ai 7.0.55 → 7.0.56

package/CHANGELOG.md

@@ -1,5 +1,26 @@
 # ai
 
+## 7.0.56
+
+### Patch Changes
+
+- 25c9120: Expose provider metadata on language-model-call end callbacks and telemetry spans.
+- 89080c8: fix (ai/gateway): make retried `doStart` calls idempotent
+
+  `generateVideo` retries `doStart`, which creates a billable generation, so a
+  retry after a lost response could start a second one. It now mints one
+  idempotency token per logical start — outside the retry closure — and forwards it
+  as an `idempotency-key` header, so a provider that deduplicates (the Vercel AI
+  Gateway does) sees the same key on every attempt. `GatewayVideoModel` simply
+  forwards the caller's headers rather than inferring retry identity from an
+  options object, which would collide across unrelated calls.
+
+- 79d6195: Stop pending and active resumed chat streams after cancellation, and prevent
+  overlapping resumptions from applying stale updates.
+- Updated dependencies [89080c8]
+- Updated dependencies [89080c8]
+  - @ai-sdk/gateway@4.0.44
+
 ## 7.0.55
 
 ### Patch Changes

package/dist/index.d.ts

@@ -1544,6 +1544,8 @@ type LanguageModelCallEndEvent<TOOLS extends ToolSet = ToolSet> = ModelInfo & {
     readonly content: ReadonlyArray<ContentPart<TOOLS>>;
     /** The provider-returned response id for this model call. */
     readonly responseId: string;
+    /** Optional provider-specific metadata for this model call. */
+    readonly providerMetadata?: ProviderMetadata;
     /** Performance metrics for the model call. */
     readonly performance: {
         /** Time spent waiting for the language model response in milliseconds. */
@@ -5314,6 +5316,7 @@ interface ChatTransport<UI_MESSAGE extends UIMessage> {
      *
      * @param options - Configuration object containing:
      * @param options.chatId - Unique identifier for the chat session to reconnect to
+     * @param options.abortSignal - Signal to abort the reconnection request if needed
      * @param options.headers - Additional HTTP headers to include in the reconnection request
      * @param options.body - Additional JSON properties to include in the request body
      * @param options.metadata - Custom metadata to attach to the request
@@ -5327,6 +5330,8 @@ interface ChatTransport<UI_MESSAGE extends UIMessage> {
     reconnectToStream: (options: {
         /** Unique identifier for the chat session to reconnect to */
         chatId: string;
+        /** Signal to abort the reconnection request if needed */
+        abortSignal?: AbortSignal;
     } & ChatRequestOptions) => Promise<ReadableStream<UIMessageChunk> | null>;
 }
 
@@ -5487,6 +5492,7 @@ declare abstract class AbstractChat<UI_MESSAGE extends UIMessage> {
     private onData?;
     private sendAutomaticallyWhen?;
     private activeResponse;
+    private activeResumeRequest;
     private jobExecutor;
     constructor({ generateId, id, transport, messageMetadataSchema, dataPartSchemas, state, onError, onToolCall, onFinish, onData, sendAutomaticallyWhen, }: Omit<ChatInit<UI_MESSAGE>, 'messages'> & {
         state: ChatState<UI_MESSAGE>;
@@ -7431,9 +7437,10 @@ declare function generateObject<SCHEMA extends FlexibleSchema<unknown> = Flexibl
  * @param options.onError - Optional callback to handle errors that occur during consumption.
  * @returns A promise that resolves when the stream is fully consumed.
  */
-declare function consumeStream({ stream, onError, }: {
+declare function consumeStream({ stream, onError, abortSignal, }: {
     stream: ReadableStream;
     onError?: (error: unknown) => void;
+    abortSignal?: AbortSignal;
 }): Promise<void>;
 
 /**

package/dist/index.js

@@ -11,7 +11,7 @@ import {
   createIdGenerator as createIdGenerator9,
   dynamicTool,
   experimental_toolCaller,
-  generateId,
+  generateId as generateId2,
   jsonSchema,
   parseJsonEventStream as parseJsonEventStream3,
   tool,
@@ -1143,7 +1143,7 @@ import {
 } from "@ai-sdk/provider-utils";
 
