Function read_http_request

ds4_server.c:11120–11166

    if (r && r->model_syntax == SERVER_MODEL_SYNTAX_GLM) {
        return build_invalid_glm_tool_error_suffix(r, thinking, detail);
    }
    return build_invalid_dsml_tool_error_suffix(r, thinking, detail);
}

static int server_next_prefill_slot_locked(const server *s) {
    if (!s || s->slot_count <= 0) return -1;
    for (int n = 1; n <= s->slot_count; n++) {
        int id = (s->last_prefill_slot + n) % s->slot_count;
        if (s->slots[id].prefill_waiting) return id;
    }
    return -1;
}

static bool server_prefill_enter(server *s, server_slot *slot) {
    if (!s || !slot || slot_job_cancelled(slot)) return false;
    if (!s->batched_mode) {
        pthread_mutex_lock(&s->inference_mu);
        if (slot_job_cancelled(slot)) {
            pthread_mutex_unlock(&s->inference_mu);
            return false;
        }
        return true;
    }

    pthread_mutex_lock(&s->model_mu);
    slot->prefill_waiting = true;
    pthread_cond_broadcast(&s->model_cv);
    while (!g_stop_requested && !slot_job_cancelled(slot) &&
           (s->model_busy || s->decode_pending > 0 ||
            server_next_prefill_slot_locked(s) != slot->id)) {
        pthread_cond_wait(&s->model_cv, &s->model_mu);
    }
    if (g_stop_requested || slot_job_cancelled(slot)) {
        slot->prefill_waiting = false;
        pthread_cond_broadcast(&s->model_cv);
        pthread_mutex_unlock(&s->model_mu);
        return false;
    }
    slot->prefill_waiting = false;
    s->last_prefill_slot = slot->id;
    s->model_busy = true;
    pthread_mutex_unlock(&s->model_mu);
    pthread_mutex_lock(&s->inference_mu);
    return true;
}

static void server_prefill_leave(server *s) {
    if (!s) return;
    pthread_mutex_unlock(&s->inference_mu);
    if (!s->batched_mode) return;