15165238383

Committed 21 May 2025 02:42PM UTC coverage: 90.404% (-0.04%) from 90.443%

Build # 15165238383

Build Type

Pull #9275

github

Committed by

web-flow

Commit Message

Merge 82e69fe2c into 17432f710

Pull Request Pull Request #9275: feat: return common type in SuperComponent type compatibility check

Run Details

11135 of 12317 relevant lines covered (90.4%)

0.9 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

88.73

haystack/components/validators/json_schema.py

# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
#
# SPDX-License-Identifier: Apache-2.0

import json
from typing import Any, Dict, List, Optional

from jsonschema import ValidationError, validate

from haystack import component
from haystack.dataclasses import ChatMessage


def is_valid_json(s: str) -> bool:
    """
    Check if the provided string is a valid JSON.

    :param s: The string to be checked.
    :returns: `True` if the string is a valid JSON; otherwise, `False`.
    """
    try:
        json.loads(s)
    except ValueError:
        return False
    return True


@component
class JsonSchemaValidator:
    """
    Validates JSON content of `ChatMessage` against a specified [JSON Schema](https://json-schema.org/).

    If JSON content of a message conforms to the provided schema, the message is passed along the "validated" output.
    If the JSON content does not conform to the schema, the message is passed along the "validation_error" output.
    In the latter case, the error message is constructed using the provided `error_template` or a default template.
    These error ChatMessages can be used by LLMs in Haystack 2.x recovery loops.

    Usage example:

    ```python
    from typing import List

    from haystack import Pipeline
    from haystack.components.generators.chat import OpenAIChatGenerator
    from haystack.components.joiners import BranchJoiner
    from haystack.components.validators import JsonSchemaValidator
    from haystack import component
    from haystack.dataclasses import ChatMessage


    @component
    class MessageProducer:

        @component.output_types(messages=List[ChatMessage])
        def run(self, messages: List[ChatMessage]) -> dict:
            return {"messages": messages}


    p = Pipeline()
    p.add_component("llm", OpenAIChatGenerator(model="gpt-4-1106-preview",
                                               generation_kwargs={"response_format": {"type": "json_object"}}))
    p.add_component("schema_validator", JsonSchemaValidator())
    p.add_component("joiner_for_llm", BranchJoiner(List[ChatMessage]))
    p.add_component("message_producer", MessageProducer())

    p.connect("message_producer.messages", "joiner_for_llm")
    p.connect("joiner_for_llm", "llm")
    p.connect("llm.replies", "schema_validator.messages")
    p.connect("schema_validator.validation_error", "joiner_for_llm")

    result = p.run(data={
        "message_producer": {
            "messages":[ChatMessage.from_user("Generate JSON for person with name 'John' and age 30")]},
            "schema_validator": {
                "json_schema": {
                    "type": "object",
                    "properties": {"name": {"type": "string"},
                    "age": {"type": "integer"}
                }
            }
        }
    })
    print(result)
    >> {'schema_validator': {'validated': [ChatMessage(_role=<ChatRole.ASSISTANT: 'assistant'>,
    _content=[TextContent(text="\\n{\\n  "name": "John",\\n  "age": 30\\n}")],
    _name=None, _meta={'model': 'gpt-4-1106-preview', 'index': 0,
    'finish_reason': 'stop', 'usage': {'completion_tokens': 17, 'prompt_tokens': 20, 'total_tokens': 37}})]}}
    ```
    """

    # Default error description template
    default_error_template = (
        "The following generated JSON does not conform to the provided schema.\n"
        "Generated JSON: {failing_json}\n"
        "Error details:\n- Message: {error_message}\n"
        "- Error Path in JSON: {error_path}\n"
        "- Schema Path: {error_schema_path}\n"
        "Please match the following schema:\n"
        "{json_schema}\n"
        "and provide the corrected JSON content ONLY. Please do not output anything else than the raw corrected "
        "JSON string, this is the most important part of the task. Don't use any markdown and don't add any comment."
    )

    def __init__(self, json_schema: Optional[Dict[str, Any]] = None, error_template: Optional[str] = None):
        """
        Initialize the JsonSchemaValidator component.