 // src/version.ts
-var VERSION = true ? "7.0.55" : "0.0.0-test";
+var VERSION = true ? "7.0.56" : "0.0.0-test";
 
 // src/util/download/download.ts
 var download = async ({
@@ -5227,7 +5227,7 @@ async function generateText({
   experimental_include,
   include = experimental_include,
   _internal: {
-    generateId: generateId2 = originalGenerateId,
+    generateId: generateId3 = originalGenerateId,
     generateCallId = originalGenerateCallId,
     now: now2 = now
   } = {},
@@ -5612,7 +5612,7 @@ async function generateText({
                   }
                 );
                 const responseData = {
-                  id: (_b3 = (_a26 = result.response) == null ? void 0 : _a26.id) != null ? _b3 : generateId2(),
+                  id: (_b3 = (_a26 = result.response) == null ? void 0 : _a26.id) != null ? _b3 : generateId3(),
                   timestamp: (_d3 = (_c3 = result.response) == null ? void 0 : _c3.timestamp) != null ? _d3 : /* @__PURE__ */ new Date(),
                   modelId: (_f2 = (_e2 = result.response) == null ? void 0 : _e2.modelId) != null ? _f2 : stepModel.modelId,
                   headers: (_g2 = result.response) == null ? void 0 : _g2.headers,
@@ -5658,6 +5658,9 @@ async function generateText({
                   usage: stepUsage,
                   content: modelCallContent,
                   responseId: currentModelResponse.response.id,
+                  ...currentModelResponse.providerMetadata != null ? {
+                    providerMetadata: currentModelResponse.providerMetadata
+                  } : {},
                   performance: {
                     responseTimeMs,
                     effectiveOutputTokensPerSecond: calculateTokensPerSecond({
@@ -5717,7 +5720,7 @@ async function generateText({
                 if (toolApprovalStatus.type === "not-applicable") {
                   continue;
                 }
-                const approvalId = generateId2();
+                const approvalId = generateId3();
                 const signature = await maybeSignApproval({
                   secret: experimental_toolApprovalSecret,
                   approvalId,
@@ -7886,9 +7889,19 @@ function asAsyncIterableStream(stream) {
 // src/util/consume-stream.ts
 async function consumeStream({
   stream,
-  onError
+  onError,
+  abortSignal
 }) {
   const reader = stream.getReader();
+  const cancelOnAbort = () => {
+    reader.cancel().catch(() => {
+    });
+  };
+  if (abortSignal == null ? void 0 : abortSignal.aborted) {
+    cancelOnAbort();
+  } else {
+    abortSignal == null ? void 0 : abortSignal.addEventListener("abort", cancelOnAbort, { once: true });
+  }
   try {
     while (true) {
       const { done } = await reader.read();
@@ -7898,6 +7911,7 @@ async function consumeStream({
   } catch (error) {
     onError == null ? void 0 : onError(error);
   } finally {
+    abortSignal == null ? void 0 : abortSignal.removeEventListener("abort", cancelOnAbort);
     reader.releaseLock();
   }
 }
@@ -8021,7 +8035,7 @@ function executeToolsFromStream({
   toolApproval,
   runtimeContext,
   toolApprovalSecret,
-  generateId: generateId2,
+  generateId: generateId3,
   onToolExecutionStart,
   onToolExecutionEnd,
   executeToolInTelemetryContext,
@@ -8056,7 +8070,7 @@ function executeToolsFromStream({
               }
               return;
             }
-            const approvalId = generateId2();
+            const approvalId = generateId3();
             const signature = await maybeSignApproval({
               secret: toolApprovalSecret,
               approvalId,
@@ -8262,7 +8276,7 @@ async function streamLanguageModelCall({
   toolsContext,
   experimental_sandbox: sandbox,
   _internal: {
-    generateId: generateId2 = originalGenerateId2,
+    generateId: generateId3 = originalGenerateId2,
     generateCallId = originalGenerateCallId2,
     now: now2 = now
   } = {},
@@ -8347,7 +8361,7 @@ async function streamLanguageModelCall({
