billion-context 0.1.59-pr.303.152 → 0.1.59-pr.304.154

package/dist/index.js

@@ -47852,6 +47852,7 @@ function rejectLegacyRoute(key, value) {
 // src/server.ts
 import http from "http";
 import fs6 from "fs";
+import { randomUUID as randomUUID2 } from "crypto";
 
 // src/compress-settings.ts
 function resolveContextLimitValue(raw, nativeLimit) {
@@ -52129,15 +52130,8 @@ ${systemPrompt}` : systemPrompt;
         } else if (type === "message_delta") {
           const u2 = data.usage ?? {};
           if (typeof u2.output_tokens === "number") roundOutput = u2.output_tokens;
-          const input = typeof u2.input_tokens === "number" ? u2.input_tokens : void 0;
-          const cached = typeof u2.cache_read_input_tokens === "number" ? u2.cache_read_input_tokens : void 0;
-          if (input !== void 0 && input > 0 && cached !== void 0) {
-            roundInput = input;
-            roundCached = cached;
-          } else {
-            if (input !== void 0 && input > 0) roundInput = input;
-            if (cached !== void 0 && cached > 0) roundCached = cached;
-          }
+          if (typeof u2.input_tokens === "number" && u2.input_tokens > 0) roundInput = u2.input_tokens;
+          if (typeof u2.cache_read_input_tokens === "number" && u2.cache_read_input_tokens > 0) roundCached = u2.cache_read_input_tokens;
           const d = data.delta ?? {};
           if (typeof d.stop_reason === "string") stopReason = d.stop_reason;
           if (!usageYielded) {
@@ -54728,6 +54722,7 @@ async function decodeRequestBody(contentEncoding, input, maxOutputBytes) {
 function bodyDumpEnabled() {
   return process.env.ACP_DUMP_BODY === "1";
 }
+var BILI_HOP_HEADER = "x-bili-hop";
 function parseLauncherModelWindows(raw) {
   if (!raw) return {};
   try {
@@ -54823,6 +54818,7 @@ async function startServer(opts) {
   const core = createCore();
   const config = opts.kernelConfig;
   const log2 = (level, msg) => logMsg(opts, level, msg);
+  const instanceId = randomUUID2();
   await initSessions();
   loadConversations();
   log2("info", `[persist] ${getStore().enabled ? "enabled" : "disabled"}`);
@@ -54832,7 +54828,7 @@ async function startServer(opts) {
   void loadRegistry();
   const server = http.createServer(async (req, res) => {
     try {
-      await handle(req, res, opts, core, config, log2);
+      await handle(req, res, opts, core, config, log2, instanceId);
     } catch (err2) {
       const msg = String(err2);
       log2("error", msg);
@@ -54957,7 +54953,7 @@ function adminTrustedHosts(bindHost, port) {
   }
   return set;
 }
-async function handle(req, res, opts, core, config, log2) {
+async function handle(req, res, opts, core, config, log2, instanceId) {
   const isAdminPath = req.url === "/__bili/" || req.url?.startsWith("/__bili/") || req.url === "/__acp/" || req.url?.startsWith("/__acp/");
   if (isAdminPath && !isLoopbackAddress(req.socket.remoteAddress)) {
     res.writeHead(403, { "content-type": "application/json" });
@@ -55118,6 +55114,11 @@ async function handle(req, res, opts, core, config, log2) {
     res.end(JSON.stringify({ error: { type: "invalid_request", message: String(err2) } }));
     return;
   }
+  const hopMarker = headerValue2(req, BILI_HOP_HEADER);
+  if (hopMarker !== void 0) {
+    const selfLoop = hopMarker === instanceId;
+    log2("warn", selfLoop ? `[chain] inbound request carries THIS instance's ${BILI_HOP_HEADER} marker (${hopMarker}) \u2014 self-loop detected. Passing through without processing; check your upstream config (it may point back to this instance).` : `[chain] inbound request carries ${BILI_HOP_HEADER} from another bili instance (${hopMarker}) \u2014 bili\u2192bili chain detected. Passing through without processing to avoid double compression; keep only one bili instance in the chain.`);
+  }
   const countTokens = isCountTokensRequest(req.method ?? "GET", urlPath, bodyBuffer.length > 0);
   let parsed = null;
   if (protocol && bodyBuffer.length > 0) {
@@ -55181,7 +55182,7 @@ ${bodyBuffer.toString("utf8")}`);
     }
   }
   let prepared = null;
-  if (!opts.passthrough && protocol && parsed && typeof parsed === "object") {
+  if (!opts.passthrough && hopMarker === void 0 && protocol && parsed && typeof parsed === "object") {
     const sessionHeader = headerValue2(req, opts.sessionHeader);
     let pluginAgent = pluginAgentHeader(req.headers);
     let pluginConversation = pluginConversationHeader(req.headers);
@@ -55281,10 +55282,11 @@ ${bodyBuffer.toString("utf8")}`);
             parsed.model,
             route,
             affinity,
-            log2
+            log2,
+            instanceId
           );
         }
-        await forward(req, res, opts, prepared.body, prepared, core, reqConfig, log2, route, affinity);
