feat: implement experimental features system

- Add experiments.ts to manage experimental features
- Refactor experimental diff strategy into experiments system
- Add UI components for managing experimental features
- Add tests for experimental tools
- Update system prompts to handle experiments
This commit is contained in:
sam hoang
2025-01-27 03:48:24 +07:00
parent 2c97b59ed1
commit ad552ea026
14 changed files with 429 additions and 69 deletions

View File

@@ -61,6 +61,7 @@ import { OpenRouterHandler } from "../api/providers/openrouter"
import { McpHub } from "../services/mcp/McpHub"
import crypto from "crypto"
import { insertGroups } from "./diff/insert-groups"
import { EXPERIMENT_IDS } from "../shared/experiments"
const cwd =
vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0) ?? path.join(os.homedir(), "Desktop") // may or may not exist but fs checking existence would immediately ask for permission which would be bad UX, need to come up with a better solution
@@ -151,9 +152,8 @@ export class Cline {
async updateDiffStrategy(experimentalDiffStrategy?: boolean) {
// If not provided, get from current state
if (experimentalDiffStrategy === undefined) {
const { experimentalDiffStrategy: stateExperimentalDiffStrategy } =
(await this.providerRef.deref()?.getState()) ?? {}
experimentalDiffStrategy = stateExperimentalDiffStrategy ?? false
const { experiments: stateExperimental } = (await this.providerRef.deref()?.getState()) ?? {}
experimentalDiffStrategy = stateExperimental?.[EXPERIMENT_IDS.DIFF_STRATEGY] ?? false
}
this.diffStrategy = getDiffStrategy(this.api.getModel().id, this.fuzzyMatchThreshold, experimentalDiffStrategy)
}
@@ -810,7 +810,7 @@ export class Cline {
})
}
const { browserViewportSize, mode, customModePrompts, preferredLanguage } =
const { browserViewportSize, mode, customModePrompts, preferredLanguage, experiments } =
(await this.providerRef.deref()?.getState()) ?? {}
const { customModes } = (await this.providerRef.deref()?.getState()) ?? {}
const systemPrompt = await (async () => {
@@ -831,6 +831,7 @@ export class Cline {
this.customInstructions,
preferredLanguage,
this.diffEnabled,
experiments,
)
})()

