Skip to content

armgit / dify

Public Imported from GitHub
0

Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.

fix(api): decode complete suggested question arrays (#44916)

MindOfqc668c73ca53b2
diff --git a/api/services/message_suggested_questions_generator.py b/api/services/message_suggested_questions_generator.py
index a5d58336c49..696e0aa8d43 100644
--- a/api/services/message_suggested_questions_generator.py
+++ b/api/services/message_suggested_questions_generator.py
@@ -11,7 +11,6 @@ from flask import current_app
from core.app.entities.app_invoke_entities import CreditUsageCreatedBy, get_credit_usage_app_type
from core.llm_generator.llm_generator import _normalize_completion_params
-from core.llm_generator.output_parser.suggested_questions_after_answer import SuggestedQuestionsAfterAnswerOutputParser
from core.model_context import use_credit_usage_metadata, with_credit_usage_created_by
from core.model_manager import ModelInstance, ModelManager
from core.ops.entities.trace_entity import TraceTaskName
@@ -23,6 +22,7 @@ from extensions.ext_database import db
from graphon.model_runtime.entities.llm_entities import LLMResult
from graphon.model_runtime.entities.message_entities import PromptMessage, UserPromptMessage
from graphon.model_runtime.entities.model_entities import ModelType, ParameterType
+from services.message_suggested_questions_parser import SuggestedQuestionsAfterAnswerOutputParser
if TYPE_CHECKING:
from core.memory.token_buffer_memory import PreparedHistory
diff --git a/api/core/llm_generator/output_parser/suggested_questions_after_answer.py b/api/services/message_suggested_questions_parser.py
similarity index 81%
rename from api/core/llm_generator/output_parser/suggested_questions_after_answer.py
rename to api/services/message_suggested_questions_parser.py
index 7ac340926db..4b74264d759 100644
--- a/api/core/llm_generator/output_parser/suggested_questions_after_answer.py
+++ b/api/services/message_suggested_questions_parser.py
@@ -1,6 +1,7 @@
+"""Parse model responses for the shared suggested-questions service."""
+
import json
import logging
-import re
from collections.abc import Sequence
from core.llm_generator.prompts import DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT
@@ -24,11 +25,12 @@ class SuggestedQuestionsAfterAnswerOutputParser:
def parse(self, text: str) -> Sequence[str]:
stripped_text = text.strip()
- action_match = re.search(r"\[.*?\]", stripped_text, re.DOTALL)
+ array_start = stripped_text.find("[")
questions: list[str] = []
- if action_match is not None:
+ if array_start != -1:
try:
- json_obj = json.loads(action_match.group(0).strip())
+ # Let the decoder honor quoted brackets, escapes, and nested values.
+ json_obj, _ = json.JSONDecoder().raw_decode(stripped_text[array_start:])
except json.JSONDecodeError as exc:
logger.warning("Failed to decode suggested questions payload: %s", exc)
else:
diff --git a/api/tests/unit_tests/services/test_message_suggested_questions_generator.py b/api/tests/unit_tests/services/test_message_suggested_questions_generator.py
index 96846ef7c32..3df7c726f8b 100644
--- a/api/tests/unit_tests/services/test_message_suggested_questions_generator.py
+++ b/api/tests/unit_tests/services/test_message_suggested_questions_generator.py
@@ -19,8 +19,6 @@ from yarl import URL
import core.plugin.impl.base as plugin_base
import core.plugin.impl.model_runtime as model_runtime_module
from core.llm_generator.llm_generator import _normalize_completion_params
-from core.llm_generator.output_parser.suggested_questions_after_answer import SuggestedQuestionsAfterAnswerOutputParser
-from core.llm_generator.prompts import DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT
from core.memory.token_buffer_memory import HistoryPrompt, PreparedHistory
from core.model_context import get_credit_usage_metadata
from core.plugin.entities.plugin_daemon import PluginDaemonInnerError
