File size: 12,440 Bytes
bbb1195 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 | // Common utilities for request mapping across all protocols
// Provides unified grounding/networking logic
use serde_json::{json, Value};
/// Request configuration after grounding resolution
#[derive(Debug, Clone)]
pub struct RequestConfig {
/// The request type: "agent", "web_search", or "image_gen"
pub request_type: String,
/// Whether to inject the googleSearch tool
pub inject_google_search: bool,
/// The final model name (with suffixes stripped)
pub final_model: String,
/// Image generation configuration (if request_type is image_gen)
pub image_config: Option<Value>,
}
pub fn resolve_request_config(
original_model: &str,
mapped_model: &str,
tools: &Option<Vec<Value>>
) -> RequestConfig {
// 1. Image Generation Check (Priority)
if mapped_model.starts_with("gemini-3-pro-image") {
let (image_config, parsed_base_model) = parse_image_config(original_model);
return RequestConfig {
request_type: "image_gen".to_string(),
inject_google_search: false,
final_model: parsed_base_model,
image_config: Some(image_config),
};
}
// 检测是否有联网工具定义 (内置功能调用)
let has_networking_tool = detects_networking_tool(tools);
// 检测是否包含非联网工具 (如 MCP 本地工具)
let has_non_networking = contains_non_networking_tool(tools);
// Strip -online suffix from original model if present (to detect networking intent)
let is_online_suffix = original_model.ends_with("-online");
// High-quality grounding allowlist (Only for models known to support search and be relatively 'safe')
let is_high_quality_model = mapped_model == "gemini-2.5-flash"
|| mapped_model == "gemini-1.5-pro"
|| mapped_model.starts_with("gemini-1.5-pro-")
|| mapped_model.starts_with("gemini-2.5-flash-")
|| mapped_model.starts_with("gemini-2.0-flash")
|| mapped_model.starts_with("gemini-3-")
|| mapped_model.contains("claude-3-5-sonnet")
|| mapped_model.contains("claude-3-opus")
|| mapped_model.contains("claude-sonnet")
|| mapped_model.contains("claude-opus")
|| mapped_model.contains("claude-4");
// Determine if we should enable networking
// [FIX] 禁用基于模型的自动联网逻辑,防止图像请求被联网搜索结果覆盖。
// 仅在用户显式请求联网时启用:1) -online 后缀 2) 携带联网工具定义
let enable_networking = is_online_suffix || has_networking_tool;
// The final model to send upstream should be the MAPPED model,
// but if searching, we MUST ensure the model name is one the backend associates with search.
// Based on ref_Antigravity2Api practice, we force a stable search model for search requests.
let mut final_model = mapped_model.trim_end_matches("-online").to_string();
if enable_networking {
// If it's a thinking model (which doesn't support tools) or a Claude-style alias,
// fallback to gemini-2.5-flash which is the standard workhorse for search.
if final_model.contains("thinking") || !final_model.starts_with("gemini-") {
final_model = "gemini-2.5-flash".to_string();
}
}
RequestConfig {
request_type: if enable_networking {
"web_search".to_string()
} else {
"agent".to_string()
},
inject_google_search: enable_networking,
final_model,
image_config: None,
}
}
/// Parse image configuration from model name suffixes
/// Returns (image_config, clean_model_name)
fn parse_image_config(model_name: &str) -> (Value, String) {
let mut aspect_ratio = "1:1";
let _image_size = "1024x1024"; // Default, not explicitly sent unless 4k/hd
if model_name.contains("-16x9") { aspect_ratio = "16:9"; }
else if model_name.contains("-9x16") { aspect_ratio = "9:16"; }
else if model_name.contains("-4x3") { aspect_ratio = "4:3"; }
else if model_name.contains("-3x4") { aspect_ratio = "3:4"; }
else if model_name.contains("-1x1") { aspect_ratio = "1:1"; }
let is_hd = model_name.contains("-4k") || model_name.contains("-hd");
let mut config = serde_json::Map::new();
config.insert("aspectRatio".to_string(), json!(aspect_ratio));
if is_hd {
config.insert("imageSize".to_string(), json!("4K"));
}
// The upstream model must be EXACTLY "gemini-3-pro-image"
(serde_json::Value::Object(config), "gemini-3-pro-image".to_string())
}
/// Inject current googleSearch tool and ensure no duplicate legacy search tools
pub fn inject_google_search_tool(body: &mut Value) {
if let Some(obj) = body.as_object_mut() {
let tools_entry = obj.entry("tools").or_insert_with(|| json!([]));
if let Some(tools_arr) = tools_entry.as_array_mut() {
// [安全校验] 如果数组中已经包含 functionDeclarations,严禁注入 googleSearch
// 因为 Gemini v1internal 不支持在一次请求中混用 search 和 functions
let has_functions = tools_arr.iter().any(|t| {
t.as_object().map_or(false, |o| o.contains_key("functionDeclarations"))
});
if has_functions {
tracing::info!("Skipping googleSearch injection due to existing functionDeclarations");
return;
}
// 首先清理掉已存在的 googleSearch 或 googleSearchRetrieval,以防重复产生冲突
tools_arr.retain(|t| {
if let Some(o) = t.as_object() {
!(o.contains_key("googleSearch") || o.contains_key("googleSearchRetrieval"))
} else {
true
}
});
// 注入统一的 googleSearch (v1internal 规范)
tools_arr.push(json!({
"googleSearch": {}
}));
}
}
}
/// 深度迭代清理客户端发送的 [undefined] 脏字符串,防止 Gemini 接口校验失败
pub fn deep_clean_undefined(value: &mut Value) {
match value {
Value::Object(map) => {
// 移除值为 "[undefined]" 的键
map.retain(|_, v| {
if let Some(s) = v.as_str() {
s != "[undefined]"
} else {
true
}
});
// 递归处理嵌套
for v in map.values_mut() {
deep_clean_undefined(v);
}
}
Value::Array(arr) => {
for v in arr.iter_mut() {
deep_clean_undefined(v);
}
}
_ => {}
}
}
/// Detects if the tool list contains a request for networking/web search.
