MediaMetz/gemini-cli
1
0
Fork 0
mirror of https://github.com/google-gemini/gemini-cli.git synced 2026-07-15 20:40:35 -07:00
Code Issues Packages Projects Releases Wiki Activity

fix disables

Browse Source
This commit is contained in:
Your Name
2026-04-07 04:46:04 +00:00
parent 1754797929
commit 370e2b9e1d
46 changed files with 1602 additions and 884 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/core/src/context/system-tests/SimulationHarness.ts
+84 -59
View File
@@ -1,9 +1,3 @@
/**
* @license
* Copyright 2026 Google LLC
* SPDX-License-Identifier: Apache-2.0
*/
import type { BaseLlmClient } from "../../core/baseLlmClient.js";
/**
* @license
* Copyright 2026 Google LLC
@@ -19,8 +13,11 @@ import { ContextTracer } from '../tracer.js';
import { ContextEventBus } from '../eventBus.js';
import { PipelineOrchestrator } from '../sidecar/orchestrator.js';
import { registerBuiltInProcessors } from '../sidecar/builtins.js';
import { debugLogger } from "../../utils/debugLogger.js";
import { ProcessorRegistry } from "../sidecar/registry.js";
import { debugLogger } from '../../utils/debugLogger.js';
import { ProcessorRegistry } from '../sidecar/registry.js';
import { DeterministicIdGenerator } from '../system/DeterministicIdGenerator.js';
import { InMemoryFileSystem } from '../system/InMemoryFileSystem.js';
import type { BaseLlmClient } from '../../core/baseLlmClient.js';
export interface TurnSummary {
turnIndex: number;
@@ -39,7 +36,11 @@ export class SimulationHarness {
private currentTurnIndex = 0;
private tokenTrajectory: TurnSummary[] = [];
static async create(config: SidecarConfig, mockLlmClient: BaseLlmClient, mockTempDir = '/tmp/sim'): Promise<SimulationHarness> {
static async create(
config: SidecarConfig,
mockLlmClient: BaseLlmClient,
mockTempDir = '/tmp/sim',
): Promise<SimulationHarness> {
const harness = new SimulationHarness();
await harness.init(config, mockLlmClient, mockTempDir);
return harness;
@@ -53,19 +54,17 @@ export class SimulationHarness {
private async init(
config: SidecarConfig,
mockLlmClient: BaseLlmClient,
mockTempDir: string
mockTempDir: string,
) {
this.config = config;
const registry = new ProcessorRegistry();
// Register all standard processors
registerBuiltInProcessors(registry);
this.tracer = new ContextTracer({ targetDir: mockTempDir, sessionId: 'sim-session' });
// Using real token calculator instead of mock, so we test actual string sizes
const InMemoryFS = (await import('../system/InMemoryFileSystem.js')).InMemoryFileSystem;
const DetIdGen = (await import('../system/DeterministicIdGenerator.js')).DeterministicIdGenerator;
this.tracer = new ContextTracer({
targetDir: mockTempDir,
sessionId: 'sim-session',
});
this.env = new ContextEnvironmentImpl(
mockLlmClient,
'sim-prompt',
@@ -75,12 +74,24 @@ export class SimulationHarness {
this.tracer,
4, // 4 chars per token average
this.eventBus,
new InMemoryFS(),
new DetIdGen()
new InMemoryFileSystem(),
new DeterministicIdGenerator(),
);
this.orchestrator = new PipelineOrchestrator(config, this.env, this.eventBus, this.tracer, registry);
this.contextManager = ContextManager.create(config, this.env, this.tracer, this.orchestrator, registry);
this.orchestrator = new PipelineOrchestrator(
config,
this.env,
this.eventBus,
this.tracer,
registry,
);
this.contextManager = ContextManager.create(
config,
this.env,
this.tracer,
this.orchestrator,
registry,
);
this.contextManager.subscribeToHistory(this.chatHistory);
}
@@ -92,61 +103,74 @@ export class SimulationHarness {
// 1. Append the new messages
const currentHistory = this.chatHistory.get();
this.chatHistory.set([...currentHistory, ...messages]);
// 2. Measure tokens immediately after append (Before background processing)
const tokensBefore = this.env.tokenCalculator.calculateEpisodeListTokens(
this.contextManager.getWorkingBufferView()
this.contextManager.getWorkingBufferView(),
);
debugLogger.log(`[Turn ${this.currentTurnIndex}] Tokens BEFORE: ${tokensBefore}`);
debugLogger.log(
`[Turn ${this.currentTurnIndex}] Tokens BEFORE: ${tokensBefore}`,
