fix(core): enable mid-stream retries for all models and re-enable compression test (#24302)

integration-tests/api-resilience.responses

@@ -1 +1 @@
-{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"text":"Part 1. "}],"role":"model"},"index":0}]},{"usageMetadata":{"promptTokenCount":100,"candidatesTokenCount":10,"totalTokenCount":110}},{"candidates":[{"content":{"parts":[{"text":"Part 2."}],"role":"model"},"index":0}],"finishReason":"STOP"}]}
+{"method":"generateContentStream","response":[{"candidates":[{"content":{"parts":[{"text":"Part 1. "}],"role":"model"},"index":0}]},{"usageMetadata":{"promptTokenCount":100,"candidatesTokenCount":10,"totalTokenCount":110}},{"candidates":[{"content":{"parts":[{"text":"Part 2."}],"role":"model"},"index":0,"finishReason":"STOP"}]}]}

integration-tests/context-compress-interactive.test.ts

@@ -19,7 +19,7 @@ describe('Interactive Mode', () => {
     await rig.cleanup();
   });
 
-  it.skip('should trigger chat compression with /compress command', async () => {
+  it('should trigger chat compression with /compress command', async () => {
     await rig.setup('interactive-compress-success', {
       fakeResponsesPath: join(
         import.meta.dirname,

packages/core/src/core/client.test.ts

@@ -2157,48 +2157,6 @@ ${JSON.stringify(
       expect(mockTurnRunFn).toHaveBeenCalledTimes(1);
     });
 
-    it('should not retry with "Please continue." when InvalidStream event is received for non-Gemini-2 models', async () => {
-      vi.spyOn(client['config'], 'getContinueOnFailedApiCall').mockReturnValue(
-        true,
-      );
-      // Arrange - router returns a non-Gemini-2 model
-      mockRouterService.route.mockResolvedValue({
-        model: 'gemini-3.0-pro',
-        reason: 'test',
-      });
-
-      const mockStream1 = (async function* () {
-        yield { type: GeminiEventType.InvalidStream };
-      })();
-
-      mockTurnRunFn.mockReturnValueOnce(mockStream1);
-
-      const mockChat: Partial<GeminiChat> = {
-        addHistory: vi.fn(),
-        setTools: vi.fn(),
-        getHistory: vi.fn().mockReturnValue([]),
-        getLastPromptTokenCount: vi.fn(),
-      };
-      client['chat'] = mockChat as GeminiChat;
-
-      const initialRequest = [{ text: 'Hi' }];
-      const promptId = 'prompt-id-invalid-stream-non-g2';
-      const signal = new AbortController().signal;
-
-      // Act
-      const stream = client.sendMessageStream(initialRequest, signal, promptId);
-      const events = await fromAsync(stream);
-
-      // Assert
-      expect(events).toEqual([
-        { type: GeminiEventType.ModelInfo, value: 'gemini-3.0-pro' },
-        { type: GeminiEventType.InvalidStream },
-      ]);
-
-      // Verify that turn.run was called only once (no retry)
-      expect(mockTurnRunFn).toHaveBeenCalledTimes(1);
-    });
-
     it('should stop recursing after one retry when InvalidStream events are repeatedly received', async () => {
       vi.spyOn(client['config'], 'getContinueOnFailedApiCall').mockReturnValue(
         true,

packages/core/src/core/client.ts

@@ -71,11 +71,7 @@ import {
   applyModelSelection,
   createAvailabilityContextProvider,
 } from '../availability/policyHelpers.js';
-import {
-  getDisplayString,
-  resolveModel,
-  isGemini2Model,
-} from '../config/models.js';
+import { getDisplayString, resolveModel } from '../config/models.js';
 import { partToString } from '../utils/partUtils.js';
 import { coreEvents, CoreEvent } from '../utils/events.js';
 
@@ -820,10 +816,7 @@ export class GeminiClient {
     }
 
     if (isInvalidStream) {
-      if (
-        this.config.getContinueOnFailedApiCall() &&
-        isGemini2Model(modelToUse)
-      ) {
+      if (this.config.getContinueOnFailedApiCall()) {
         if (isInvalidStreamRetry) {
           logContentRetryFailure(
             this.config,

packages/core/src/core/geminiChat.test.ts

@@ -1140,41 +1140,6 @@ describe('GeminiChat', () => {
   });
 
   describe('sendMessageStream with retries', () => {
-    it('should not retry on invalid content if model does not start with gemini-2', async () => {
-      // Mock the stream to fail.
-      vi.mocked(mockContentGenerator.generateContentStream).mockImplementation(
-        async () =>
-          (async function* () {
-            yield {
-              candidates: [{ content: { parts: [{ text: '' }] } }],
-            } as unknown as GenerateContentResponse;
-          })(),
-      );
-
-      const stream = await chat.sendMessageStream(
-        { model: 'gemini-1.5-pro' },
-        'test',
-        'prompt-id-no-retry',
-        new AbortController().signal,
-        LlmRole.MAIN,
-      );
-
-      await expect(
-        (async () => {
-          for await (const _ of stream) {
-            // Must loop to trigger the internal logic that throws.
-          }
-        })(),
-      ).rejects.toThrow(InvalidStreamError);
-
-      // Should be called only 1 time (no retry)
-      expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(
-        1,
-      );
-      expect(mockLogContentRetry).not.toHaveBeenCalled();
-      expect(mockLogContentRetryFailure).toHaveBeenCalledTimes(1);
-    });
-
     it('should yield a RETRY event when an invalid stream is encountered', async () => {
       // ARRANGE: Mock the stream to fail once, then succeed.
       vi.mocked(mockContentGenerator.generateContentStream)

packages/core/src/core/geminiChat.ts

@@ -25,11 +25,7 @@ import {
   getRetryErrorType,
 } from '../utils/retry.js';
 import type { ValidationRequiredError } from '../utils/googleQuotaErrors.js';
-import {
-  resolveModel,
-  isGemini2Model,
-  supportsModernFeatures,
-} from '../config/models.js';
+import { resolveModel, supportsModernFeatures } from '../config/models.js';
 import { hasCycleInSchema } from '../tools/tools.js';
 import type { StructuredError } from './turn.js';
 import type { CompletedToolCall } from '../scheduler/types.js';
@@ -423,10 +419,7 @@ export class GeminiChat {
               ? error.type
               : getRetryErrorType(error);
 
-            if (
-              (isContentError && isGemini2Model(model)) ||
-              (isRetryable && !signal.aborted)
-            ) {
+            if (isContentError || (isRetryable && !signal.aborted)) {
               // The issue requests exactly 3 retries (4 attempts) for API errors during stream iteration.
               // Regardless of the global maxAttempts (e.g. 10), we only want to retry these mid-stream API errors
               // up to 3 times before finally throwing the error to the user.