
|
|
parts = lines[i].split()
|
|
name = parts[1]
|
|
size = int(parts[2], 16)
|
|
# Pad to 0x10-byte alignment
|
|
result[parts[1]] = (size + 0xF) & ~0xF
|
|
return result
|
|
|
|
|
|
def get_section_hex_dump(path: Path, section: str) -> List[str]:
|
|
lines = run_objdump(path, ["-s", "-j", section]).splitlines()
|
|
return lines[4:]
|
|
|
|
|
|
def parse_hex_dump(lines: List[str]) -> bytes:
|
|
result = bytearray()
|
|
for line in lines:
|
|
data = line[6:41].replace(" ", "")
|
|
result.extend(bytes.fromhex(data))
|
|
|
|
# pad to 0x10-byte alignment
|
|
while len(result) % 0x10:
|
|
result.append(0)
|
|
|
|
return result
|
|
|
|
|
|
def find_functions_with_diffs(version: str, c_path: str):
|
|
object_path = Path(c_path).with_suffix(".o")
|
|
|
|
expected_dir = Path("expected/build") / version
|
|
build_dir = Path("build") / version
|
|
|
|
insts1 = disassemble(expected_dir / object_path)
|
|
insts2 = disassemble(build_dir / object_path)
|
|
|
|
functions_with_diffs = collections.OrderedDict()
|
|
for inst1, inst2 in pair_instructions(insts1, insts2):
|
|
if inst1 is None and inst2 is not None:
|
|
functions_with_diffs[inst2.func_name] = True
|
|
elif inst1 is not None and inst2 is None:
|
|
functions_with_diffs[inst1.func_name] = True
|
|
elif inst1 is not None and inst2 is not None and has_diff(inst1, inst2):
|
|
functions_with_diffs[inst1.func_name] = True
|
|
functions_with_diffs[inst2.func_name] = True
|
|
|
|
if not functions_with_diffs:
|
|
print(f"{c_path} OK")
|
|
return
|
|
|
|
print(f"{c_path} functions with diffs:")
|
|
for func_name in functions_with_diffs:
|
|
print(f" {func_name}")
|
|
|
|
|
|
def find_data_diffs(version: str, c_path: str):
|
|
object_path = Path(c_path).with_suffix(".o")
|
|
|
|
expected_dir = Path("expected/build") / version
|
|
build_dir = Path("build") / version
|
|
|
|
sizes1 = get_section_sizes(expected_dir / object_path)
|
|
sizes2 = get_section_sizes(build_dir / object_path)
|
|
rodata_dump1 = get_section_hex_dump(expected_dir / object_path, ".rodata")
|
|
rodata_dump2 = get_section_hex_dump(build_dir / object_path, ".rodata")
|
|
rodata1 = parse_hex_dump(rodata_dump1)
|
|
rodata2 = parse_hex_dump(rodata_dump2)
|
|
|
|
rodata_matches = rodata1 == rodata2
|
|
data_size_matches = sizes1.get(".data", 0) == sizes2.get(".data", 0)
|
|
bss_size_matches = sizes1.get(".bss", 0) == sizes2.get(".bss", 0)
|
|
|
|
if rodata_matches:
|
|
print(f"{c_path} .rodata OK")
|
|
else:
|
|
print(
|
|
f"{c_path} .rodata differs: expected size 0x{sizes1.get('.rodata', 0):04x} vs build size 0x{sizes2.get('.rodata', 0):04x}"
|
|
)
|
|
print(f" expected:")
|
|
print("\n".join(rodata_dump1))
|
|
print(f" build:")
|
|
print("\n".join(rodata_dump2))
|
|
|
|
if data_size_matches:
|
|
print(f"{c_path} .data size OK")
|
|
else:
|
|
print(
|
|
f"{c_path} .data size differs: expected size 0x{sizes1.get('.data', 0):04x} vs build size 0x{sizes2.get('.data', 0):04x}"
|
|
)
|
|
|
|
if bss_size_matches:
|
|
print(f"{c_path} .bss size OK")
|
|
else:
|
|
print(
|
|
f"{c_path} .bss size differs: expected size 0x{sizes1.get('.bss', 0):04x} vs build size 0x{sizes2.get('.bss', 0):04x}"
|
|
)
|
|
|
|
|
|
@dataclass
|
|
class ObjectDataForComparison:
|
|
insts1: List[Inst]
|
|
insts2: List[Inst]
|
|
sizes1: Dict[str, int]
|
|
sizes2: Dict[str, int]
|
|
rodata1: bytes
|
|
rodata2: bytes
|
|
|
|
|
|
def get_object_data_for_comparison(object1: Path, object2: Path):
|
|
insts1 = disassemble(object1)
|
|
insts2 = disassemble(object2)
|
|
sizes1 = get_section_sizes(object1)
|
|
sizes2 = get_section_sizes(object2)
|
|
rodata_dump1 = get_section_hex_dump(object1, ".rodata")
|
|
rodata_dump2 = get_section_hex_dump(object2, ".rodata")
|
|
rodata1 = parse_hex_dump(rodata_dump1)
|
|
