isledecomp/isle
1
0
Fork 0
mirror of https://github.com/isledecomp/isle.git synced 2026-01-24 00:31:16 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Use isledecomp lib in reccmp

Browse Source
This commit is contained in:
disinvite 2023-11-14 14:49:27 -05:00
parent 0fa6d207f8
commit 6a1e05c270
2 changed files with 100 additions and 118 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
tools/isledecomp/isledecomp/dir.py
View File

@ -7,11 +7,15 @@ def file_is_cpp(filename: str) -> bool:
return ext.lower() in ('.h', '.cpp') return ext.lower() in ('.h', '.cpp')
def walk_source_dir(source: str) -> Iterator[str]: def walk_source_dir(source: str, recursive: bool = True) -> Iterator[str]:
"""Generator to walk the given directory recursively and return """Generator to walk the given directory recursively and return
any C++ files found.""" any C++ files found."""
source = os.path.abspath(source)
for subdir, dirs, files in os.walk(source): for subdir, dirs, files in os.walk(source):
for file in files: for file in files:
if file_is_cpp(file): if file_is_cpp(file):
yield os.path.join(subdir, file) yield os.path.join(subdir, file)
if not recursive:
break

212
tools/reccmp/reccmp.py
View File

@ -12,6 +12,8 @@
import colorama import colorama
import html import html
import re import re
from isledecomp.dir import walk_source_dir
from isledecomp.parser import find_code_blocks
parser = argparse.ArgumentParser(allow_abbrev=False, parser = argparse.ArgumentParser(allow_abbrev=False,
description='Recompilation Compare: compare an original EXE with a recompiled EXE + PDB.') description='Recompilation Compare: compare an original EXE with a recompiled EXE + PDB.')
@ -414,143 +416,119 @@ def can_resolve_register_differences(original_asm, new_asm):
basename = os.path.basename(os.path.splitext(original)[0]) basename = os.path.basename(os.path.splitext(original)[0])
pattern = '// OFFSET:' pattern = '// OFFSET:'
for subdir, dirs, files in os.walk(source): for srcfilename in walk_source_dir(source):
for file in files: with open(srcfilename, 'r') as srcfile:
srcfilename = os.path.join(os.path.abspath(subdir), file) blocks = find_code_blocks(srcfile)
with open(srcfilename, 'r') as srcfile:
line_no = 0
while True: for block in blocks:
try: if block.is_stub:
line = srcfile.readline() continue
line_no += 1
if not line: if block.module != basename:
break continue
line = line.strip() addr = block.offset
# Verbose flag handling
if verbose:
if addr == verbose:
found_verbose_target = True
else:
continue
if line.startswith(pattern) and not line.endswith('STUB'): if block.is_template:
par = line[len(pattern):].strip().split() recinfo = syminfo.get_recompiled_address_from_name(block.signature)
module = par[0] if not recinfo:
if module != basename: continue
continue else:
recinfo = syminfo.get_recompiled_address(srcfilename, block.start_line)
if not recinfo:
continue
addr = int(par[1], 16) # The effective_ratio is the ratio when ignoring differing register
# allocation vs the ratio is the true ratio.
ratio = 0.0
effective_ratio = 0.0
if recinfo.size:
origasm = parse_asm(origfile, addr + recinfo.start, recinfo.size)
recompasm = parse_asm(recompfile, recinfo.addr + recinfo.start, recinfo.size)
# Verbose flag handling diff = difflib.SequenceMatcher(None, origasm, recompasm)
if verbose: ratio = diff.ratio()
if addr == verbose: effective_ratio = ratio
found_verbose_target = True
else:
continue
if line.endswith('TEMPLATE'): if ratio != 1.0:
line = srcfile.readline() # Check whether we can resolve register swaps which are actually
line_no += 1 # perfect matches modulo compiler entropy.
# Name comes after // comment if can_resolve_register_differences(origasm, recompasm):
name = line.strip()[2:].strip() effective_ratio = 1.0
else:
ratio = 0
recinfo = syminfo.get_recompiled_address_from_name(name) percenttext = f'{(effective_ratio * 100):.2f}%'
if not recinfo: if not plain:
continue if effective_ratio == 1.0:
else: percenttext = colorama.Fore.GREEN + percenttext + colorama.Style.RESET_ALL
find_open_bracket = line elif effective_ratio > 0.8:
while '{' not in find_open_bracket: percenttext = colorama.Fore.YELLOW + percenttext + colorama.Style.RESET_ALL
find_open_bracket = srcfile.readline() else:
line_no += 1 percenttext = colorama.Fore.RED + percenttext + colorama.Style.RESET_ALL
recinfo = syminfo.get_recompiled_address(srcfilename, line_no) if effective_ratio == 1.0 and ratio != 1.0:
if not recinfo: if plain:
