DevFlow/crates/df-ai/examples/glm_stream_test.rs

//! 独立诊断:用 df-ai 真实调用 GLM anthropic 流式端点,验证
//! sse_parser + apply_anthropic_event + provider.stream() 整条链路。
//!
//! 二分定位「发消息卡掉」: 若本例能正常吐 chunk → provider 层(df-ai)OK,
//! 问题在 devflow 应用层(provider 配置/emit/前端); 若卡/空/Err → df-ai 有 bug。
//!
//! 运行: cd crates/df-ai && cargo run --example glm_stream_test
use df_ai::build_provider;
use df_ai_core::CompletionRequest;
use futures::StreamExt;

#[tokio::main]
async fn main() {
    let token = std::env::var("ANTHROPIC_AUTH_TOKEN")
        .or_else(|_| std::env::var("ANTHROPIC_API_KEY"))
        .expect("需要环境变量 ANTHROPIC_AUTH_TOKEN");
    eprintln!("[glm-test] token len={}", token.len());

    let provider = build_provider(
        "anthropic",
        "https://open.bigmodel.cn/api/anthropic",
        &token,
        "glm-5.2",
    );

    // 用 JSON 反序列化构造请求,绕开字段列表(devflow 实际用 glm-5.2)
    let req: CompletionRequest = serde_json::from_str(
        r#"{"model":"glm-5.2","stream":true,"max_tokens":16,"messages":[{"role":"user","content":"说你好"}]}"#,
    )
    .expect("parse CompletionRequest");

    eprintln!("[glm-test] 调用 provider.stream() ...");
    let t0 = std::time::Instant::now();
    let mut s = match provider.stream(req).await {
        Ok(s) => {
            eprintln!("[glm-test] stream() Ok, 建连耗时 {:?}", t0.elapsed());
            s
        }
        Err(e) => {
            eprintln!("[glm-test] stream() Err: {:#}", e);
            return;
        }
    };

    let mut n = 0;
    let mut got_text = false;
    while let Some(chunk_result) = s.next().await {
        n += 1;
        match chunk_result {
            Ok(chunk) => {
                if !chunk.delta.is_empty() {
                    got_text = true;
                }
                eprintln!(
                    "[glm-test] chunk#{} delta={:?} reasoning={:?} finished={} usage={:?} err={:?}",
                    n, chunk.delta,
                    chunk.reasoning_content.as_deref().map(|s| if s.len() > 30 { format!("{}..", &s[..30]) } else { s.to_string() }),
                    chunk.finished, chunk.usage, chunk.error
                );
            }
            Err(e) => {
                eprintln!("[glm-test] chunk#{} Err: {}", n, e);
            }
        }
        if n > 60 {
            eprintln!("[glm-test] 超 60 chunk 截断");
            break;
        }
    }
    eprintln!(
        "[glm-test] 流结束 共 {} chunk, 是否拿到文本={}, 总耗时 {:?}",
        n,
        got_text,
        t0.elapsed()
    );
}