       callId: effectiveCallId,
       provider: resolvedModel.provider,
       modelId: resolvedModel.modelId,
-      generateId: generateId2,
+      generateId: generateId3,
       now: now2,
       callStartTimestampMs,
       onLanguageModelCallEnd
@@ -8368,7 +8382,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({
   callId,
   provider,
   modelId,
-  generateId: generateId2,
+  generateId: generateId3,
   now: now2,
   callStartTimestampMs,
   onLanguageModelCallEnd
@@ -8377,7 +8391,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({
   const modelCallContent = [];
   const textPartIndexes = /* @__PURE__ */ new Map();
   const reasoningPartIndexes = /* @__PURE__ */ new Map();
-  let responseId = generateId2();
+  let responseId = generateId3();
   let responseModelId = modelId;
   let timeToFirstOutputMs;
   let previousOutputChunkTimestampMs;
@@ -8522,6 +8536,7 @@ function createLanguageModelV4StreamPartToLanguageModelStreamPartTransform({
               usage,
               content: modelCallContent,
               responseId,
+              ...chunk.providerMetadata != null ? { providerMetadata: chunk.providerMetadata } : {},
               performance
             },
             callbacks: onLanguageModelCallEnd
@@ -8811,7 +8826,7 @@ function streamText({
   include = experimental_include,
   _internal: {
     now: now2 = now,
-    generateId: generateId2 = originalGenerateId3,
+    generateId: generateId3 = originalGenerateId3,
     generateCallId = originalGenerateCallId3
   } = {},
   ...settings
@@ -8885,7 +8900,7 @@ function streamText({
     onToolExecutionStart: resolvedOnToolExecutionStart,
     onToolExecutionEnd: resolvedOnToolExecutionEnd,
     now: now2,
-    generateId: generateId2,
+    generateId: generateId3,
     generateCallId,
     download: download2,
     // assign default values to include:
@@ -9003,7 +9018,7 @@ var DefaultStreamTextResult = class {
     providerOptions,
     prepareStep,
     now: now2,
-    generateId: generateId2,
+    generateId: generateId3,
     generateCallId,
     timeout,
     onChunk,
@@ -9771,7 +9786,7 @@ var DefaultStreamTextResult = class {
             toolApproval,
             runtimeContext,
             toolApprovalSecret: experimental_toolApprovalSecret,
-            generateId: generateId2,
+            generateId: generateId3,
             // the callbacks need to be passed down and handled by executeToolCall
             // to guarantee that the onToolExecutionStart callback is invoked before the tool execute function
             onToolExecutionStart: filterNullable2(
@@ -9813,7 +9828,7 @@ var DefaultStreamTextResult = class {
           };
           const toolExecutionMs = {};
           let stepResponse = {
-            id: generateId2(),
+            id: generateId3(),
             timestamp: /* @__PURE__ */ new Date(),
             modelId: model.modelId
           };
@@ -10577,7 +10592,7 @@ function createUIMessageStream({
   onStepFinish,
   onEnd,
   onFinish,
-  generateId: generateId2 = generateIdFunc
+  generateId: generateId3 = generateIdFunc
 }) {
   let controller;
   const ongoingStreamPromises = [];
@@ -10649,7 +10664,7 @@ function createUIMessageStream({
   });
   return handleUIMessageStreamFinish({
     stream,
-    messageId: generateId2(),
+    messageId: generateId3(),
     originalMessages,
     onStepEnd: onStepEnd != null ? onStepEnd : onStepFinish,
     onEnd: onEnd != null ? onEnd : onFinish,
@@ -12925,7 +12940,7 @@ async function generateObject(options) {
     onStepFinish,
     onFinish,
     _internal: {
-      generateId: generateId2 = originalGenerateId4,
+      generateId: generateId3 = originalGenerateId4,
       currentDate = () => /* @__PURE__ */ new Date()
     } = {},
     ...settings
@@ -12965,7 +12980,7 @@ async function generateObject(options) {
   const resolvedOnStepStart = onStepStart != null ? onStepStart : experimental_onStepStart;