rodata2 = parse_hex_dump(rodata_dump2)
|
|
return ObjectDataForComparison(insts1, insts2, sizes1, sizes2, rodata1, rodata2)
|
|
|
|
|
|
def print_summary(version: str, csv: bool, only_not_ok: bool):
|
|
expected_dir = Path("expected/build") / version
|
|
build_dir = Path("build") / version
|
|
|
|
expected_object_files = sorted(expected_dir.glob("src/**/*.o"))
|
|
|
|
comparison_data_list: List[multiprocessing.pool.AsyncResult] = []
|
|
|
|
with multiprocessing.Pool(initializer=set_sigint_ignored) as p:
|
|
for expected_object in expected_object_files:
|
|
build_object = build_dir / expected_object.relative_to(expected_dir)
|
|
comparison_data_list.append(
|
|
p.apply_async(
|
|
get_object_data_for_comparison,
|
|
(expected_object, build_object),
|
|
)
|
|
)
|
|
if csv:
|
|
print("path,expected,actual,.text,.rodata,.data size,.bss size")
|
|
for expected_object, data_async in zip(
|
|
expected_object_files, comparison_data_list
|
|
):
|
|
c_path = expected_object.relative_to(expected_dir).with_suffix(".c")
|
|
data = data_async.get()
|
|
|
|
insts1 = data.insts1
|
|
insts2 = data.insts2
|
|
|
|
added = 0
|
|
removed = 0
|
|
changed = 0
|
|
for inst1, inst2 in pair_instructions(insts1, insts2):
|
|
if inst1 is None and inst2 is not None:
|
|
added += 1
|
|
elif inst1 is not None and inst2 is None:
|
|
removed += 1
|
|
elif inst1 is not None and inst2 is not None and has_diff(inst1, inst2):
|
|
changed += 1
|
|
|
|
if insts1:
|
|
text_progress = max(1.0 - (added + removed + changed) / len(insts1), 0)
|
|
else:
|
|
text_progress = 1.0
|
|
|
|
sizes1 = data.sizes1
|
|
sizes2 = data.sizes2
|
|
rodata1 = data.rodata1
|
|
rodata2 = data.rodata2
|
|
|
|
rodata_matches = rodata1 == rodata2
|
|
data_size_matches = sizes1.get(".data", 0) == sizes2.get(".data", 0)
|
|
bss_size_matches = sizes1.get(".bss", 0) == sizes2.get(".bss", 0)
|
|
|
|
if only_not_ok:
|
|
if (
|
|
text_progress == 1
|
|
and rodata_matches
|
|
and data_size_matches
|
|
and bss_size_matches
|
|
):
|
|
continue
|
|
|
|
if csv:
|
|
print(
|
|
f"{c_path},{len(insts1)},{len(insts2)},{text_progress:.3f},{rodata_matches},{data_size_matches},{bss_size_matches}"
|
|
)
|
|
else:
|
|
ok = green("OK")
|
|
diff = red("diff")
|
|
text_progress_str = (
|
|
ok
|
|
if text_progress == 1
|
|
else red(f"{math.floor(text_progress * 100):>2}%")
|
|
)
|
|
rodata_str = ok if rodata_matches else diff
|
|
data_size_str = ok if data_size_matches else diff
|
|
bss_size_str = ok if bss_size_matches else diff
|
|
print(
|
|
f"text:{text_progress_str:<13} rodata:{rodata_str:<13} data size:{data_size_str:<13} bss size:{bss_size_str:<13} {c_path}"
|
|
)
|
|
sys.stdout.flush()
|
|
|
|
|
|
if __name__ == "__main__":
|
|
parser = argparse.ArgumentParser(description="Calculate progress matching retail")
|
|
parser.add_argument(
|
|
"file",
|
|
metavar="FILE",
|
|
nargs="?",
|
|
help="find functions with diffs in the given source file (if omitted, print summary of diffs for all files)",
|
|
)
|
|
parser.add_argument(
|
|
"-v",
|
|
"--version",
|
|
dest="oot_version",
|
|
help="version to compare",
|
|
default="ntsc-1.2",
|
|
)
|
|
parser.add_argument(
|
|
"--data",
|
|
help="diff .data size, .bss size, and .rodata contents instead of text",
|
|
action="store_true",
|
|
)
|
|
parser.add_argument(
|
|
"--not-ok",
|
|
help="only print non-OK files",
|
|
action="store_true",
|
|
dest="only_not_ok",
|
|
)
|
|
parser.add_argument("--csv", help="print summary CSV", action="store_true")
|
|
args = parser.parse_args()
|
|
|
|
if args.file is not None:
|
|
if args.data:
|
|
find_data_diffs(args.oot_version, args.file)
|
|
else:
|
|
find_functions_with_diffs(args.oot_version, args.file)
|
|
else:
|
|
print_summary(args.oot_version, args.csv, args.only_not_ok)
|