);
// 3. Yield to event loop to allow internal async subscribers and orchestrator to finish
await new Promise(resolve => setTimeout(resolve, 50));
await new Promise((resolve) => setTimeout(resolve, 50));
// 3.1 Simulate what projectCompressedHistory does with the sync handlers
let currentView = this.contextManager.getWorkingBufferView();
const currentTokens = this.env.tokenCalculator.calculateEpisodeListTokens(currentView);
const currentTokens =
this.env.tokenCalculator.calculateEpisodeListTokens(currentView);
if (this.config.budget && currentTokens > this.config.budget.maxTokens) {
debugLogger.log(`[Turn ${this.currentTurnIndex}] Sync panic triggered! ${currentTokens} > ${this.config.budget.maxTokens}`);
const syncPipelines = this.config.pipelines.filter(p => p.execution === 'blocking');
debugLogger.log(
`[Turn ${this.currentTurnIndex}] Sync panic triggered! ${currentTokens} > ${this.config.budget.maxTokens}`,
);
const syncPipelines = this.config.pipelines.filter(
(p) => p.execution === 'blocking',
);
const orchestrator = this.orchestrator;
for (const pipe of syncPipelines) {
await orchestrator.executePipeline(pipe.name, currentView, {
currentTokens,
maxTokens: this.config.budget.maxTokens,
retainedTokens: this.config.budget.retainedTokens,
isBudgetSatisfied: false,
deficitTokens: currentTokens - this.config.budget.maxTokens,
protectedEpisodeIds: new Set()
});
currentView = this.contextManager.getWorkingBufferView();
await orchestrator.executePipeline(pipe.name, currentView, {
currentTokens,
maxTokens: this.config.budget.maxTokens,
retainedTokens: this.config.budget.retainedTokens,
isBudgetSatisfied: false,
deficitTokens: currentTokens - this.config.budget.maxTokens,
protectedEpisodeIds: new Set(),
});
currentView = this.contextManager.getWorkingBufferView();
}
// Inject the truncated view back into the graph
for (let i = 0; i < currentView.length; i++) {
const ep = currentView[i];
if (!this.contextManager.getWorkingBufferView().find(c => c.id === ep.id)) {
this.eventBus.emitVariantReady({
targetId: ep.id,
variantId: 'v-emergency',
variant: {
status: 'ready',
type: 'masked', // Truncation is technically a mask
text: ep.yield?.text || '',
recoveredTokens: 0,
}
});
}
const ep = currentView[i];
if (
!this.contextManager
.getWorkingBufferView()
.find((c) => c.id === ep.id)
) {
this.eventBus.emitVariantReady({
targetId: ep.id,
variantId: 'v-emergency',
variant: {
status: 'ready',
type: 'masked', // Truncation is technically a mask
text: ep.yield?.text || '',
recoveredTokens: 0,
},
});
}
}
// Wait for variant propagation
await new Promise(resolve => setTimeout(resolve, 50));
await new Promise((resolve) => setTimeout(resolve, 50));
}
// 4. Measure tokens after background processors have (hopefully) emitted variants
const tokensAfter = this.env.tokenCalculator.calculateEpisodeListTokens(
this.contextManager.getWorkingBufferView()
this.contextManager.getWorkingBufferView(),
);
debugLogger.log(`[Turn ${this.currentTurnIndex}] Tokens AFTER: ${tokensAfter}`);
debugLogger.log(
`[Turn ${this.currentTurnIndex}] Tokens AFTER: ${tokensAfter}`,
);
this.tokenTrajectory.push({
turnIndex: this.currentTurnIndex++,
tokensBeforeBackground: tokensBefore,
@@ -155,10 +179,11 @@ export class SimulationHarness {
}
async getGoldenState() {
const finalProjection = await this.contextManager.projectCompressedHistory();
const finalProjection =
await this.contextManager.projectCompressedHistory();
return {
tokenTrajectory: this.tokenTrajectory,
finalProjection
finalProjection,
};
}
}
packages/core/src/context/system-tests/lifecycle.golden.test.ts
+65 -23
View File
@@ -12,9 +12,14 @@ import type { BaseLlmClient } from '../../core/baseLlmClient.js';
expect.addSnapshotSerializer({
test: (val) =>
typeof val === 'string' &&
(/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(val) ||
/^\/tmp\/sim/.test(val)), // Mask temp directories and UUIDs
print: (val) => (typeof val === 'string' && /^\/tmp\/sim/.test(val) ? '"<MOCKED_DIR>"' : '"<UUID>"'),
