Refactor/make output schema mandatory

[srijanpatel]

output_schema is made part of the config like regular schemas.
has_fixed_schema is also made part of the config. NodeSidebar is configured to not render it.

Refactoring Fixed Output Schema Handling:

• backend/app/api/node_management.py: Removed the logic for adding the output schema to the configuration for fixed output nodes.
• backend/app/nodes/base.py: Added output_schema and has_fixed_output fields to FixedOutputBaseNodeConfig and removed the abstract output_schema property from FixedOutputBaseNode. Updated the setup method to use the new output_schema field. [1] [2]
• backend/app/nodes/llm/retriever.py: Updated RetrieverNodeConfig to include the output_schema field and removed the redundant output_schema property from RetrieverNode. [1] [2]

Code Readability Improvements:

• backend/app/nodes/llm/retriever.py: Removed unused imports and reformatted code for better readability, including breaking long lines and adding newlines for separation. [1] [2] [3] [4]

Frontend Updates:

• frontend/src/components/nodes/nodeSidebar/NodeSidebar.tsx: Updated the logic to determine if a node has a fixed output by checking currentNodeConfig instead of nodeSchema. Also, prevented rendering of the has_fixed_output field. [1] [2] [3]
• frontend/src/store/nodeTypesSlice.ts: Added the has_fixed_schema field to the FlowWorkflowNodeType interface.

---

Important

Refactor to make output schema mandatory in node configuration, updating both backend and frontend to handle fixed output schemas appropriately.

• Backend Changes:
  • node_management.py: Removed logic for adding output schema to config for fixed output nodes.
  • base.py: Added output_schema and has_fixed_output to FixedOutputBaseNodeConfig, removed abstract output_schema property.
  • retriever.py: Updated RetrieverNodeConfig to include output_schema, removed redundant output_schema property.
• Frontend Changes:
  • NodeSidebar.tsx: Updated logic to check currentNodeConfig for has_fixed_output, prevented rendering of has_fixed_output field.
  • nodeTypesSlice.ts: Added has_fixed_schema to FlowWorkflowNodeType interface.

^{This description was created by for ca7e0a6. It will automatically update as commits are pushed.}

[ellipsis-dev]

👍 Looks good to me! Reviewed everything up to 5e40e5d in 1 minute and 25 seconds

More details

• Looked at 285 lines of code in 5 files
• Skipped 0 files when reviewing.
• Skipped posting 5 drafted comments based on config settings.

1. backend/app/nodes/llm/retriever.py:27

• Draft comment:
  The output_schema field is defined as Dict[str, str], but the schema types (e.g., list[RetrievalResultSchema]) are not strings. Consider using Dict[str, Any] for more accurate type representation.
• Reason this comment was not posted:
  Marked as duplicate.

2. backend/app/nodes/base.py:276

• Draft comment:
  The output_schema field is defined as Dict[str, str], but the schema types (e.g., list[RetrievalResultSchema]) are not strings. Consider using Dict[str, Any] for more accurate type representation. This comment also applies to RetrieverNodeConfig in retriever.py.
• Reason this comment was not posted:
  Decided after close inspection that this draft comment was likely wrong and/or not actionable:
  The code in create_output_model_class() shows that string type names are expected and handled properly. The field_type_to_python_type mapping converts string type names to actual types. If the type isn't in the mapping, it tries to use it as-is (line 121). This suggests the current design is intentional and working as expected.
  The comment might have a point - using Dict[str, Any] would be more technically accurate since non-string types can be passed. However, the current implementation handles both cases gracefully.
  The current implementation is working as designed - it expects string type names but has a fallback for direct type references. Changing to Dict[str, Any] wouldn't improve functionality.
  Delete the comment. The current typing is intentional and the implementation handles all cases correctly. This is not a real issue that needs fixing.

3. backend/app/nodes/llm/retriever.py:116

• Draft comment:
  Instead of using assert embedding_model_info is not None, consider raising a ValueError with a descriptive message if embedding_model_info is None. This provides clearer error handling.
• Reason this comment was not posted:
  Confidence changes required: 50%
  The assert embedding_model_info is not None is used to ensure embedding_model_info is not None, but it would be more informative to raise a specific error if this condition fails.

4. frontend/src/components/nodes/nodeSidebar/NodeSidebar.tsx:568

• Draft comment:
  The hasFixedOutput logic is repeated in multiple places. Consider centralizing this logic to avoid redundancy and potential errors. This comment also applies to line 668.
• Reason this comment was not posted:
  Confidence changes required: 50%
  The has_fixed_output field is being checked in multiple places in the frontend code. It might be beneficial to centralize this logic to avoid redundancy and potential errors.