        :param json_schema: A dictionary representing the [JSON schema](https://json-schema.org/) against which
            the messages' content is validated.
        :param error_template: A custom template string for formatting the error message in case of validation failure.
        """
        self.json_schema = json_schema
        self.error_template = error_template

    @component.output_types(validated=List[ChatMessage], validation_error=List[ChatMessage])
    def run(
        self,
        messages: List[ChatMessage],
        json_schema: Optional[Dict[str, Any]] = None,
        error_template: Optional[str] = None,
    ) -> Dict[str, List[ChatMessage]]:
        """
        Validates the last of the provided messages against the specified json schema.

        If it does, the message is passed along the "validated" output. If it does not, the message is passed along
        the "validation_error" output.

        :param messages: A list of ChatMessage instances to be validated. The last message in this list is the one
            that is validated.
        :param json_schema: A dictionary representing the [JSON schema](https://json-schema.org/)
            against which the messages' content is validated. If not provided, the schema from the component init
            is used.
        :param error_template: A custom template string for formatting the error message in case of validation. If not
            provided, the `error_template` from the component init is used.
        :return:  A dictionary with the following keys:
            - "validated": A list of messages if the last message is valid.
            - "validation_error": A list of messages if the last message is invalid.
        :raises ValueError: If no JSON schema is provided or if the message content is not a dictionary or a list of
            dictionaries.
        """
        last_message = messages[-1]
        if last_message.text is None:
            raise ValueError(f"The provided ChatMessage has no text. ChatMessage: {last_message}")
        if not is_valid_json(last_message.text):
            return {
                "validation_error": [
                    ChatMessage.from_user(
                        f"The message '{last_message.text}' is not a valid JSON object. "
                        f"Please provide only a valid JSON object in string format."
                        f"Don't use any markdown and don't add any comment."
                    )
                ]
            }

        last_message_content = json.loads(last_message.text)
        json_schema = json_schema or self.json_schema
        error_template = error_template or self.error_template or self.default_error_template

        if not json_schema:
            raise ValueError("Provide a JSON schema for validation either in the run method or in the component init.")
        # fc payload is json object but subtree `parameters` is string - we need to convert to json object
        # we need complete json to validate it against schema
        last_message_json = self._recursive_json_to_object(last_message_content)
        using_openai_schema: bool = self._is_openai_function_calling_schema(json_schema)
        if using_openai_schema:
            validation_schema = json_schema["parameters"]
        else:
            validation_schema = json_schema
        try:
            last_message_json = [last_message_json] if not isinstance(last_message_json, list) else last_message_json
            for content in last_message_json:
                if using_openai_schema:
                    validate(instance=content["function"]["arguments"], schema=validation_schema)
                else:
                    validate(instance=content, schema=validation_schema)

            return {"validated": [last_message]}
        except ValidationError as e:
            error_path = " -> ".join(map(str, e.absolute_path)) if e.absolute_path else "N/A"
            error_schema_path = " -> ".join(map(str, e.absolute_schema_path)) if e.absolute_schema_path else "N/A"

            error_template = error_template or self.default_error_template

            recovery_prompt = self._construct_error_recovery_message(
                error_template, str(e), error_path, error_schema_path, validation_schema, failing_json=last_message.text
            )
            return {"validation_error": [ChatMessage.from_user(recovery_prompt)]}

    def _construct_error_recovery_message(  # pylint: disable=too-many-positional-arguments
        self,
        error_template: str,
        error_message: str,
        error_path: str,
        error_schema_path: str,
        json_schema: Dict[str, Any],
        failing_json: str,
    ) -> str:
        """
        Constructs an error recovery message using a specified template or the default one if none is provided.

        :param error_template: A custom template string for formatting the error message in case of validation failure.
        :param error_message: The error message returned by the JSON schema validator.
        :param error_path: The path in the JSON content where the error occurred.
        :param error_schema_path: The path in the JSON schema where the error occurred.
        :param json_schema: The JSON schema against which the content is validated.
        :param failing_json: The generated invalid JSON string.
        """
        error_template = error_template or self.default_error_template

        return error_template.format(
            error_message=error_message,
            error_path=error_path,
            error_schema_path=error_schema_path,
            json_schema=json_schema,
            failing_json=failing_json,
        )

    def _is_openai_function_calling_schema(self, json_schema: Dict[str, Any]) -> bool:
        """
        Checks if the provided schema is a valid OpenAI function calling schema.