(/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(
val,
) ||
/^\/tmp\/sim/.test(val)), // Mask temp directories and UUIDs
print: (val) =>
typeof val === 'string' && /^\/tmp\/sim/.test(val)
? '"<MOCKED_DIR>"'
: '"<UUID>"',
});
describe('System Lifecycle Golden Tests', () => {
@@ -36,69 +41,106 @@ describe('System Lifecycle Golden Tests', () => {
triggers: ['budget_exceeded'],
processors: [
{ processorId: 'BlobDegradationProcessor' },
{ processorId: 'ToolMaskingProcessor', options: { stringLengthThresholdTokens: 50 } }, // Mask any tool string > 200 chars
{ processorId: 'StateSnapshotProcessor', options: {} } // Squash old history
]
{
processorId: 'ToolMaskingProcessor',
options: { stringLengthThresholdTokens: 50 },
}, // Mask any tool string > 200 chars
{ processorId: 'StateSnapshotProcessor', options: {} }, // Squash old history
],
},
{
name: 'Immediate Sanitization', // The magic string the projector is hardcoded to use
execution: 'blocking',
triggers: ['budget_exceeded'],
processors: [
{ processorId: 'EmergencyTruncationProcessor', options: {} }
]
}
]
{ processorId: 'EmergencyTruncationProcessor', options: {} },
],
},
],
});
const mockLlmClient = {
generateContent: vi.fn().mockResolvedValue({
text: '<MOCKED_STATE_SNAPSHOT_SUMMARY>',
})
}),
} as unknown as BaseLlmClient;
it('Scenario 1: Organic Growth with Huge Tool Output & Images', async () => {
const harness = await SimulationHarness.create(getAggressiveConfig(), mockLlmClient);
const harness = await SimulationHarness.create(
getAggressiveConfig(),
mockLlmClient,
);
// Turn 0: System Prompt
await harness.simulateTurn([
{ role: 'user', parts: [{ text: 'System Instructions' }] },
{ role: 'model', parts: [{ text: 'Ack.' }] }
{ role: 'model', parts: [{ text: 'Ack.' }] },
]);
// Turn 1: Normal conversation
await harness.simulateTurn([
{ role: 'user', parts: [{ text: 'Hello!' }] },
{ role: 'model', parts: [{ text: 'Hi, how can I help?' }] }
{ role: 'model', parts: [{ text: 'Hi, how can I help?' }] },
]);
// Turn 2: Massive Tool Output (Should trigger ToolMaskingProcessor in background)
await harness.simulateTurn([
{ role: 'user', parts: [{ text: 'Read the logs.' }] },
{ role: 'model', parts: [{ functionCall: { name: 'run_shell_command', args: { cmd: 'cat server.log' } } }] },
{ role: 'user', parts: [{ functionResponse: { name: 'run_shell_command', response: { output: 'LOG '.repeat(5000) } } }] },
{ role: 'model', parts: [{ text: 'The logs are very long.' }] }
{
role: 'model',
parts: [
{
functionCall: {
name: 'run_shell_command',
args: { cmd: 'cat server.log' },
},
},
],
},
{
role: 'user',
parts: [
{
functionResponse: {
name: 'run_shell_command',
response: { output: 'LOG '.repeat(5000) },
},
},
],
},
{ role: 'model', parts: [{ text: 'The logs are very long.' }] },
]);
// Turn 3: Multi-modal blob (Should trigger BlobDegradationProcessor)
await harness.simulateTurn([
{ role: 'user', parts: [{ text: 'Look at this architecture diagram:' }, { inlineData: { mimeType: 'image/png', data: 'fake_base64_data_'.repeat(1000) } }] },
{ role: 'model', parts: [{ text: 'Nice diagram.' }] }
{
role: 'user',
parts: [
{ text: 'Look at this architecture diagram:' },
{
inlineData: {
mimeType: 'image/png',
data: 'fake_base64_data_'.repeat(1000),
},
},
],
},
{ role: 'model', parts: [{ text: 'Nice diagram.' }] },
]);
// Turn 4: More conversation to trigger StateSnapshot
await harness.simulateTurn([
{ role: 'user', parts: [{ text: 'Can we refactor?' }] },
{ role: 'model', parts: [{ text: 'Yes we can.' }] }
{ role: 'model', parts: [{ text: 'Yes we can.' }] },
]);
// Get final state
const goldenState = await harness.getGoldenState();
// In a perfectly functioning opportunistic system, the token trajectory should show
// In a perfectly functioning opportunistic system, the token trajectory should show
// the massive spikes in Turn 2 and 3 being immediately resolved by the background tasks.
// The final projection should fit neatly under the Max Tokens limit.
expect(goldenState).toMatchSnapshot();
});
});