View File

@@ -11,6 +11,7 @@ import { defaultModeSlug, modes } from "../../../shared/modes"
import "../../../utils/path"
import { addCustomInstructions } from "../sections/custom-instructions"
import * as modesSection from "../sections/modes"
import { EXPERIMENT_IDS } from "../../../shared/experiments"
// Mock the sections
jest.mock("../sections/modes", () => ({
@@ -121,6 +122,7 @@ const createMockMcpHub = (): McpHub =>
describe("SYSTEM_PROMPT", () => {
let mockMcpHub: McpHub
let experiments: Record<string, boolean>
beforeAll(() => {
// Ensure fs mock is properly initialized
@@ -140,6 +142,10 @@ describe("SYSTEM_PROMPT", () => {
"/mock/mcp/path",
]
dirs.forEach((dir) => mockFs._mockDirectories.add(dir))
experiments = {
[EXPERIMENT_IDS.SEARCH_AND_REPLACE]: false,
[EXPERIMENT_IDS.INSERT_BLOCK]: false,
}
})
beforeEach(() => {
@@ -164,6 +170,10 @@ describe("SYSTEM_PROMPT", () => {
defaultModeSlug, // mode
undefined, // customModePrompts
undefined, // customModes
undefined, // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments,
)
expect(prompt).toMatchSnapshot()
@@ -179,7 +189,11 @@ describe("SYSTEM_PROMPT", () => {
"1280x800", // browserViewportSize
defaultModeSlug, // mode
undefined, // customModePrompts
undefined, // customModes
undefined, // customModes,
undefined, // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments,
)
expect(prompt).toMatchSnapshot()
@@ -197,7 +211,11 @@ describe("SYSTEM_PROMPT", () => {
undefined, // browserViewportSize
defaultModeSlug, // mode
undefined, // customModePrompts
undefined, // customModes
undefined, // customModes,
undefined, // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments,
)
expect(prompt).toMatchSnapshot()
@@ -213,7 +231,11 @@ describe("SYSTEM_PROMPT", () => {
undefined, // browserViewportSize
defaultModeSlug, // mode
undefined, // customModePrompts
undefined, // customModes
undefined, // customModes,
undefined, // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments,
)
expect(prompt).toMatchSnapshot()
@@ -229,7 +251,11 @@ describe("SYSTEM_PROMPT", () => {
"900x600", // different viewport size
defaultModeSlug, // mode
undefined, // customModePrompts
undefined, // customModes
undefined, // customModes,
undefined, // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments,
)
expect(prompt).toMatchSnapshot()
@@ -249,6 +275,7 @@ describe("SYSTEM_PROMPT", () => {
undefined, // globalCustomInstructions
undefined, // preferredLanguage
true, // diffEnabled
experiments,
)
expect(prompt).toContain("apply_diff")
@@ -269,6 +296,7 @@ describe("SYSTEM_PROMPT", () => {
undefined, // globalCustomInstructions
undefined, // preferredLanguage
false, // diffEnabled
experiments,
)
expect(prompt).not.toContain("apply_diff")
@@ -289,6 +317,7 @@ describe("SYSTEM_PROMPT", () => {
undefined, // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments,
)
expect(prompt).not.toContain("apply_diff")
@@ -308,6 +337,8 @@ describe("SYSTEM_PROMPT", () => {
undefined, // customModes
undefined, // globalCustomInstructions
"Spanish", // preferredLanguage
undefined, // diffEnabled
experiments,
)
expect(prompt).toContain("Language Preference:")
@@ -337,6 +368,9 @@ describe("SYSTEM_PROMPT", () => {
undefined, // customModePrompts
customModes, // customModes
"Global instructions", // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments,
)
// Role definition should be at the top
@@ -368,6 +402,10 @@ describe("SYSTEM_PROMPT", () => {
defaultModeSlug,
customModePrompts,
undefined,
undefined,
undefined,
undefined,
experiments,
)
// Role definition from promptComponent should be at the top
@@ -394,18 +432,101 @@ describe("SYSTEM_PROMPT", () => {
defaultModeSlug,
customModePrompts,
undefined,
undefined,
undefined,
undefined,
experiments,
)
// Should use the default mode's role definition
expect(prompt.indexOf(modes[0].roleDefinition)).toBeLessThan(prompt.indexOf("TOOL USE"))
})
describe("experimental tools", () => {
it("should disable experimental tools by default", async () => {
const prompt = await SYSTEM_PROMPT(
mockContext,
"/test/path",
false, // supportsComputerUse
undefined, // mcpHub
undefined, // diffStrategy
undefined, // browserViewportSize
defaultModeSlug, // mode
undefined, // customModePrompts
undefined, // customModes
undefined, // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments, // experiments - undefined should disable all experimental tools
)
// Verify experimental tools are not included in the prompt
expect(prompt).not.toContain(EXPERIMENT_IDS.SEARCH_AND_REPLACE)
expect(prompt).not.toContain(EXPERIMENT_IDS.INSERT_BLOCK)
})
it("should enable experimental tools when explicitly enabled", async () => {
const experiments = {
[EXPERIMENT_IDS.SEARCH_AND_REPLACE]: true,
[EXPERIMENT_IDS.INSERT_BLOCK]: true,
}
const prompt = await SYSTEM_PROMPT(
mockContext,
"/test/path",
false, // supportsComputerUse
undefined, // mcpHub
undefined, // diffStrategy
undefined, // browserViewportSize
defaultModeSlug, // mode
undefined, // customModePrompts
undefined, // customModes
undefined, // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments,
)
// Verify experimental tools are included in the prompt when enabled
expect(prompt).toContain(EXPERIMENT_IDS.SEARCH_AND_REPLACE)
expect(prompt).toContain(EXPERIMENT_IDS.INSERT_BLOCK)
})
it("should selectively enable experimental tools", async () => {
const experiments = {
[EXPERIMENT_IDS.SEARCH_AND_REPLACE]: true,
[EXPERIMENT_IDS.INSERT_BLOCK]: false,
}
const prompt = await SYSTEM_PROMPT(
mockContext,
"/test/path",
false, // supportsComputerUse
undefined, // mcpHub
undefined, // diffStrategy
undefined, // browserViewportSize
defaultModeSlug, // mode
undefined, // customModePrompts
undefined, // customModes
undefined, // globalCustomInstructions
undefined, // preferredLanguage
undefined, // diffEnabled
experiments,
)
// Verify only enabled experimental tools are included
expect(prompt).toContain(EXPERIMENT_IDS.SEARCH_AND_REPLACE)
expect(prompt).not.toContain(EXPERIMENT_IDS.INSERT_BLOCK)
})
})
afterAll(() => {
jest.restoreAllMocks()
})
})
describe("addCustomInstructions", () => {
let experiments: Record<string, boolean>
beforeAll(() => {
// Ensure fs mock is properly initialized
const mockFs = jest.requireMock("fs/promises")
@@ -417,6 +538,11 @@ describe("addCustomInstructions", () => {
}
throw new Error(`ENOENT: no such file or directory, mkdir '${path}'`)
})
experiments = {
[EXPERIMENT_IDS.SEARCH_AND_REPLACE]: false,
[EXPERIMENT_IDS.INSERT_BLOCK]: false,
}
})
beforeEach(() => {
@@ -434,6 +560,10 @@ describe("addCustomInstructions", () => {
"architect", // mode
undefined, // customModePrompts
undefined, // customModes
undefined,
undefined,
undefined,
experiments,
)
expect(prompt).toMatchSnapshot()
@@ -450,6 +580,10 @@ describe("addCustomInstructions", () => {
"ask", // mode
undefined, // customModePrompts
undefined, // customModes
undefined,
undefined,
undefined,
experiments,
)
expect(prompt).toMatchSnapshot()

