youngkingdom/dify
1
0
Fork 0
mirror of https://github.com/langgenius/dify.git synced 2026-05-05 01:48:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: ensure sub-graph modal syncs immediately when

Browse Source 
applying generated code.
This commit is contained in:
zhsama
2026-02-05 06:06:38 +08:00
parent 52b34b1fdb
commit a2380c4fd3
4 changed files with 85 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
api/core/llm_generator/llm_generator.py
View File

@ -14,6 +14,7 @@ from core.llm_generator.context_models import (
)
from core.llm_generator.entities import RuleCodeGeneratePayload, RuleGeneratePayload, RuleStructuredOutputPayload
from core.llm_generator.output_models import (
CodeNodeOutputItem,
CodeNodeStructuredOutput,
InstructionModifyOutput,
SuggestedQuestionsOutput,
@ -479,8 +480,10 @@ class LLMGenerator:
model_parameters=model_parameters,
)
response_payload = response.model_dump()
response_payload["outputs"] = cls._format_code_outputs(response.outputs)
return {
**response.model_dump(),
**response_payload,
"code_language": language,
"error": "",
}
@ -503,6 +506,20 @@ class LLMGenerator:
"error": error,
}
@classmethod
def _format_code_outputs(cls, outputs: Sequence[CodeNodeOutputItem]) -> dict[str, dict[str, str]]:
"""Normalize code outputs to a stable mapping for frontend consumers.
The LLM structured output uses an array to satisfy strict-mode schemas, but the
frontend expects a name-to-type mapping for Code node outputs.
"""
mapped: dict[str, dict[str, str]] = {}
for output_item in outputs:
if not output_item.name:
continue
mapped[output_item.name] = {"type": str(output_item.type)}
return mapped
@classmethod
def generate_suggested_questions(
cls,
@ -557,10 +574,14 @@ class LLMGenerator:
completion_params = model_config.get("completion_params", {}) if model_config else {}
try:
response = invoke_llm_with_pydantic_model(provider=model_instance.provider, model_schema=model_schema,
model_instance=model_instance, prompt_messages=prompt_messages,
output_model=SuggestedQuestionsOutput,
model_parameters=completion_params)
response = invoke_llm_with_pydantic_model(
provider=model_instance.provider,
model_schema=model_schema,
model_instance=model_instance,
prompt_messages=prompt_messages,
output_model=SuggestedQuestionsOutput,
model_parameters=completion_params,
)
return {"questions": response.questions, "error": ""}