   const resolvedOnStepEnd = onStepEnd != null ? onStepEnd : onStepFinish;
   const jsonSchema2 = await outputStrategy.jsonSchema();
-  const callId = generateId2();
+  const callId = generateId3();
   await notify({
     event: {
       callId,
@@ -13034,7 +13049,7 @@ async function generateObject(options) {
       })
     );
     const responseData = {
-      id: (_b = (_a23 = generateResult.response) == null ? void 0 : _a23.id) != null ? _b : generateId2(),
+      id: (_b = (_a23 = generateResult.response) == null ? void 0 : _a23.id) != null ? _b : generateId3(),
       timestamp: (_d = (_c = generateResult.response) == null ? void 0 : _c.timestamp) != null ? _d : currentDate(),
       modelId: (_f = (_e = generateResult.response) == null ? void 0 : _e.modelId) != null ? _f : model.modelId,
       headers: (_g = generateResult.response) == null ? void 0 : _g.headers,
@@ -13314,7 +13329,7 @@ function streamObject(options) {
     },
     onFinish,
     _internal: {
-      generateId: generateId2 = originalGenerateId5,
+      generateId: generateId3 = originalGenerateId5,
       currentDate = () => /* @__PURE__ */ new Date(),
       now: now2 = now
     } = {},
@@ -13361,7 +13376,7 @@ function streamObject(options) {
     onError,
     onFinish,
     download: download2,
-    generateId: generateId2,
+    generateId: generateId3,
     currentDate,
     now: now2
   });
@@ -13390,7 +13405,7 @@ var DefaultStreamObjectResult = class {
     onError,
     onFinish,
     download: download2,
-    generateId: generateId2,
+    generateId: generateId3,
     currentDate,
     now: now2
   }) {
@@ -13421,7 +13436,7 @@ var DefaultStreamObjectResult = class {
       }
     });
     this.baseStream = stitchableStream.stream.pipeThrough(eventProcessor);
-    const callId = generateId2();
+    const callId = generateId3();
     (async () => {
       const jsonSchema2 = await outputStrategy.jsonSchema();
       await notify({
@@ -13518,7 +13533,7 @@ var DefaultStreamObjectResult = class {
       let accumulatedText = "";
       let textDelta = "";
       let fullResponse = {
-        id: generateId2(),
+        id: generateId3(),
         timestamp: currentDate(),
         modelId: model.modelId
       };
@@ -14136,6 +14151,7 @@ function detectToolDrift(current, baseline) {
 import {
   convertBase64ToUint8Array as convertBase64ToUint8Array5,
   delay as defaultDelay,
+  generateId,
   withUserAgentSuffix as withUserAgentSuffix10,
   detectMediaType as detectMediaType4
 } from "@ai-sdk/provider-utils";
@@ -14337,7 +14353,7 @@ async function executeStartStatusFlow({
   webhook: webhookFactory,
   retry
 }) {
-  var _a23, _b, _c;
+  var _a23, _b, _c, _d;
   const earlyWarnings = [];
   let webhookUrl;
   let webhookReceived;
@@ -14356,17 +14372,23 @@ async function executeStartStatusFlow({
       });
     }
   }
-  const startResult = await retry(
-    () => model.doStart({
-      ...callOptions,
-      webhookUrl
-    })
+  const callerIdempotencyKey = Object.entries((_a23 = callOptions.headers) != null ? _a23 : {}).find(
+    ([key, value]) => key.toLowerCase() === "idempotency-key" && value !== void 0
   );
+  const startCallOptions = {
+    ...callOptions,
+    headers: {
+      ...callOptions.headers,
+      ...callerIdempotencyKey ? {} : { "idempotency-key": `aisdk_vid_${generateId()}` }
+    },
+    webhookUrl
+  };
+  const startResult = await retry(() => model.doStart(startCallOptions));
   const allWarnings = [...earlyWarnings, ...startResult.warnings];
   let operationProviderMetadata = startResult.providerMetadata == null ? void 0 : { ...startResult.providerMetadata };
-  const intervalMs = (_a23 = pollConfig == null ? void 0 : pollConfig.intervalMs) != null ? _a23 : 5e3;
-  const timeoutMs = (_b = pollConfig == null ? void 0 : pollConfig.timeoutMs) != null ? _b : 6e5;
-  const delay = (_c = pollConfig == null ? void 0 : pollConfig.delay) != null ? _c : defaultDelay;