/// Supported keywords: "web_search", "google_search", "web_search_20250305"
pub fn detects_networking_tool(tools: &Option<Vec<Value>>) -> bool {
if let Some(list) = tools {
for tool in list {
// 1. 直发风格 (Claude/Simple OpenAI/Anthropic Builtin/Vertex): { "name": "..." } 或 { "type": "..." }
if let Some(n) = tool.get("name").and_then(|v| v.as_str()) {
if n == "web_search" || n == "google_search" || n == "web_search_20250305" || n == "google_search_retrieval" {
return true;
}
}
if let Some(t) = tool.get("type").and_then(|v| v.as_str()) {
if t == "web_search_20250305" || t == "google_search" || t == "web_search" || t == "google_search_retrieval" {
return true;
}
}
// 2. OpenAI 嵌套风格: { "type": "function", "function": { "name": "..." } }
if let Some(func) = tool.get("function") {
if let Some(n) = func.get("name").and_then(|v| v.as_str()) {
let keywords = ["web_search", "google_search", "web_search_20250305", "google_search_retrieval"];
if keywords.contains(&n) {
return true;
}
}
}
// 3. Gemini 原生风格: { "functionDeclarations": [ { "name": "..." } ] }
if let Some(decls) = tool.get("functionDeclarations").and_then(|v| v.as_array()) {
for decl in decls {
if let Some(n) = decl.get("name").and_then(|v| v.as_str()) {
if n == "web_search" || n == "google_search" || n == "google_search_retrieval" {
return true;
}
}
}
}
// 4. Gemini googleSearch 声明 (含 googleSearchRetrieval 变体)
if tool.get("googleSearch").is_some() || tool.get("googleSearchRetrieval").is_some() {
return true;
}
}
}
false
}
/// 探测是否包含非联网相关的本地函数工具
pub fn contains_non_networking_tool(tools: &Option<Vec<Value>>) -> bool {
if let Some(list) = tools {
for tool in list {
let mut is_networking = false;
// 简单逻辑:如果它是一个函数声明且名字不是联网关键词,则视为非联网工具
if let Some(n) = tool.get("name").and_then(|v| v.as_str()) {
let keywords = ["web_search", "google_search", "web_search_20250305", "google_search_retrieval"];
if keywords.contains(&n) { is_networking = true; }
} else if let Some(func) = tool.get("function") {
if let Some(n) = func.get("name").and_then(|v| v.as_str()) {
let keywords = ["web_search", "google_search", "web_search_20250305", "google_search_retrieval"];
if keywords.contains(&n) { is_networking = true; }
}
} else if tool.get("googleSearch").is_some() || tool.get("googleSearchRetrieval").is_some() {
is_networking = true;
} else if tool.get("functionDeclarations").is_some() {
// 如果是 Gemini 风格的 functionDeclarations,进去看一眼
if let Some(decls) = tool.get("functionDeclarations").and_then(|v| v.as_array()) {
for decl in decls {
if let Some(n) = decl.get("name").and_then(|v| v.as_str()) {
let keywords = ["web_search", "google_search", "google_search_retrieval"];
if !keywords.contains(&n) {
return true; // 发现本地函数
}
}
}
}
is_networking = true; // 即使全是联网,外层也标记为联网
}
if !is_networking {
return true;
}
}
}
false
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_high_quality_model_auto_grounding() {
let config = resolve_request_config("gpt-4o", "gemini-2.5-flash", &None);
assert_eq!(config.request_type, "web_search");
assert!(config.inject_google_search);
assert_eq!(config.final_model, "gemini-2.5-flash"); // 修正断言: final_model = mapped_model
}
#[test]
fn test_gemini_native_tool_detection() {
let tools = Some(vec![json!({
"functionDeclarations": [
{ "name": "web_search", "parameters": {} }
]
})]);
assert!(detects_networking_tool(&tools));
}
#[test]
fn test_online_suffix_force_grounding() {
let config = resolve_request_config("gemini-3-flash-online", "gemini-3-flash", &None);
assert_eq!(config.request_type, "web_search");
assert!(config.inject_google_search);
assert_eq!(config.final_model, "gemini-3-flash");
}
#[test]
fn test_default_no_grounding() {
let config = resolve_request_config("claude-sonnet", "gemini-3-flash", &None);
assert_eq!(config.request_type, "agent");
assert!(!config.inject_google_search);
}
#[test]
fn test_image_model_excluded() {
let config = resolve_request_config("gemini-3-pro-image", "gemini-3-pro-image", &None);
assert_eq!(config.request_type, "image_gen");
assert!(!config.inject_google_search);
}
}
|