mirror of
https://github.com/langgenius/dify.git
synced 2026-11-13 23:06:40 +08:00
twwu
89504b7dce
Move fixed copy into the difyBuilder i18n namespace and resolve scene IDs using reply_language. Preserve dynamic streaming, historical receipts, and submitted values. Show fixed replies immediately with a short transition, detect language before initial progress, and retire new notices and recovery messages. Keep recovery and publication validation in place. Update generated contracts and regression coverage, including progress snapshot and execution context expectations and test type annotations.
47 lines
1.8 KiB
Python
47 lines
1.8 KiB
Python
"""Detect the Builder reply language; fixed copy is localized by Web i18n."""
|
|
|
|
import logging
|
|
import re
|
|
from collections.abc import Callable
|
|
|
|
from services.dify_builder.agent import llm
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
_DETECT_SYSTEM = (
|
|
"Identify the language of the user's text. Reply with ONLY its IETF BCP-47 "
|
|
"code, e.g. en, zh-Hans, zh-Hant, ja, fr, es. No other text."
|
|
)
|
|
|
|
# A plausible BCP-47 language tag: a 2-3 letter primary subtag optionally
|
|
# followed by one or more hyphenated subtags (script/region/variant, e.g.
|
|
# "zh-Hans", "zh-Hant-HK"). Guards against a model that ignores "reply with
|
|
# ONLY the code" and returns a sentence instead.
|
|
_LANG_CODE_RE = re.compile(r"^[A-Za-z]{2,3}(-[A-Za-z0-9]{1,8})*$")
|
|
|
|
|
|
class LanguageDetector:
|
|
def __init__(self, model_provider: Callable[[], object | None]) -> None:
|
|
self._model_provider = model_provider
|
|
|
|
def detect_language(self, text: str) -> str:
|
|
if not text or not text.strip():
|
|
return "en"
|
|
try:
|
|
model = self._model_provider()
|
|
if model is None:
|
|
return "en"
|
|
raw = llm.invoke_text(model, system=_DETECT_SYSTEM, user=text[:2000]).strip()
|
|
except Exception:
|
|
logger.exception("dify_builder: language detection failed; defaulting to en")
|
|
return "en"
|
|
# Reject anything that isn't a single short, code-shaped token: the model
|
|
# disobeying "reply with ONLY the code" (e.g. "The language is Japanese.")
|
|
# must fall back to en rather than silently becoming the target language.
|
|
if not raw or re.search(r"\s", raw) or len(raw) > 35:
|
|
return "en"
|
|
code = raw.strip('."')
|
|
if not code or not _LANG_CODE_RE.match(code):
|
|
return "en"
|
|
return code
|