Revert "fix: handle request retries and model fallback correctly" (#11164)

Author: gsquared94

packages/cli/src/ui/hooks/useQuotaAndFallback.test.ts

@@ -19,14 +19,25 @@ import {
   type FallbackModelHandler,
   UserTierId,
   AuthType,
-  TerminalQuotaError,
+  isGenericQuotaExceededError,
+  isProQuotaExceededError,
   makeFakeConfig,
-  type GoogleApiError,
 } from '@google/gemini-cli-core';
 import { useQuotaAndFallback } from './useQuotaAndFallback.js';
 import type { UseHistoryManagerReturn } from './useHistoryManager.js';
 import { AuthState, MessageType } from '../types.js';
 
+// Mock the error checking functions from the core package to control test scenarios
+vi.mock('@google/gemini-cli-core', async (importOriginal) => {
+  const original =
+    await importOriginal<typeof import('@google/gemini-cli-core')>();
+  return {
+    ...original,
+    isGenericQuotaExceededError: vi.fn(),
+    isProQuotaExceededError: vi.fn(),
+  };
+});
+
 // Use a type alias for SpyInstance as it's not directly exported
 type SpyInstance = ReturnType<typeof vi.spyOn>;
 
@@ -36,15 +47,12 @@ describe('useQuotaAndFallback', () => {
   let mockSetAuthState: Mock;
   let mockSetModelSwitchedFromQuotaError: Mock;
   let setFallbackHandlerSpy: SpyInstance;
-  let mockGoogleApiError: GoogleApiError;
+
+  const mockedIsGenericQuotaExceededError = isGenericQuotaExceededError as Mock;
+  const mockedIsProQuotaExceededError = isProQuotaExceededError as Mock;
 
   beforeEach(() => {
     mockConfig = makeFakeConfig();
-    mockGoogleApiError = {
-      code: 429,
-      message: 'mock error',
-      details: [],
-    };
 
     // Spy on the method that requires the private field and mock its return.
     // This is cleaner than modifying the config class for tests.
@@ -64,6 +72,9 @@ describe('useQuotaAndFallback', () => {
 
     setFallbackHandlerSpy = vi.spyOn(mockConfig, 'setFallbackModelHandler');
     vi.spyOn(mockConfig, 'setQuotaErrorOccurred');
+
+    mockedIsGenericQuotaExceededError.mockReturnValue(false);
+    mockedIsProQuotaExceededError.mockReturnValue(false);
   });
 
   afterEach(() => {
@@ -128,6 +139,22 @@ describe('useQuotaAndFallback', () => {
 
     describe('Automatic Fallback Scenarios', () => {
       const testCases = [
+        {
+          errorType: 'generic',
+          tier: UserTierId.FREE,
+          expectedMessageSnippets: [
+            'Automatically switching from model-A to model-B',
+            'upgrade to a Gemini Code Assist Standard or Enterprise plan',
+          ],
+        },
+        {
+          errorType: 'generic',
+          tier: UserTierId.STANDARD, // Paid tier
+          expectedMessageSnippets: [
+            'Automatically switching from model-A to model-B',
+            'switch to using a paid API key from AI Studio',
+          ],
+        },
         {
           errorType: 'other',
           tier: UserTierId.FREE,
@@ -148,11 +175,15 @@ describe('useQuotaAndFallback', () => {
 
       for (const { errorType, tier, expectedMessageSnippets } of testCases) {
         it(`should handle ${errorType} error for ${tier} tier correctly`, async () => {
+          mockedIsGenericQuotaExceededError.mockReturnValue(
+            errorType === 'generic',
+          );
+
           const handler = getRegisteredHandler(tier);
           const result = await handler(
             'model-A',
             'model-B',
-            new Error('some error'),
+            new Error('quota exceeded'),
           );
 
           // Automatic fallbacks should return 'stop'
@@ -176,6 +207,10 @@ describe('useQuotaAndFallback', () => {
     });
 
     describe('Interactive Fallback (Pro Quota Error)', () => {
+      beforeEach(() => {
+        mockedIsProQuotaExceededError.mockReturnValue(true);
+      });
+
       it('should set an interactive request and wait for user choice', async () => {
         const { result } = renderHook(() =>
           useQuotaAndFallback({
@@ -194,7 +229,7 @@ describe('useQuotaAndFallback', () => {
         const promise = handler(
           'gemini-pro',
           'gemini-flash',
-          new TerminalQuotaError('pro quota', mockGoogleApiError),
+          new Error('pro quota'),
         );
 
         await act(async () => {});
@@ -233,7 +268,7 @@ describe('useQuotaAndFallback', () => {
         const promise1 = handler(
           'gemini-pro',
           'gemini-flash',
-          new TerminalQuotaError('pro quota 1', mockGoogleApiError),
+          new Error('pro quota 1'),
         );
         await act(async () => {});
 
