bugfix: eliminate use-after-move undefined behavior.

xllm/api_service/chat_service_impl.cpp

@@ -625,6 +625,10 @@ void ChatServiceImpl::process_async_impl(std::shared_ptr<ChatCall> call) {
     CHECK(stream_parser != nullptr) << "create StreamOutputParser failed!";
   }
 
+  auto saved_tools = request_params.tools;
+  auto saved_streaming = request_params.streaming;
+  auto saved_request_id = request_params.request_id;
+
   master_->handle_request(
       std::move(messages),
       std::move(prompt_tokens),
@@ -633,12 +637,12 @@ void ChatServiceImpl::process_async_impl(std::shared_ptr<ChatCall> call) {
       [call,
        model,
        master = master_,
-       stream = request_params.streaming,
+       stream = std::move(saved_streaming),
        include_usage = include_usage,
        first_message_sent = std::unordered_set<size_t>(),
-       request_id = request_params.request_id,
+       request_id = std::move(saved_request_id),
        created_time = absl::ToUnixSeconds(absl::Now()),
-       json_tools = request_params.tools,
+       json_tools = std::move(saved_tools),
        tool_call_parser_format = tool_call_parser_format_,
        reasoning_parser_format = reasoning_parser_format_,
        is_force_reasoning = is_force_reasoning_,
@@ -741,6 +745,9 @@ void MMChatServiceImpl::process_async_impl(std::shared_ptr<MMChatCall> call) {
     include_usage = rpc_request.stream_options().include_usage();
   }
 
+  auto saved_streaming = request_params.streaming;
+  auto saved_request_id = request_params.request_id;
+
   // schedule the request
   master_->handle_request(
       std::move(messages),
@@ -749,10 +756,10 @@ void MMChatServiceImpl::process_async_impl(std::shared_ptr<MMChatCall> call) {
       [call,
        model,
        master = master_,
-       stream = request_params.streaming,
+       stream = std::move(saved_streaming),
        include_usage = include_usage,
        first_message_sent = std::unordered_set<size_t>(),
-       request_id = request_params.request_id,
+       request_id = std::move(saved_request_id),
        created_time = absl::ToUnixSeconds(absl::Now())](
           const RequestOutput& req_output) mutable -> bool {
         if (req_output.status.has_value()) {

xllm/api_service/completion_service_impl.cpp

@@ -196,6 +196,8 @@ void CompletionServiceImpl::process_async_impl(
     request_params.decode_address = rpc_request.routing().decode_name();
   }
 
+  auto saved_streaming = request_params.streaming;
+  auto saved_request_id = request_params.request_id;
   // schedule the request
   master_->handle_request(
       std::move(rpc_request.prompt()),
@@ -205,9 +207,9 @@ void CompletionServiceImpl::process_async_impl(
       [call,
        model,
        master = master_,
-       stream = request_params.streaming,
+       stream = std::move(saved_streaming),
        include_usage = include_usage,
-       request_id = request_params.request_id,
+       request_id = std::move(saved_request_id),
        created_time = absl::ToUnixSeconds(absl::Now())](
           const RequestOutput& req_output) -> bool {
         if (req_output.status.has_value()) {

xllm/api_service/embedding_service_impl.cpp

@@ -94,6 +94,7 @@ void EmbeddingServiceImpl::process_async_impl(
   // TODO only support input_str for now
   auto& input = rpc_request.input();
 
+  auto saved_request_id = request_params.request_id;
   // schedule the request
   master_->handle_request(
       std::move(input),
@@ -102,7 +103,7 @@ void EmbeddingServiceImpl::process_async_impl(
       call.get(),
       [call,
        model,
-       request_id = request_params.request_id,
+       request_id = std::move(saved_request_id),
        created_time = absl::ToUnixSeconds(absl::Now())](
           const RequestOutput& req_output) -> bool {
         if (req_output.status.has_value()) {

xllm/api_service/image_generation_service_impl.cpp

@@ -83,13 +83,14 @@ void ImageGenerationServiceImpl::process_async_impl(
   DiTRequestParams request_params(
       rpc_request, call->get_x_request_id(), call->get_x_request_time());
 
+  auto saved_request_id = request_params.request_id;
   // schedule the request
   master_->handle_request(
       std::move(request_params),
       call.get(),
       [call,
        model,
-       request_id = request_params.request_id,
+       request_id = std::move(saved_request_id),
        created_time = absl::ToUnixSeconds(absl::Now())](
           const DiTRequestOutput& req_output) -> bool {
         if (req_output.status.has_value()) {