| """Deterministic, CPU-only locale inference and output script containment. |
| |
| Count letters/marks, not shared danda punctuation or decimal digits. Devanagari |
| cannot distinguish Hindi/Marathi: auto uses Hindi; explicit Marathi is preferred. |
| """ |
| import unicodedata |
| from collections import Counter |
|
|
| BLOCKS = ((0x0900, 0x097F, 'hi'), (0x0980, 0x09FF, 'bn'), |
| (0x0A00, 0x0A7F, 'pa'), (0x0A80, 0x0AFF, 'gu'), |
| (0x0B00, 0x0B7F, 'or'), (0x0B80, 0x0BFF, 'ta'), |
| (0x0C00, 0x0C7F, 'te'), (0x0C80, 0x0CFF, 'kn'), |
| (0x0D00, 0x0D7F, 'ml')) |
|
|
|
|
| def script_counts(text): |
| counts = Counter() |
| for c in text: |
| if unicodedata.category(c)[0] not in 'LM': |
| continue |
| if unicodedata.name(c, '').startswith('LATIN'): |
| counts['en-IN'] += 1 |
| continue |
| for low, high, code in BLOCKS: |
| if low <= ord(c) <= high: |
| counts[code] += 1 |
| break |
| return counts |
|
|
|
|
| def detect_language(text): |
| counts = script_counts(text) |
| code = max(counts, key=counts.get) if counts else 'en-IN' |
| if code == 'bn' and any(c in text for c in 'ৰৱ'): |
| return 'as' |
| return code |
|
|
|
|
| def guard_output(original, result, locale): |
| allowed = set(script_counts(original)) |
| |
| if not allowed or set(script_counts(result)) <= allowed: |
| return result |
| |
| |
| if __package__: |
| from .normalize import language_table |
| else: |
| from normalize import language_table |
| locale = language_table(detect_language(original)) |
| result = ''.join(' ' + locale.digit(unicodedata.decimal(c)) + ' ' |
| if c.isdecimal() else c for c in original) |
| return ' '.join(result.split()) if set(script_counts(result)) <= allowed else original |
|
|