@@ -243,7 +278,7 @@ describe('useQuotaAndFallback', () => {
         const result2 = await handler(
           'gemini-pro',
           'gemini-flash',
-          new TerminalQuotaError('pro quota 2', mockGoogleApiError),
+          new Error('pro quota 2'),
         );
 
         // The lock should have stopped the second request
@@ -262,6 +297,10 @@ describe('useQuotaAndFallback', () => {
   });
 
   describe('handleProQuotaChoice', () => {
+    beforeEach(() => {
+      mockedIsProQuotaExceededError.mockReturnValue(true);
+    });
+
     it('should do nothing if there is no pending pro quota request', () => {
       const { result } = renderHook(() =>
         useQuotaAndFallback({
@@ -297,7 +336,7 @@ describe('useQuotaAndFallback', () => {
       const promise = handler(
         'gemini-pro',
         'gemini-flash',
-        new TerminalQuotaError('pro quota', mockGoogleApiError),
+        new Error('pro quota'),
       );
       await act(async () => {}); // Allow state to update
 
@@ -328,7 +367,7 @@ describe('useQuotaAndFallback', () => {
       const promise = handler(
         'gemini-pro',
         'gemini-flash',
-        new TerminalQuotaError('pro quota', mockGoogleApiError),
+        new Error('pro quota'),
       );
       await act(async () => {}); // Allow state to update
 

packages/cli/src/ui/hooks/useQuotaAndFallback.ts

@@ -9,7 +9,8 @@ import {
   type Config,
   type FallbackModelHandler,
   type FallbackIntent,
-  TerminalQuotaError,
+  isGenericQuotaExceededError,
+  isProQuotaExceededError,
   UserTierId,
 } from '@google/gemini-cli-core';
 import { useCallback, useEffect, useRef, useState } from 'react';
@@ -62,7 +63,7 @@ export function useQuotaAndFallback({
 
       let message: string;
 
-      if (error instanceof TerminalQuotaError) {
+      if (error && isProQuotaExceededError(error)) {
         // Pro Quota specific messages (Interactive)
         if (isPaidTier) {
           message = `⚡ You have reached your daily ${failedModel} quota limit.
@@ -73,6 +74,19 @@ export function useQuotaAndFallback({
 ⚡ You can choose to authenticate with a paid API key or continue with the fallback model.
 ⚡ To increase your limits, upgrade to a Gemini Code Assist Standard or Enterprise plan with higher limits at https://goo.gle/set-up-gemini-code-assist
 ⚡ Or you can utilize a Gemini API Key. See: https://goo.gle/gemini-cli-docs-auth#gemini-api-key
+⚡ You can switch authentication methods by typing /auth`;
+        }
+      } else if (error && isGenericQuotaExceededError(error)) {
+        // Generic Quota (Automatic fallback)
+        const actionMessage = `⚡ You have reached your daily quota limit.\n⚡ Automatically switching from ${failedModel} to ${fallbackModel} for the remainder of this session.`;
+
+        if (isPaidTier) {
+          message = `${actionMessage}
+⚡ To continue accessing the ${failedModel} model today, consider using /auth to switch to using a paid API key from AI Studio at https://aistudio.google.com/apikey`;
+        } else {
+          message = `${actionMessage}
+⚡ To increase your limits, upgrade to a Gemini Code Assist Standard or Enterprise plan with higher limits at https://goo.gle/set-up-gemini-code-assist
+⚡ Or you can utilize a Gemini API Key. See: https://goo.gle/gemini-cli-docs-auth#gemini-api-key
 ⚡ You can switch authentication methods by typing /auth`;
         }
       } else {
@@ -105,7 +119,7 @@ export function useQuotaAndFallback({
       config.setQuotaErrorOccurred(true);
 
       // Interactive Fallback for Pro quota
-      if (error instanceof TerminalQuotaError) {
+      if (error && isProQuotaExceededError(error)) {
         if (isDialogPending.current) {
           return 'stop'; // A dialog is already active, so just stop this request.
         }

packages/core/index.ts

@@ -44,5 +44,3 @@ export { makeFakeConfig } from './src/test-utils/config.js';
 export * from './src/utils/pathReader.js';
 export { ClearcutLogger } from './src/telemetry/clearcut-logger/clearcut-logger.js';
 export { logModelSlashCommand } from './src/telemetry/loggers.js';
-export * from './src/utils/googleQuotaErrors.js';
-export type { GoogleApiError } from './src/utils/googleErrors.js';