official error message is sometimes useful

[xixixi2000]

diff api.go api.go.orig
61,77d60
<
< if resp.StatusCode != 200 {
< // Try read response body as JSON
< var error_response map[string]interface{}
< err = json.NewDecoder(resp.Body).Decode(&error_response)
< if err != nil {
< c.AbortWithStatusJSON(resp.StatusCode, err)
< return
< }
< c.AbortWithStatusJSON(resp.StatusCode, gin.H{
< "error": error_response,
< })
<
< return
< }
<
< /*
103d85
< */

[linweiyuan]

不明白你意思

[xixixi2000]

sometimes we not only need to know the statuscode, but also the real reason, it's sit in the offical error message

[linweiyuan]

https://github.com/linweiyuan/go-chatgpt-api/blob/main/api/chatgpt/constant.go#L20-L26

[xixixi2000]

for example 429 stands for at least 3 reasons , not only too many requests

[linweiyuan]

除了要刷新当前对话和限流，还有哪种情况？

旧版本是有处理的，新版本被我去掉了：https://github.com/linweiyuan/go-chatgpt-api/blob/legacy/api/chatgpt/js.go#L69-L97

官网不同状态码的返回结构都不一样，我统一返回 errorMessage，除了 429 其他不适配了

[xixixi2000]

ok, if using gpt-4, needs change model, if sending message together needs wait, if conversation_id error, needs refresh, if too many messages in a fixed time, needs to change id :)

[xixixi2000]

just get the string , the struct is not important :)

[linweiyuan]

试一下会不会返回对应错误信息？我这边没 GPT-4 测试，试过需要刷新的情况是正常的

[xixixi2000]

does not work, it returns: {"details":{"errorMessage":""},"error":"error sending request"}

[xixixi2000]

this code works fine, can deal with the 403,404,412,429,500 ...

    if resp.StatusCode != 200 {
            // Try read response body as JSON
            var error_response map[string]interface{}
            err = json.NewDecoder(resp.Body).Decode(&error_response)
            if err != nil {
                    c.AbortWithStatusJSON(resp.StatusCode, err)
                    return
            }
            c.AbortWithStatusJSON(resp.StatusCode, gin.H{
                    "error": error_response,
            })

            return
    }

it returns:
{"details":{"error":{"detail":{"clears_in":5052,"code":"model_cap_exceeded","message":"You have sent too many messages to the model. Please try again later."}}},"error":"error sending request"}

[xixixi2000]

And I tried to modified 200 to http.StatusOK before, but it didn't work :(

[linweiyuan]

所以我就说它好多错误的结构都不一样，比如普通 429，它返回的就是：

{
  "detail": "Too many request. （类似）"
}

这个 GPT-4 的 429，返回的就是：

{
  "details": {
    "error": {
      "detail": {
        "clears_in": 5052,
        "code": "model_cap_exceeded",
        "message": "You have sent too many messages to the model. Please try again later."
      }
    }
  },
  "error": "error sending request"
}

---

你发的代码是要把错误信息原封不动返回给前端，然后每个前端项目都要适配一遍，目前我做的是统一返回 errorMessage，那么前端项目只需取即值即可

我改成你那种吧，错误信息会比较准确和及时

[xixixi2000]

yes, so I use a string to find the keyword :( and ignoring the struct it returns, it can deal with the error in the future, a bit ugly but works :)