+        await forward(req, res, opts, prepared.body, prepared, core, reqConfig, log2, route, instanceId, affinity);
         if (prepared) {
           rememberPluginMessages(sessionId, prepared.processedMessages, prepared.originalMessages, prepared.nudge);
         }
@@ -55297,7 +55299,7 @@ ${bodyBuffer.toString("utf8")}`);
     if (protocol === null && !opts.passthrough) {
       log2("warn", `unrecognized path ${maskUrlsInText(req.url ?? "")} \u2014 not a known protocol (/chat/completions, /v1/messages, /responses, /responses/compact); forwarding unchanged`);
     }
-    await forward(req, res, opts, bodyBuffer, null, core, reqConfig, log2, route, void 0);
+    await forward(req, res, opts, bodyBuffer, null, core, reqConfig, log2, route, instanceId, void 0);
   }
 }
 var ACP_TAG_MARK = "<acp ";
@@ -55658,7 +55660,7 @@ function injectResponsesTool(tools, toolsToAdd = ACP_TOOLS_RESPONSES) {
   const additions = toolsToAdd.filter((t) => !present.has(t.name));
   return [...tools, ...additions];
 }
-function buildForwardTarget(req, opts, route, affinity) {
+function buildForwardTarget(req, opts, route, affinity, hopMarker) {
   const reqUrl = req.url ?? "";
   const isAbsoluteUrl = /^https?:\/\//i.test(reqUrl);
   const rewritten = route ? route.rewrittenUrl : isAbsoluteUrl ? reqUrl : opts.upstream + reqUrl;
@@ -55670,6 +55672,7 @@ function buildForwardTarget(req, opts, route, affinity) {
     if (UPSTREAM_HOP_HEADERS.has(lower) || reqConnNamed.has(lower) || v2 === void 0) continue;
     headers[k2] = Array.isArray(v2) ? v2.join(", ") : v2;
   }
+  if (hopMarker !== void 0) headers[BILI_HOP_HEADER] = hopMarker;
   headers["host"] = new URL(upstreamUrl).host;
   const betaKey = Object.keys(headers).find((h) => h.toLowerCase() === "x-codex-beta-features");
   if (betaKey) {
@@ -55683,14 +55686,14 @@ function buildForwardTarget(req, opts, route, affinity) {
   const proxyUrl = resolveProxy(opts.routes, opts.proxy, route?.rewrittenUrl ?? upstreamUrl, opts.proxyFallback);
   return { upstreamUrl, headers, proxyUrl };
 }
-async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, core, config, model, route, affinity, log2) {
+async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, core, config, model, route, affinity, log2, instanceId) {
   const session = prepared.session;
   const limit = config.modelContextLimit;
   const tokenCount = Math.max(session.stats.lastInputTokens, estimateCoreMessages(prepared.processedMessages));
   if (limit <= 0 || !model || tokenCount < limit) return prepared;
   if ((prepared.nudge?.compressibleRanges ?? []).length === 0) return prepared;
   log2("warn", `[${session.id}] context ${tokenCount} tokens exceeds model window ${limit} (model=${model}); preflight compressing before forward`);
-  const { upstreamUrl, headers, proxyUrl } = buildForwardTarget(req, opts, route, affinity);
+  const { upstreamUrl, headers, proxyUrl } = buildForwardTarget(req, opts, route, affinity, instanceId);
   const clientAbort = new AbortController();
   res.on("close", () => {
     if (!res.writableEnded) clientAbort.abort();
@@ -55721,8 +55724,8 @@ async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, c
   log2("warn", `[${session.id}] preflight could not compress enough (~${result.savedTokens} tokens saved); forwarding as-is`);
   return prepared;
 }
-async function forward(req, res, opts, body, prepared, core, config, log2, route, affinity) {
-  const { upstreamUrl, headers, proxyUrl } = buildForwardTarget(req, opts, route, affinity);
+async function forward(req, res, opts, body, prepared, core, config, log2, route, instanceId, affinity) {
+  const { upstreamUrl, headers, proxyUrl } = buildForwardTarget(req, opts, route, affinity, prepared !== null ? instanceId : void 0);
   log2("info", `forward ${req.method} \u2192 ${maskUrlForLog(upstreamUrl)}`);
   if (process.env.ACP_DEBUG && prepared) {
     const sid = prepared.session.id;
@@ -60062,7 +60065,7 @@ function pluginStatusAll() {
 }
 
 // src/launcher.ts
-import { randomUUID as randomUUID2 } from "crypto";
+import { randomUUID as randomUUID3 } from "crypto";
 import fs10 from "fs";
 import net2 from "net";
 import os5 from "os";
@@ -61042,7 +61045,7 @@ async function runLaunch(params, deps = {}) {
     const dshAcpPatch = writeDshAcpPatch(dshHomeDir);
     if (dshAcpPatch) clientArgs = dshArgsWithPatch(clientArgs, dshAcpPatch);
   } else if (base === "codex") {
-    const codexConversationId = injectMcp ? randomUUID2() : void 0;
+    const codexConversationId = injectMcp ? randomUUID3() : void 0;
     if (directUrl) {
       env = { ...process.env, BILLION_CONTEXT_PROXY: origin };
       if (codexConversationId) clientArgs = [...buildCodexMcpArgs(origin, codexConversationId), ...clientArgs];