@@ -177,34 +175,6 @@ def test_token_transport_failure_propagates_before_best_effort_generation(
assert _get_plugin_daemon_request_timeout() == timeout
- ("text", "expected"),
- [
- ('["Question 1?", "还有吗?"]', ["Question 1?", "还有吗?"]),
- ('Here are questions:\n["Next?"]', ["Next?"]),
- ('["Next?", 1, null, true]', ["Next?"]),
- ("[]", []),
- ("", []),
- ("not a question array", []),
- ('["unfinished]', []),
- ],
-)
-def test_response_parser_handles_question_arrays(text: str, expected: list[str]) -> None:
- assert SuggestedQuestionsAfterAnswerOutputParser().parse(text) == expected
-
-
[email protected]("instruction", [None, "", " "])
-def test_empty_instruction_uses_default_prompt(instruction: str | None) -> None:
- parser = SuggestedQuestionsAfterAnswerOutputParser(instruction_prompt=instruction)
- assert parser.get_format_instructions() == DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT
-
-
-def test_custom_instruction_preserves_json_output_requirement() -> None:
- instructions = SuggestedQuestionsAfterAnswerOutputParser("Ask a follow-up").get_format_instructions()
- assert instructions.startswith("Ask a follow-up\n")
- assert 'JSON array like ["question1", "question2", "question3"]' in instructions
-
-
@pytest.mark.parametrize("limit", [0, -1, -0.5])
def test_non_positive_token_limits_are_removed_without_mutating_configuration(limit: int | float) -> None:
configured: dict[str, object] = {
diff --git a/api/tests/unit_tests/services/test_message_suggested_questions_parser.py b/api/tests/unit_tests/services/test_message_suggested_questions_parser.py
new file mode 100644
index 00000000000..5d4e018fd9b
--- /dev/null
+++ b/api/tests/unit_tests/services/test_message_suggested_questions_parser.py
@@ -0,0 +1,54 @@
+"""Suggested-question payload parsing without model/provider I/O."""
+
+import json
+
+import pytest
+
+from core.llm_generator.prompts import DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT
+from services.message_suggested_questions_parser import SuggestedQuestionsAfterAnswerOutputParser
+
+
+ ("text", "expected"),
+ [
+ ('["How do I use arr[0]?", "Next?"]', ["How do I use arr[0]?", "Next?"]),
+ ('["What does ] mean?", "And [?"]', ["What does ] mean?", "And [?"]),
+ ('Questions:\n```json\n["Can I use [a-z]?", "Next?"]\n```', ["Can I use [a-z]?", "Next?"]),
+ (json.dumps(['How do I quote "]"?', "Next?"]), ['How do I quote "]"?', "Next?"]),
+ ('["Next?", ["nested"], {"name": "value"}, 1, null, true]', ["Next?"]),
+ ('["Next?"] trailing prose', ["Next?"]),
+ ('["unfinished]', []),
+ ("not a question array", []),
+ ("[]", []),
+ ],
+)
+def test_parse_suggested_question_payload(text: str, expected: list[str]) -> None:
+ assert SuggestedQuestionsAfterAnswerOutputParser().parse(text) == expected
+
+
+ ("text", "expected"),
+ [
+ ('["Question 1?", "还有吗?"]', ["Question 1?", "还有吗?"]),
+ ('Here are questions:\n["Next?"]', ["Next?"]),
+ ('["Next?", 1, null, true]', ["Next?"]),
+ ("[]", []),
+ ("", []),
+ ("not a question array", []),
+ ('["unfinished]', []),
+ ],
+)
+def test_response_parser_handles_question_arrays(text: str, expected: list[str]) -> None:
+ assert SuggestedQuestionsAfterAnswerOutputParser().parse(text) == expected
+
+
[email protected]("instruction", [None, "", " "])
+def test_empty_instruction_uses_default_prompt(instruction: str | None) -> None:
+ parser = SuggestedQuestionsAfterAnswerOutputParser(instruction_prompt=instruction)
+ assert parser.get_format_instructions() == DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT
+
+
+def test_custom_instruction_preserves_json_output_requirement() -> None:
+ instructions = SuggestedQuestionsAfterAnswerOutputParser("Ask a follow-up").get_format_instructions()
+ assert instructions.startswith("Ask a follow-up\n")
+ assert 'JSON array like ["question1", "question2", "question3"]' in instructions