1
0
Fork 0
VoxeLibre/tools/spell_check_translate_files.py

61 lines
2.1 KiB
Python
Raw Normal View History

# Output indicator
# >> Spell checking error
# !> Indicates a text line with too many '=' in translate file (.tr)
import os
import argparse
import hunspell
import re
parser = argparse.ArgumentParser(description='Check translation file using pyhunspell for a given language.')
parser.add_argument("language", help='language code')
parser.add_argument("dic_file", help='path to .dic file')
parser.add_argument("aff_file", help='path to .aff file')
parser.add_argument('-a', "--add", help='path to file with additional words', default=None)
parser.add_argument('-s', "--suggestions", action="store_true", help='display spelling suggestions for incorrectly spelled words')
args = parser.parse_args()
PATH = "../mods/"
code_lang = args.language
hs = hunspell.HunSpell(args.dic_file, args.aff_file)
if args.add is not None:
with open(args.add, 'r') as file:
for word in file:
hs.add(word.strip())
def get_errors(file):
result = set()
for line in file:
line = line.strip()
if line.startswith('#') or line == '':
continue
if '=' in line:
try:
_, translated = re.split(r'[^@]=', line)
except:
print("!> Too many '='s in line:", line)
continue
for word in re.split(r'\@.|[\W ]',translated):
if not hs.spell(word):
result.add(word)
return result
def spell_check(filename):
with open(filename, 'r', encoding="utf-8") as file:
errors = get_errors(file)
if len(errors) > 0:
print("Spell checking errors in '", filename[len(PATH):], "':", sep='')
for word in errors:
print('>>', word)
if args.suggestions:
print(">> Did you mean:", ", ".join(hs.suggest(word)), "?")
for root, _, _ in os.walk(PATH):
if root.endswith('locale'):
translated_file = os.path.join(root, os.path.basename(os.path.dirname(root))) + "." + code_lang + ".tr"
if os.path.exists(translated_file) and os.path.isfile(translated_file):
spell_check(translated_file)