调用AzureOpenAIModelFactory.StreamingChatCompletions触发NullReferenceException问题

问题根源 #

调用AzureOpenAIModelFactory.StreamingChatCompletions()时立即触发System.NullReferenceException，问题出在StreamingChatCompletions类的内部构造函数：私有只读字段_baseChatCompletions未被初始化，构造函数直接调用其Add()方法，导致空引用异常。相关代码片段如下：

public class StreamingChatCompletions : IDisposable
{
    private readonly IList<ChatCompletions> _baseChatCompletions;

    //...
        
    internal StreamingChatCompletions(
            ChatCompletions baseChatCompletions = null,
            List<StreamingChatChoice> streamingChatChoices = null)
    {
        _baseChatCompletions.Add(baseChatCompletions);
        _streamingChatChoices = streamingChatChoices;
        _streamingTaskComplete = true;
    }
}

解决方案 #

1. 初始化私有字段_baseChatCompletions #

由于_baseChatCompletions是readonly类型，必须在声明时或构造函数内完成初始化。推荐在声明时直接初始化：

private readonly IList<ChatCompletions> _baseChatCompletions = new List<ChatCompletions>();

或者在构造函数开头初始化：

internal StreamingChatCompletions(
        ChatCompletions baseChatCompletions = null,
        List<StreamingChatChoice> streamingChatChoices = null)
{
    _baseChatCompletions = new List<ChatCompletions>();
    // 后续逻辑...
}

2. 处理可选参数的null情况 #

baseChatCompletions是可选参数，可能传入null，调用Add()前需判断，避免向列表中添加null元素（根据业务需求调整）：

if (baseChatCompletions != null)
{
    _baseChatCompletions.Add(baseChatCompletions);
}

完整修正后的构造函数示例 #

public class StreamingChatCompletions : IDisposable
{
    private readonly IList<ChatCompletions> _baseChatCompletions = new List<ChatCompletions>();
    private List<StreamingChatChoice> _streamingChatChoices;
    private bool _streamingTaskComplete;

    internal StreamingChatCompletions(
            ChatCompletions baseChatCompletions = null,
            List<StreamingChatChoice> streamingChatChoices = null)
    {
        if (baseChatCompletions != null)
        {
            _baseChatCompletions.Add(baseChatCompletions);
        }
        _streamingChatChoices = streamingChatChoices;
        _streamingTaskComplete = true;
    }

    // IDisposable接口实现...
}

内容的提问来源于stack exchange，提问作者Intelligence6