        :param json_schema: The JSON schema to check
        :return: `True` if the schema is a valid OpenAI function calling schema; otherwise, `False`.
        """
        return all(key in json_schema for key in ["name", "description", "parameters"])

    def _recursive_json_to_object(self, data: Any) -> Any:
        """
        Convert any string values that are valid JSON objects into dictionary objects.

        Returns a new data structure.

        :param data: The data structure to be traversed.
        :return: A new data structure with JSON strings converted to dictionary objects.
        """
        if isinstance(data, list):
            return [self._recursive_json_to_object(item) for item in data]

        if isinstance(data, dict):
            new_dict = {}
            for key, value in data.items():
                if isinstance(value, str):
                    try:
                        json_value = json.loads(value)
                        if isinstance(json_value, (dict, list)):
                            new_dict[key] = self._recursive_json_to_object(json_value)
                        else:
                            new_dict[key] = value  # Preserve the original string value
                    except json.JSONDecodeError:
                        new_dict[key] = value
                elif isinstance(value, dict):
                    new_dict[key] = self._recursive_json_to_object(value)
                else:
                    new_dict[key] = value
            return new_dict

        # If it's neither a list nor a dictionary, return the value directly
        raise ValueError("Input must be a dictionary or a list of dictionaries.")

1	# SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
2	#
3	# SPDX-License-Identifier: Apache-2.0
4
5	import json	1✔
6	from typing import Any, Dict, List, Optional	1✔
7
8	from jsonschema import ValidationError, validate	1✔
9
10	from haystack import component	1✔
11	from haystack.dataclasses import ChatMessage	1✔
12
13
14	def is_valid_json(s: str) -> bool:	1✔
15	"""
16	Check if the provided string is a valid JSON.
17
18	:param s: The string to be checked.
19	:returns: `True` if the string is a valid JSON; otherwise, `False`.
20	"""
21	try:	1✔
22	json.loads(s)	1✔
23	except ValueError:	×
24	return False	×
25	return True	1✔
26
27
28	@component	1✔
29	class JsonSchemaValidator:	1✔
30	"""
31	Validates JSON content of `ChatMessage` against a specified [JSON Schema](https://json-schema.org/).
32
33	If JSON content of a message conforms to the provided schema, the message is passed along the "validated" output.
34	If the JSON content does not conform to the schema, the message is passed along the "validation_error" output.
35	In the latter case, the error message is constructed using the provided `error_template` or a default template.
36	These error ChatMessages can be used by LLMs in Haystack 2.x recovery loops.
37
38	Usage example:
39
40	```python
41	from typing import List
42
43	from haystack import Pipeline
44	from haystack.components.generators.chat import OpenAIChatGenerator
45	from haystack.components.joiners import BranchJoiner
46	from haystack.components.validators import JsonSchemaValidator
47	from haystack import component
48	from haystack.dataclasses import ChatMessage
49
50
51	@component
52	class MessageProducer:
53
54	@component.output_types(messages=List[ChatMessage])
55	def run(self, messages: List[ChatMessage]) -> dict:
56	return {"messages": messages}
57
58
59	p = Pipeline()
60	p.add_component("llm", OpenAIChatGenerator(model="gpt-4-1106-preview",
61	generation_kwargs={"response_format": {"type": "json_object"}}))
62	p.add_component("schema_validator", JsonSchemaValidator())
63	p.add_component("joiner_for_llm", BranchJoiner(List[ChatMessage]))
64	p.add_component("message_producer", MessageProducer())
65
66	p.connect("message_producer.messages", "joiner_for_llm")
67	p.connect("joiner_for_llm", "llm")
68	p.connect("llm.replies", "schema_validator.messages")
69	p.connect("schema_validator.validation_error", "joiner_for_llm")
70
71	result = p.run(data={
72	"message_producer": {
