mirror of
https://github.com/zeldaret/oot.git
synced 2024-11-29 12:04:29 +00:00
39 lines
937 B
Python
39 lines
937 B
Python
|
#!/usr/bin/python3
|
||
|
|
||
|
import argparse
|
||
|
|
||
|
def read_csv(csv_file):
|
||
|
|
||
|
f_lines = ""
|
||
|
with open(csv_file) as f:
|
||
|
f_lines = f.readlines()[1:]
|
||
|
|
||
|
overlays = {}
|
||
|
|
||
|
for line in f_lines:
|
||
|
row = line.split(",")
|
||
|
overlays[row[0]] = int(row[3])
|
||
|
|
||
|
return overlays
|
||
|
|
||
|
def main():
|
||
|
parser = argparse.ArgumentParser(description="Compares two csv produced by `get_actor_sizes.py`.")
|
||
|
parser.add_argument("firstcsv", help="")
|
||
|
parser.add_argument("secondcsv", help="")
|
||
|
args = parser.parse_args()
|
||
|
|
||
|
first = read_csv(args.firstcsv)
|
||
|
second = read_csv(args.secondcsv)
|
||
|
|
||
|
print(f"overlay,{args.firstcsv},{args.secondcsv},diff")
|
||
|
|
||
|
common_actors = set(first.keys()) & set(second.keys())
|
||
|
for actor_name in common_actors:
|
||
|
total_1 = first[actor_name]
|
||
|
total_2 = second[actor_name]
|
||
|
if total_1 != total_2:
|
||
|
print(f"{actor_name},{total_1},{total_2},{total_1-total_2}")
|
||
|
|
||
|
|
||
|
main()
|