View File

@@ -39,6 +39,7 @@ async function generatePrompt(
globalCustomInstructions?: string,
preferredLanguage?: string,
diffEnabled?: boolean,
experiments?: Record<string, boolean>,
): Promise<string> {
if (!context) {
throw new Error("Extension context is required for generating system prompt")
@@ -68,6 +69,7 @@ ${getToolDescriptionsForMode(
browserViewportSize,
mcpHub,
customModeConfigs,
experiments,
)}
${getToolUseGuidelinesSection()}
@@ -102,6 +104,7 @@ export const SYSTEM_PROMPT = async (
globalCustomInstructions?: string,
preferredLanguage?: string,
diffEnabled?: boolean,
experiments?: Record<string, boolean>,
): Promise<string> => {
if (!context) {
throw new Error("Extension context is required for generating system prompt")
@@ -135,5 +138,6 @@ export const SYSTEM_PROMPT = async (
globalCustomInstructions,
preferredLanguage,
diffEnabled,
experiments,
)
}

View File

@@ -46,6 +46,7 @@ export function getToolDescriptionsForMode(
browserViewportSize?: string,
mcpHub?: McpHub,
customModes?: ModeConfig[],
experiments?: Record<string, boolean>,
): string {
const config = getModeConfig(mode, customModes)
const args: ToolArgs = {
@@ -64,7 +65,7 @@ export function getToolDescriptionsForMode(
const toolGroup = TOOL_GROUPS[groupName]
if (toolGroup) {
toolGroup.forEach((tool) => {
if (isToolAllowedForMode(tool as ToolName, mode, customModes ?? [])) {
if (isToolAllowedForMode(tool as ToolName, mode, customModes ?? [], experiments ?? {})) {
tools.add(tool)
}
})