+  const intervalMs = (_b = pollConfig == null ? void 0 : pollConfig.intervalMs) != null ? _b : 5e3;
+  const timeoutMs = (_c = pollConfig == null ? void 0 : pollConfig.timeoutMs) != null ? _c : 6e5;
+  const delay = (_d = pollConfig == null ? void 0 : pollConfig.delay) != null ? _d : defaultDelay;
   const startTime = Date.now();
   if (webhookReceived != null) {
     await waitForWebhook({
@@ -17473,7 +17495,8 @@ var HttpChatTransport = class {
     const response = await fetch2(api, {
       method: "GET",
       headers,
-      credentials
+      credentials,
+      signal: options.abortSignal
     });
     if (response.status === 204) {
       return null;
@@ -17515,8 +17538,8 @@ var DefaultChatTransport = class extends HttpChatTransport {
 // src/ui/chat.ts
 var AbstractChat = class {
   constructor({
-    generateId: generateId2 = generateIdFunc2,
-    id = generateId2(),
+    generateId: generateId3 = generateIdFunc2,
+    id = generateId3(),
     transport = new DefaultChatTransport(),
     messageMetadataSchema,
     dataPartSchemas,
@@ -17528,6 +17551,7 @@ var AbstractChat = class {
     sendAutomaticallyWhen
   }) {
     this.activeResponse = void 0;
+    this.activeResumeRequest = void 0;
     this.jobExecutor = new SerialJobExecutor();
     /**
      * Appends or replaces a user message to the chat list. This triggers the API call to fetch
@@ -17697,16 +17721,13 @@ var AbstractChat = class {
      * Abort the current request immediately, keep the generated tokens if any.
      */
     this.stop = async () => {
-      var _a23;
-      if (this.status !== "streaming" && this.status !== "submitted")
-        return;
-      if ((_a23 = this.activeResponse) == null ? void 0 : _a23.abortController) {
-        this.activeResponse.abortController.abort();
-      }
+      var _a23, _b;
+      (_a23 = this.activeResumeRequest) == null ? void 0 : _a23.abortController.abort();
+      (_b = this.activeResponse) == null ? void 0 : _b.abortController.abort();
     };
     this.id = id;
     this.transport = transport;
-    this.generateId = generateId2;
+    this.generateId = generateId3;
     this.messageMetadataSchema = messageMetadataSchema;
     this.dataPartSchemas = dataPartSchemas;
     this.state = state;
@@ -17766,25 +17787,60 @@ var AbstractChat = class {
     body,
     messageId
   }) {
-    var _a23, _b;
+    var _a23, _b, _c;
+    const abortController = new AbortController();
+    const activeResumeRequest = trigger === "resume-stream" ? { abortController } : void 0;
+    if (activeResumeRequest) {
+      (_a23 = this.activeResumeRequest) == null ? void 0 : _a23.abortController.abort();
+      this.activeResumeRequest = activeResumeRequest;
+    }
+    const isCurrentRequest = () => activeResumeRequest == null || this.activeResumeRequest === activeResumeRequest;
+    const clearActiveResumeRequest = () => {
+      if (this.activeResumeRequest === activeResumeRequest) {
+        this.activeResumeRequest = void 0;
+      }
+    };
     let resumeStream;
     if (trigger === "resume-stream") {
       try {
         const reconnect = await this.transport.reconnectToStream({
           chatId: this.id,
+          abortSignal: abortController.signal,
           metadata,
           headers,
           body
         });
+        if (abortController.signal.aborted || !isCurrentRequest()) {
+          await (reconnect == null ? void 0 : reconnect.cancel().catch(() => {
+          }));
+          if (isCurrentRequest()) {
+            this.setStatus({ status: "ready" });
+          }
+          clearActiveResumeRequest();
+          return;
+        }
         if (reconnect == null) {
+          this.setStatus({ status: "ready" });
+          clearActiveResumeRequest();
           return;
         }
         resumeStream = reconnect;
       } catch (err) {
+        if (abortController.signal.aborted || err.name === "AbortError") {