continue percenttext += '*'
else:
percenttext += colorama.Fore.RED + '*' + colorama.Style.RESET_ALL
# The effective_ratio is the ratio when ignoring differing register if args.print_rec_addr:
# allocation vs the ratio is the true ratio. addrs = f'0x{addr:x} / 0x{recinfo.addr:x}'
ratio = 0.0 else:
effective_ratio = 0.0 addrs = hex(addr)
if recinfo.size:
origasm = parse_asm(origfile, addr + recinfo.start, recinfo.size)
recompasm = parse_asm(recompfile, recinfo.addr + recinfo.start, recinfo.size)
diff = difflib.SequenceMatcher(None, origasm, recompasm) if not verbose:
ratio = diff.ratio() print(f' {recinfo.name} ({addrs}) is {percenttext} similar to the original')
effective_ratio = ratio
if ratio != 1.0: function_count += 1
# Check whether we can resolve register swaps which are actually total_accuracy += ratio
# perfect matches modulo compiler entropy. total_effective_accuracy += effective_ratio
if can_resolve_register_differences(origasm, recompasm):
effective_ratio = 1.0
else:
ratio = 0
percenttext = f'{(effective_ratio * 100):.2f}%' if recinfo.size:
if not plain: udiff = difflib.unified_diff(origasm, recompasm, n=10)
if effective_ratio == 1.0:
percenttext = colorama.Fore.GREEN + percenttext + colorama.Style.RESET_ALL
elif effective_ratio > 0.8:
percenttext = colorama.Fore.YELLOW + percenttext + colorama.Style.RESET_ALL
else:
percenttext = colorama.Fore.RED + percenttext + colorama.Style.RESET_ALL
if effective_ratio == 1.0 and ratio != 1.0: # If verbose, print the diff for that function to the output
if verbose:
if effective_ratio == 1.0:
ok_text = 'OK!' if plain else (colorama.Fore.GREEN + '✨ OK! ✨' + colorama.Style.RESET_ALL)
if ratio == 1.0:
print(f'{addrs}: {recinfo.name} 100% match.\n\n{ok_text}\n\n')
else:
print(f'{addrs}: {recinfo.name} Effective 100%% match. (Differs in register allocation only)\n\n{ok_text} (still differs in register allocation)\n\n')
else:
for line in udiff:
if line.startswith('++') or line.startswith('@@') or line.startswith('--'):
# Skip unneeded parts of the diff for the brief view
pass
elif line.startswith('+'):
if plain: if plain:
percenttext += '*' print(line)
else: else:
percenttext += colorama.Fore.RED + '*' + colorama.Style.RESET_ALL print(colorama.Fore.GREEN + line)
elif line.startswith('-'):
if args.print_rec_addr: if plain:
addrs = f'0x{addr:x} / 0x{recinfo.addr:x}' print(line)
else:
print(colorama.Fore.RED + line)
else: else:
addrs = hex(addr) print(line)
if not plain:
print(colorama.Style.RESET_ALL, end='')
if not verbose: print(f'\n{recinfo.name} is only {percenttext} similar to the original, diff above')
print(f' {recinfo.name} ({addrs}) is {percenttext} similar to the original')
function_count += 1 # If html, record the diffs to an HTML file
total_accuracy += ratio if html_path:
total_effective_accuracy += effective_ratio escaped = html.escape('\\n'.join(udiff).replace('"', '\\"').replace('\n', '\\n'))
htmlinsert.append(f'{{address: "0x{addr:x}", name: "{html.escape(recinfo.name)}", matching: {effective_ratio}, diff: "{escaped}"}}')
if recinfo.size:
udiff = difflib.unified_diff(origasm, recompasm, n=10)
# If verbose, print the diff for that function to the output
if verbose:
if effective_ratio == 1.0:
ok_text = 'OK!' if plain else (colorama.Fore.GREEN + '✨ OK! ✨' + colorama.Style.RESET_ALL)
if ratio == 1.0:
print(f'{addrs}: {recinfo.name} 100% match.\n\n{ok_text}\n\n')
else:
print(f'{addrs}: {recinfo.name} Effective 100%% match. (Differs in register allocation only)\n\n{ok_text} (still differs in register allocation)\n\n')
else:
for line in udiff:
if line.startswith('++') or line.startswith('@@') or line.startswith('--'):
# Skip unneeded parts of the diff for the brief view
pass
elif line.startswith('+'):
if plain:
print(line)
else:
print(colorama.Fore.GREEN + line)
elif line.startswith('-'):
if plain:
print(line)
else:
print(colorama.Fore.RED + line)
else:
print(line)
if not plain:
print(colorama.Style.RESET_ALL, end='')
print(f'\n{recinfo.name} is only {percenttext} similar to the original, diff above')
# If html, record the diffs to an HTML file
if html_path:
escaped = html.escape('\\n'.join(udiff).replace('"', '\\"').replace('\n', '\\n'))
htmlinsert.append(f'{{address: "0x{addr:x}", name: "{html.escape(recinfo.name)}", matching: {effective_ratio}, diff: "{escaped}"}}')
except UnicodeDecodeError:
break
def gen_html(html_path, data): def gen_html(html_path, data):
templatedata = None templatedata = None