From 44c9f415b778c63c655853d6be8cfdcf38deb188 Mon Sep 17 00:00:00 2001
From: "tao.chen" <93983997+taochen-ct@users.noreply.github.com>
Date: Mon, 13 Jul 2026 19:55:16 +0800
Subject: [PATCH] =?UTF-8?q?tools:=20=E4=BF=AE=20spark=5Fsubmit=20=E6=A0=A1?=
 =?UTF-8?q?=E9=AA=8C=E7=9A=84=E4=B8=A4=E5=A4=84=E5=B0=8F=E9=97=AE=E9=A2=98?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

c9e54a3 引入了 server-minted 校验, 但有两处需要调整:

1. 错误信息用 fmt.Errorf(... %w).Error() 构造, %w 包装语义被丢
   掉 (errResult 接收 string, Error() 之后 %w 已经不可达). 改用
   fmt.Sprintf 拼接, %q 引用路径, 错误信息保持不变但代码不再误
   导.

2. 校验放在 scriptPath 解析之后、queue 解析之前. 错误信息会按字
   段出现顺序报 (cluster_id, master, deploy_mode, script_path,
   queue, ...), 但当前顺序是 script_path 校验先报, 然后才报 queue
   缺失. 把校验挪到所有 RequireString 之后、parseStringMap 之前,
   LLM 看错误时字段顺序跟 schema 顺序一致.

nil-safe 校验逻辑不变 (d.UploadStore == nil 时跳过). 现有测试
全部通过:
  - TestSparkSubmit_RejectsNonMintedPath 仍通过 (校验位置不影响
    行为)
  - TestSparkSubmit_StructuredCommand 仍通过 (走 mint 路径)

Co-Authored-By: Claude <noreply@anthropic.com>
---
 internal/mcp/tools/spark_submit.go | 27 +++++++++++++++------------
 1 file changed, 15 insertions(+), 12 deletions(-)

diff --git a/internal/mcp/tools/spark_submit.go b/internal/mcp/tools/spark_submit.go
index 3c79078..33c83d7 100644
--- a/internal/mcp/tools/spark_submit.go
+++ b/internal/mcp/tools/spark_submit.go
@@ -80,26 +80,29 @@ func (d *Deps) SparkSubmitHandler(ctx context.Context, req mcp.CallToolRequest)
 		return errResult("spark_submit: " + err.Error()), nil
 	}
 
-	// Reject paths that didn't come through upload_file on this server.
-	// Cluster-local paths and arbitrary host paths are unreachable from the MCP
-	// server's spark-submit, so we fail fast with a clear message rather than
-	// letting spark-submit produce a confusing FileNotFoundException later.
-	if d.UploadStore != nil {
-		if _, err := d.UploadStore.Validate(scriptPath); err != nil {
-			return errResult(fmt.Errorf("spark_submit: script_path %q is not from upload_file on this server; upload the file first via upload_file and pass back the path field: %w", scriptPath, err).Error()), nil
-		}
+	executorMemory, err := req.RequireString("executor_memory")
+	if err != nil {
+		return errResult("spark_submit: " + err.Error()), nil
 	}
 	queue, err := req.RequireString("queue")
 	if err != nil {
 		return errResult("spark_submit: " + err.Error()), nil
 	}
-	executorMemory, err := req.RequireString("executor_memory")
-	if err != nil {
-		return errResult("spark_submit: " + err.Error()), nil
-	}
 
 	args := req.GetArguments()
 
+	// Reject paths that didn't come through upload_file on this server.
+	// Cluster-local paths and arbitrary host paths are unreachable from the MCP
+	// server's spark-submit, so we fail fast with a clear message rather than
+	// letting spark-submit produce a confusing FileNotFoundException later.
+	// Placed after all required-string parses so error messages surface in
+	// field order (cluster_id, master, ..., script_path, queue, ...).
+	if d.UploadStore != nil {
+		if _, err := d.UploadStore.Validate(scriptPath); err != nil {
+			return errResult(fmt.Sprintf("spark_submit: script_path %q is not from upload_file on this server; upload the file first via upload_file and pass back the path field", scriptPath)), nil
+		}
+	}
+
 	// deploy_mode is required for forward compatibility but the builder does not
 	// emit --deploy-mode, matching the Python reference's actual behavior. If the
 	// Go side needs --deploy-mode later, add it here and document the divergence.