+          if (isCurrentRequest()) {
+            this.setStatus({ status: "ready" });
+          }
+          clearActiveResumeRequest();
+          return;
+        }
+        if (!isCurrentRequest()) {
+          return;
+        }
         if (this.onError && err instanceof Error) {
           this.onError(err);
         }
         this.setStatus({ status: "error", error: err });
+        clearActiveResumeRequest();
         return;
       }
     }
@@ -17800,7 +17856,7 @@ var AbstractChat = class {
           lastMessage: trigger === "regenerate-message" ? void 0 : this.state.snapshot(lastMessage),
           messageId: this.generateId()
         }),
-        abortController: new AbortController()
+        abortController
       };
       activeResponse = response;
       response.abortController.signal.addEventListener("abort", () => {
@@ -17824,11 +17880,17 @@ var AbstractChat = class {
       }
       const runUpdateMessageJob = (job) => (
         // serialize the job execution to avoid race conditions:
-        this.jobExecutor.run(
-          () => job({
+        this.jobExecutor.run(() => {
+          if (response.abortController.signal.aborted) {
+            return Promise.resolve();
+          }
+          return job({
             state: response.state,
             write: () => {
               var _a24;
+              if (response.abortController.signal.aborted) {
+                return;
+              }
               this.setStatus({ status: "streaming" });
               const replaceLastMessage = response.state.message.id === ((_a24 = this.lastMessage) == null ? void 0 : _a24.id);
               if (replaceLastMessage) {
@@ -17840,8 +17902,8 @@ var AbstractChat = class {
                 this.state.pushMessage(response.state.message);
               }
             }
-          })
-        )
+          });
+        })
       );
       await consumeStream({
         stream: processUIMessageStream({
@@ -17855,15 +17917,29 @@ var AbstractChat = class {
             throw error;
           }
         }),
+        abortSignal: response.abortController.signal,
         onError: (error) => {
           throw error;
         }
       });
-      this.setStatus({ status: "ready" });
+      if (isAbort) {
+        if (isCurrentRequest()) {
+          this.setStatus({ status: "ready" });
+        }
+        return null;
+      }
+      if (isCurrentRequest()) {
+        this.setStatus({ status: "ready" });
+      }
     } catch (err) {
       if (isAbort || err.name === "AbortError") {
         isAbort = true;
-        this.setStatus({ status: "ready" });
+        if (isCurrentRequest()) {
+          this.setStatus({ status: "ready" });
+        }
+        return null;
+      }
+      if (!isCurrentRequest()) {
         return null;
       }
       isError = true;
@@ -17877,7 +17953,7 @@ var AbstractChat = class {
     } finally {
       try {
         if (activeResponse) {
-          (_a23 = this.onFinish) == null ? void 0 : _a23.call(this, {
+          (_b = this.onFinish) == null ? void 0 : _b.call(this, {
             message: activeResponse.state.message,
             messages: this.state.messages,
             isAbort,
@@ -17892,11 +17968,12 @@ var AbstractChat = class {
       if (this.activeResponse === activeResponse) {
         this.activeResponse = void 0;
       }
+      clearActiveResumeRequest();
     }
     if (!isError && await this.shouldSendAutomatically()) {
       await this.makeRequest({
         trigger: "submit-message",
-        messageId: (_b = this.lastMessage) == null ? void 0 : _b.id,
+        messageId: (_c = this.lastMessage) == null ? void 0 : _c.id,
         metadata,
         headers,
         body
@@ -18213,7 +18290,7 @@ export {
   extractReasoningMiddleware,
   fingerprintTools,
   gateway2 as gateway,
-  generateId,
+  generateId2 as generateId,
   generateImage,
   generateObject,
   generateSpeech,