NumberFormatException when model returns scientific notation (e.g., “1.5009969E7”) for int parameter in tool method

[Oren-Peer]

Bug description
When using the @tool annotation to define a method with an int parameter, and the model returns a numeric value in scientific notation (e.g., "1.5009969E7"), Spring AI’s internal JsonParser attempts to parse it using Integer.parseInt, which fails with a NumberFormatException.

java.lang.NumberFormatException: For input string: \"1.5009969E7\"
	 java.base/java.lang.NumberFormatException.forInputString(NumberFormatException.java:67) ~[?:?]
	 java.base/java.lang.Integer.parseInt(Integer.java:588) ~[?:?]
	 java.base/java.lang.Integer.parseInt(Integer.java:685) ~[?:?]
	 org.springframework.ai.util.json.JsonParser.toTypedObject(JsonParser.java:131) ~[spring-ai-model-1.0.0-M8.jar:1.0.0-M8]
	 org.springframework.ai.tool.method.MethodToolCallback.buildTypedArgument(MethodToolCallback.java:148) ~[spring-ai-model-1.0.0-M8.jar:1.0.0-M8]
	 org.springframework.ai.tool.method.MethodToolCallback.lambda$buildMethodArguments$1(MethodToolCallback.java:139) ~[spring-ai-model-1.0.0-M8.jar:1.0.0-M8]
	 java.base/java.util.stream.ReferencePipeline$3$1.accept(ReferencePipeline.java:215) ~[?:?]
	 java.base/java.util.Spliterators$ArraySpliterator.forEachRemaining(Spliterators.java:1024) ~[?:?]
	 java.base/java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:570) ~[?:?]
	 java.base/java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:560) ~[?:?]
	 java.base/java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:636) ~[?:?]
	 java.base/java.util.stream.AbstractPipeline.evaluateToArrayNode(AbstractPipeline.java:291) ~[?:?]
	 java.base/java.util.stream.ReferencePipeline.toArray(ReferencePipeline.java:656) ~[?:?]
	 java.base/java.util.stream.ReferencePipeline.toArray(ReferencePipeline.java:662) ~[?:?]
	 org.springframework.ai.tool.method.MethodToolCallback.buildMethodArguments(MethodToolCallback.java:140) ~[spring-ai-model-1.0.0-M8.jar:1.0.0-M8]
	 org.springframework.ai.tool.method.MethodToolCallback.call(MethodToolCallback.java:107) ~[spring-ai-model-1.0.0-M8.jar:1.0.0-M8]
	 org.springframework.ai.model.tool.DefaultToolCallingManager.executeToolCall(DefaultToolCallingManager.java:205) ~[spring-ai-model-1.0.0-M8.jar:1.0.0-M8]
	 org.springframework.ai.model.tool.DefaultToolCallingManager.executeToolCalls(DefaultToolCallingManager.java:128) ~[spring-ai-model-1.0.0-M8.jar:1.0.0-M8]
	 org.springframework.ai.vertexai.gemini.schema.VertexToolCallingManager.executeToolCalls(VertexToolCallingManager.java:94) ~[spring-ai-vertex-ai-gemini-1.0.0-M8.jar:1.0.0-M8]
	 org.springframework.ai.vertexai.gemini.VertexAiGeminiChatModel.lambda$internalStream$8(VertexAiGeminiChatModel.java:514) ~[spring-ai-vertex-ai-gemini-1.0.0-M8.jar:1.0.0-M8]
	 reactor.core.publisher.FluxDefer.subscribe(FluxDefer.java:46) ~[reactor-core-3.7.2.jar:3.7.2]
	 reactor.core.publisher.FluxSubscribeOn$SubscribeOnSubscriber.run(FluxSubscribeOn.java:194) ~[reactor-core-3.7.2.jar:3.7.2]
	 reactor.core.scheduler.WorkerTask.call(WorkerTask.java:84) ~[reactor-core-3.7.2.jar:3.7.2]
	 reactor.core.scheduler.WorkerTask.call(WorkerTask.java:37) ~[reactor-core-3.7.2.jar:3.7.2]
	 java.base/java.util.concurrent.FutureTask.run(FutureTask.java:317) ~[?:?]
	 java.base/java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:304) ~[?:?]
	 java.base/java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1144) ~[?:?]
	 java.base/java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642) ~[?:?]
	 java.base/java.lang.Thread.run(Thread.java:1575) [?:?]

Environment
• Spring AI version: 1.0.0-M8
• Java version: 22
• Model: Vertex AI Gemini
• Tool method used with @tool

Steps to reproduce

1. Create a tool method with a numeric parameter:

@Tool(description = "Get the Object by its id")
public Object getById(int id) {
    ...
}

2. Have the model generate a response calling the tool with a large value like "1.5009969E7" (i.e., scientific notation).
3. The framework tries to convert this value to int using Integer.parseInt(...), which fails.
   org.springframework.ai.util.json.JsonParser

[dev-jonghoonpark]

It seems that the parser handled the input well.
Is it not possible to receive it as a double and then convert it to an int?

[markpollack]

Can you replace the method in JsonParser with this and see if it works.

@SuppressWarnings("unchecked")
public static Object toTypedObject(Object value, Class<?> type) {
    Assert.notNull(value, "value cannot be null");
    Assert.notNull(type, "type cannot be null");

    var javaType = ClassUtils.resolvePrimitiveIfNecessary(type);

    if (javaType.isEnum()) {
        return Enum.valueOf((Class<Enum>) javaType, value.toString());
    }

    // Delegate everything else to ObjectMapper
    return OBJECT_MAPPER.convertValue(value, javaType);
}

[Oren-Peer]

An option is to use BigDecimal and then parse it to an int or long

public static Object toTypedObject(Object value, Class<?> type) {
		Assert.notNull(value, "value cannot be null");
		Assert.notNull(type, "type cannot be null");

		var javaType = ClassUtils.resolvePrimitiveIfNecessary(type);

		if (javaType == String.class) {
			return value.toString();
		}
		else if (javaType == Byte.class) {
			return Byte.parseByte(value.toString());
		}
		else if (javaType == Integer.class) {
			BigDecimal decimal = new BigDecimal(value.toString());
			return decimal.intValueExact();
		}
		else if (javaType == Short.class) {
			return Short.parseShort(value.toString());
		}
		else if (javaType == Long.class) {
			BigDecimal decimal = new BigDecimal(value.toString());
			return decimal.longValueExact();
		}
		else if (javaType == Double.class) {
			return Double.parseDouble(value.toString());
		}
		else if (javaType == Float.class) {
			return Float.parseFloat(value.toString());
		}
		else if (javaType == Boolean.class) {
			return Boolean.parseBoolean(value.toString());
		}
		else if (javaType.isEnum()) {
			return Enum.valueOf((Class<Enum>) javaType, value.toString());
		}

		String json = JsonParser.toJson(value);
		return JsonParser.fromJson(json, javaType);
	}

[BeomSeogKim]

Hello, I hope this message finds you well.
I understand that this issue is currently assigned to @tzolov.
As someone eager to make my first contribution to open source, I have reviewed the issue and believe I can provide a solution.
If @tzolov is not actively working on this at the moment, would it be possible for me to take on this issue?
I would greatly appreciate the opportunity to contribute.
Thank you for considering my request.

[markpollack]

yes, I can assign it to you.

I like the solution suggested

		else if (javaType == Integer.class) {
			BigDecimal decimal = new BigDecimal(value.toString());
			return decimal.intValueExact();
		}

as it is the least invasive.