View File

@@ -40,6 +40,12 @@ import { singleCompletionHandler } from "../../utils/single-completion-handler"
import { getCommitInfo, searchCommits, getWorkingState } from "../../utils/git"
import { ConfigManager } from "../config/ConfigManager"
import { CustomModesManager } from "../config/CustomModesManager"
import {
EXPERIMENT_IDS,
experimentConfigs,
experiments as Experiments,
experimentDefault,
} from "../../shared/experiments"
import { CustomSupportPrompts, supportPrompt } from "../../shared/support-prompt"
import { ACTION_NAMES } from "../CodeActionProvider"
@@ -118,7 +124,7 @@ type GlobalStateKey =
| "customModePrompts"
| "customSupportPrompts"
| "enhancementApiConfigId"
| "experimentalDiffStrategy"
| "experiments" // Map of experiment IDs to their enabled state
| "autoApprovalEnabled"
| "customModes" // Array of custom modes
@@ -339,7 +345,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
fuzzyMatchThreshold,
mode,
customInstructions: globalInstructions,
experimentalDiffStrategy,
experiments,
} = await this.getState()
const modePrompt = customModePrompts?.[mode] as PromptComponent
@@ -354,7 +360,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
task,
images,
undefined,
experimentalDiffStrategy,
Experiments.isEnabled(experiments, EXPERIMENT_IDS.DIFF_STRATEGY),
)
}
@@ -367,7 +373,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
fuzzyMatchThreshold,
mode,
customInstructions: globalInstructions,
experimentalDiffStrategy,
experiments,
} = await this.getState()
const modePrompt = customModePrompts?.[mode] as PromptComponent
@@ -382,7 +388,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
undefined,
undefined,
historyItem,
experimentalDiffStrategy,
Experiments.isEnabled(experiments, EXPERIMENT_IDS.DIFF_STRATEGY),
)
}
@@ -1044,14 +1050,14 @@ export class ClineProvider implements vscode.WebviewViewProvider {
diffEnabled,
mcpEnabled,
fuzzyMatchThreshold,
experimentalDiffStrategy,
experiments,
} = await this.getState()
// Create diffStrategy based on current model and settings
const diffStrategy = getDiffStrategy(
apiConfiguration.apiModelId || apiConfiguration.openRouterModelId || "",
fuzzyMatchThreshold,
experimentalDiffStrategy,
Experiments.isEnabled(experiments, EXPERIMENT_IDS.DIFF_STRATEGY),
)
const cwd =
vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0) || ""
@@ -1072,6 +1078,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
customInstructions,
preferredLanguage,
diffEnabled,
experiments,
)
await this.postMessageToWebview({
@@ -1207,14 +1214,28 @@ export class ClineProvider implements vscode.WebviewViewProvider {
vscode.window.showErrorMessage("Failed to get list api configuration")
}
break
case "experimentalDiffStrategy":
await this.updateGlobalState("experimentalDiffStrategy", message.bool ?? false)
// Update diffStrategy in current Cline instance if it exists
if (this.cline) {
await this.cline.updateDiffStrategy(message.bool ?? false)
case "updateExperimental": {
if (!message.values) {
break
}
const updatedExperiments = {
...((await this.getGlobalState("experiments")) ?? experimentDefault),
...message.values,
}
await this.updateGlobalState("experiments", updatedExperiments)
// Update diffStrategy in current Cline instance if it exists
if (message.values[EXPERIMENT_IDS.DIFF_STRATEGY] !== undefined && this.cline) {
await this.cline.updateDiffStrategy(
Experiments.isEnabled(updatedExperiments, EXPERIMENT_IDS.DIFF_STRATEGY),
)
}
await this.postStateToWebview()
break
}
case "updateMcpTimeout":
if (message.serverName && typeof message.timeout === "number") {
try {
@@ -1873,8 +1894,8 @@ export class ClineProvider implements vscode.WebviewViewProvider {
customModePrompts,
customSupportPrompts,
enhancementApiConfigId,
experimentalDiffStrategy,
autoApprovalEnabled,
experiments,
} = await this.getState()
const allowedCommands = vscode.workspace.getConfiguration("roo-cline").get<string[]>("allowedCommands") || []
@@ -1914,9 +1935,9 @@ export class ClineProvider implements vscode.WebviewViewProvider {
customModePrompts: customModePrompts ?? {},
customSupportPrompts: customSupportPrompts ?? {},
enhancementApiConfigId,
experimentalDiffStrategy: experimentalDiffStrategy ?? false,
autoApprovalEnabled: autoApprovalEnabled ?? false,
customModes: await this.customModesManager.getCustomModes(),
experiments: experiments ?? experimentDefault,
}
}
@@ -2039,9 +2060,9 @@ export class ClineProvider implements vscode.WebviewViewProvider {
customModePrompts,
customSupportPrompts,
enhancementApiConfigId,
experimentalDiffStrategy,
autoApprovalEnabled,
customModes,
experiments,
] = await Promise.all([
this.getGlobalState("apiProvider") as Promise<ApiProvider | undefined>,
this.getGlobalState("apiModelId") as Promise<string | undefined>,
@@ -2109,9 +2130,9 @@ export class ClineProvider implements vscode.WebviewViewProvider {
this.getGlobalState("customModePrompts") as Promise<CustomModePrompts | undefined>,
this.getGlobalState("customSupportPrompts") as Promise<CustomSupportPrompts | undefined>,
this.getGlobalState("enhancementApiConfigId") as Promise<string | undefined>,
this.getGlobalState("experimentalDiffStrategy") as Promise<boolean | undefined>,
this.getGlobalState("autoApprovalEnabled") as Promise<boolean | undefined>,
this.customModesManager.getCustomModes(),
this.getGlobalState("experiments") as Promise<Record<string, boolean> | undefined>,
])
let apiProvider: ApiProvider
@@ -2225,7 +2246,7 @@ export class ClineProvider implements vscode.WebviewViewProvider {
customModePrompts: customModePrompts ?? {},
customSupportPrompts: customSupportPrompts ?? {},
enhancementApiConfigId,
experimentalDiffStrategy: experimentalDiffStrategy ?? false,
experiments: experiments ?? experimentDefault,
autoApprovalEnabled: autoApprovalEnabled ?? false,
customModes,
}