5. frontend/src/components/nodes/nodeSidebar/NodeSidebar.tsx:668

• Draft comment:
  The hasFixedOutput logic is repeated in multiple places. Consider centralizing this logic to avoid redundancy and potential errors. This comment also applies to line 568.
• Reason this comment was not posted:
  Confidence changes required: 50%
  The has_fixed_output field is being checked in multiple places in the frontend code. It might be beneficial to centralize this logic to avoid redundancy and potential errors.

Workflow ID: wflow_6tAv2h3elhpoukbf

---

You can customize Ellipsis with 👍 / 👎 feedback, review rules, user-specific overrides, quiet mode, and more.

[ellipsis-dev]

👍 Looks good to me! Incremental review on ca7e0a6 in 57 seconds

More details

• Looked at 878 lines of code in 19 files
• Skipped 0 files when reviewing.
• Skipped posting 6 drafted comments based on config settings.

1. backend/app/nodes/llm/generative/best_of_n.py:39

• Draft comment:
  Ensure that output_json_schema is properly handled when it is not provided. Consider adding a check or default value to prevent inconsistencies in node configuration.
• Reason this comment was not posted:
  Confidence changes required: 50%
  The code in backend/app/nodes/llm/generative/best_of_n.py has a potential issue with the output_schema field. The default value for output_schema is set to {"response": "string"}. However, in the setup_subworkflow method, the output_schema is used to configure nodes, and there is a possibility that the output_json_schema is not properly handled if it is not provided. This could lead to inconsistencies in the node configuration.

2. backend/app/nodes/logic/coalesce.py:53

• Draft comment:
  The parameters __config__, __module__, __doc__, __validators__, and __cls_kwargs__ in create_model are set to None. These can be omitted for cleaner code. This applies to other instances in this file as well.
• Reason this comment was not posted:
  Confidence changes required: 30%
  In backend/app/nodes/logic/coalesce.py, the create_model function is used to dynamically create models. However, the __config__, __module__, __doc__, __validators__, and __cls_kwargs__ parameters are set to None. This is unnecessary and can be omitted for cleaner code.

3. backend/app/nodes/logic/router.py:133

• Draft comment:
  The parameters __config__, __module__, __doc__, __validators__, and __cls_kwargs__ in create_model are set to None. These can be omitted for cleaner code. This applies to other instances in this file as well.
• Reason this comment was not posted:
  Confidence changes required: 30%
  In backend/app/nodes/logic/router.py, the create_model function is used to dynamically create models. However, the __config__, __module__, __doc__, __validators__, and __cls_kwargs__ parameters are set to None. This is unnecessary and can be omitted for cleaner code.

4. backend/app/nodes/loops/base_loop_subworkflow_node.py:105

• Draft comment:
  The parameters __config__, __module__, __doc__, __validators__, and __cls_kwargs__ in create_model are set to None. These can be omitted for cleaner code.
• Reason this comment was not posted:
  Confidence changes required: 30%
  In backend/app/nodes/loops/base_loop_subworkflow_node.py, the create_model function is used to dynamically create models. However, the __config__, __module__, __doc__, __validators__, and __cls_kwargs__ parameters are set to None. This is unnecessary and can be omitted for cleaner code.

5. backend/app/nodes/primitives/input.py:69

• Draft comment:
  The parameters __config__, __module__, __doc__, __validators__, and __cls_kwargs__ in create_model are set to None. These can be omitted for cleaner code.
• Reason this comment was not posted:
  Confidence changes required: 30%
  In backend/app/nodes/primitives/input.py, the create_model function is used to dynamically create models. However, the __config__, __module__, __doc__, __validators__, and __cls_kwargs__ parameters are set to None. This is unnecessary and can be omitted for cleaner code.

6. backend/app/nodes/primitives/output.py:58

• Draft comment:
  The parameters __config__, __module__, __doc__, __validators__, and __cls_kwargs__ in create_model are set to None. These can be omitted for cleaner code.
• Reason this comment was not posted:
  Confidence changes required: 30%
  In backend/app/nodes/primitives/output.py, the create_model function is used to dynamically create models. However, the __config__, __module__, __doc__, __validators__, and __cls_kwargs__ parameters are set to None. This is unnecessary and can be omitted for cleaner code.

Workflow ID: wflow_9oLg5yO9sij0aPsN

---

You can customize Ellipsis with 👍 / 👎 feedback, review rules, user-specific overrides, quiet mode, and more.