73	"messages":[ChatMessage.from_user("Generate JSON for person with name 'John' and age 30")]},
74	"schema_validator": {
75	"json_schema": {
76	"type": "object",
77	"properties": {"name": {"type": "string"},
78	"age": {"type": "integer"}
79	}
80	}
81	}
82	})
83	print(result)
84	>> {'schema_validator': {'validated': [ChatMessage(_role=<ChatRole.ASSISTANT: 'assistant'>,
85	_content=[TextContent(text="\\n{\\n "name": "John",\\n "age": 30\\n}")],
86	_name=None, _meta={'model': 'gpt-4-1106-preview', 'index': 0,
87	'finish_reason': 'stop', 'usage': {'completion_tokens': 17, 'prompt_tokens': 20, 'total_tokens': 37}})]}}
88	```
89	"""
90
91	# Default error description template
92	default_error_template = (	1✔
93	"The following generated JSON does not conform to the provided schema.\n"
94	"Generated JSON: {failing_json}\n"
95	"Error details:\n- Message: {error_message}\n"
96	"- Error Path in JSON: {error_path}\n"
97	"- Schema Path: {error_schema_path}\n"
98	"Please match the following schema:\n"
99	"{json_schema}\n"
100	"and provide the corrected JSON content ONLY. Please do not output anything else than the raw corrected "
101	"JSON string, this is the most important part of the task. Don't use any markdown and don't add any comment."
102	)
103
104	def __init__(self, json_schema: Optional[Dict[str, Any]] = None, error_template: Optional[str] = None):	1✔
105	"""
106	Initialize the JsonSchemaValidator component.
107
108	:param json_schema: A dictionary representing the [JSON schema](https://json-schema.org/) against which
109	the messages' content is validated.
110	:param error_template: A custom template string for formatting the error message in case of validation failure.
111	"""
112	self.json_schema = json_schema	1✔
113	self.error_template = error_template	1✔
114
115	@component.output_types(validated=List[ChatMessage], validation_error=List[ChatMessage])	1✔
116	def run(	1✔
117	self,
118	messages: List[ChatMessage],
119	json_schema: Optional[Dict[str, Any]] = None,
120	error_template: Optional[str] = None,
121	) -> Dict[str, List[ChatMessage]]:
122	"""
123	Validates the last of the provided messages against the specified json schema.
124
125	If it does, the message is passed along the "validated" output. If it does not, the message is passed along
126	the "validation_error" output.
127
128	:param messages: A list of ChatMessage instances to be validated. The last message in this list is the one
129	that is validated.
130	:param json_schema: A dictionary representing the [JSON schema](https://json-schema.org/)
131	against which the messages' content is validated. If not provided, the schema from the component init
132	is used.
133	:param error_template: A custom template string for formatting the error message in case of validation. If not
134	provided, the `error_template` from the component init is used.
135	:return: A dictionary with the following keys:
136	- "validated": A list of messages if the last message is valid.
137	- "validation_error": A list of messages if the last message is invalid.
138	:raises ValueError: If no JSON schema is provided or if the message content is not a dictionary or a list of
139	dictionaries.
140	"""
141	last_message = messages[-1]	1✔
142	if last_message.text is None:	1✔
143	raise ValueError(f"The provided ChatMessage has no text. ChatMessage: {last_message}")	×
144	if not is_valid_json(last_message.text):	1✔
145	return {	×
146	"validation_error": [
147	ChatMessage.from_user(
148	f"The message '{last_message.text}' is not a valid JSON object. "
149	f"Please provide only a valid JSON object in string format."
150	f"Don't use any markdown and don't add any comment."
151	)
152	]
153	}
154
155	last_message_content = json.loads(last_message.text)	1✔
156	json_schema = json_schema or self.json_schema	1✔
157	error_template = error_template or self.error_template or self.default_error_template	1✔
158
159	if not json_schema:	1✔
160	raise ValueError("Provide a JSON schema for validation either in the run method or in the component init.")	×
161	# fc payload is json object but subtree `parameters` is string - we need to convert to json object
162	# we need complete json to validate it against schema