View File

@@ -4,6 +4,7 @@ import { ExtensionMessage, ExtensionState } from "../../../shared/ExtensionMessa
import { setSoundEnabled } from "../../../utils/sound"
import { defaultModeSlug, modes } from "../../../shared/modes"
import { addCustomInstructions } from "../../prompts/sections/custom-instructions"
import { experimentDefault, experiments } from "../../../shared/experiments"
// Mock custom-instructions module
const mockAddCustomInstructions = jest.fn()
@@ -320,6 +321,7 @@ describe("ClineProvider", () => {
requestDelaySeconds: 5,
mode: defaultModeSlug,
customModes: [],
experiments: experimentDefault,
}
const message: ExtensionMessage = {
@@ -617,6 +619,7 @@ describe("ClineProvider", () => {
mode: "code",
diffEnabled: true,
fuzzyMatchThreshold: 1.0,
experiments: experimentDefault,
} as any)
// Reset Cline mock
@@ -636,7 +639,7 @@ describe("ClineProvider", () => {
"Test task",
undefined,
undefined,
undefined,
false,
)
})
test("handles mode-specific custom instructions updates", async () => {
@@ -887,6 +890,7 @@ describe("ClineProvider", () => {
},
mcpEnabled: true,
mode: "code" as const,
experiments: experimentDefault,
} as any)
const handler1 = getMessageHandler()
@@ -918,6 +922,7 @@ describe("ClineProvider", () => {
},
mcpEnabled: false,
mode: "code" as const,
experiments: experimentDefault,
} as any)
const handler2 = getMessageHandler()
@@ -985,6 +990,7 @@ describe("ClineProvider", () => {
experimentalDiffStrategy: true,
diffEnabled: true,
fuzzyMatchThreshold: 0.8,
experiments: experimentDefault,
} as any)
// Mock SYSTEM_PROMPT to verify diffStrategy and diffEnabled are passed
@@ -1012,6 +1018,7 @@ describe("ClineProvider", () => {
undefined, // effectiveInstructions
undefined, // preferredLanguage
true, // diffEnabled
experimentDefault,
)
// Run the test again to verify it's consistent
@@ -1034,6 +1041,7 @@ describe("ClineProvider", () => {
experimentalDiffStrategy: true,
diffEnabled: false,
fuzzyMatchThreshold: 0.8,
experiments: experimentDefault,
} as any)
// Mock SYSTEM_PROMPT to verify diffEnabled is passed as false
@@ -1061,6 +1069,7 @@ describe("ClineProvider", () => {
undefined, // effectiveInstructions
undefined, // preferredLanguage
false, // diffEnabled
experimentDefault,
)
})
@@ -1077,6 +1086,7 @@ describe("ClineProvider", () => {
mode: "architect",
mcpEnabled: false,
browserViewportSize: "900x600",
experiments: experimentDefault,
} as any)
// Mock SYSTEM_PROMPT to call addCustomInstructions