163	last_message_json = self._recursive_json_to_object(last_message_content)	1✔
164	using_openai_schema: bool = self._is_openai_function_calling_schema(json_schema)	1✔
165	if using_openai_schema:	1✔
166	validation_schema = json_schema["parameters"]	1✔
167	else:
168	validation_schema = json_schema	1✔
169	try:	1✔
170	last_message_json = [last_message_json] if not isinstance(last_message_json, list) else last_message_json	1✔
171	for content in last_message_json:	1✔
172	if using_openai_schema:	1✔
173	validate(instance=content["function"]["arguments"], schema=validation_schema)	1✔
174	else:
175	validate(instance=content, schema=validation_schema)	1✔
176
177	return {"validated": [last_message]}	1✔
178	except ValidationError as e:	1✔
179	error_path = " -> ".join(map(str, e.absolute_path)) if e.absolute_path else "N/A"	1✔
180	error_schema_path = " -> ".join(map(str, e.absolute_schema_path)) if e.absolute_schema_path else "N/A"	1✔
181
182	error_template = error_template or self.default_error_template	1✔
183
184	recovery_prompt = self._construct_error_recovery_message(	1✔
185	error_template, str(e), error_path, error_schema_path, validation_schema, failing_json=last_message.text
186	)
187	return {"validation_error": [ChatMessage.from_user(recovery_prompt)]}	1✔
188
189	def _construct_error_recovery_message( # pylint: disable=too-many-positional-arguments	1✔
190	self,
191	error_template: str,
192	error_message: str,
193	error_path: str,
194	error_schema_path: str,
195	json_schema: Dict[str, Any],
196	failing_json: str,
197	) -> str:
198	"""
199	Constructs an error recovery message using a specified template or the default one if none is provided.
200
201	:param error_template: A custom template string for formatting the error message in case of validation failure.
202	:param error_message: The error message returned by the JSON schema validator.
203	:param error_path: The path in the JSON content where the error occurred.
204	:param error_schema_path: The path in the JSON schema where the error occurred.
205	:param json_schema: The JSON schema against which the content is validated.
206	:param failing_json: The generated invalid JSON string.
207	"""
208	error_template = error_template or self.default_error_template	1✔
209
210	return error_template.format(	1✔
211	error_message=error_message,
212	error_path=error_path,
213	error_schema_path=error_schema_path,
214	json_schema=json_schema,
215	failing_json=failing_json,
216	)
217
218	def _is_openai_function_calling_schema(self, json_schema: Dict[str, Any]) -> bool:	1✔
219	"""
220	Checks if the provided schema is a valid OpenAI function calling schema.
221
222	:param json_schema: The JSON schema to check
223	:return: `True` if the schema is a valid OpenAI function calling schema; otherwise, `False`.
224	"""
225	return all(key in json_schema for key in ["name", "description", "parameters"])	1✔
226
227	def _recursive_json_to_object(self, data: Any) -> Any:	1✔
228	"""
229	Convert any string values that are valid JSON objects into dictionary objects.
230
231	Returns a new data structure.
232
233	:param data: The data structure to be traversed.
234	:return: A new data structure with JSON strings converted to dictionary objects.
235	"""
236	if isinstance(data, list):	1✔
237	return [self._recursive_json_to_object(item) for item in data]	1✔
238
239	if isinstance(data, dict):	1✔
240	new_dict = {}	1✔
241	for key, value in data.items():	1✔
242	if isinstance(value, str):	1✔
243	try:	1✔
244	json_value = json.loads(value)	1✔
245	if isinstance(json_value, (dict, list)):	1✔
246	new_dict[key] = self._recursive_json_to_object(json_value)	1✔
247	else:
248	new_dict[key] = value # Preserve the original string value	×
249	except json.JSONDecodeError:	1✔
250	new_dict[key] = value	1✔
251	elif isinstance(value, dict):	1✔
252	new_dict[key] = self._recursive_json_to_object(value)	1✔
253	else:
254	new_dict[key] = value	×
255	return new_dict	1✔
256
257	# If it's neither a list nor a dictionary, return the value directly
258	raise ValueError("Input must be a dictionary or a list of dictionaries.")	×

deepset-ai / haystack / 15165238383

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous