Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
45 changes: 24 additions & 21 deletions evals/apps/cli/src/index.ts
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@ import {
IpcMessageType,
TaskCommandName,
rooCodeDefaults,
EvalEventName,
} from "@evals/types"
import {
type Run,
Expand All @@ -34,7 +35,7 @@ import { IpcServer, IpcClient } from "@evals/ipc"
import { __dirname, extensionDevelopmentPath, exercisesPath } from "./paths.js"
import { getExercises } from "./exercises.js"

type TaskResult = { success: boolean; retry: boolean }
type TaskResult = { success: boolean }
type TaskPromise = Promise<TaskResult>

const TASK_START_DELAY = 10 * 1_000
Expand Down Expand Up @@ -116,24 +117,25 @@ const run = async (toolbox: GluegunToolbox) => {

const runningPromises: TaskPromise[] = []

// Retries aren't implemented yet, but the return values are set up to
// support them.
const processTask = async (task: Task, delay = 0) => {
if (task.finishedAt === null) {
await new Promise((resolve) => setTimeout(resolve, delay))
const { retry } = await runExercise({ run, task, server })

if (retry) {
return { success: false, retry: true }
}
await runExercise({ run, task, server })
}

if (task.passed === null) {
const passed = await runUnitTest({ task })
await updateTask(task.id, { passed })
return { success: passed, retry: false }

server.broadcast({
type: IpcMessageType.TaskEvent,
origin: IpcOrigin.Server,
data: { eventName: passed ? EvalEventName.Pass : EvalEventName.Fail, taskId: task.id },
})

return { success: passed }
} else {
return { success: task.passed, retry: false }
return { success: task.passed }
}
}

Expand Down Expand Up @@ -200,7 +202,7 @@ const runExercise = async ({ run, task, server }: { run: Run; task: Task; server
} catch (error) {
console.log(`${Date.now()} [cli#runExercise | ${language} / ${exercise}] unable to connect`)
client.disconnect()
return { success: false, retry: false }
return { success: false }
}

let taskStartedAt = Date.now()
Expand All @@ -209,16 +211,15 @@ const runExercise = async ({ run, task, server }: { run: Run; task: Task; server
let rooTaskId: string | undefined
let isClientDisconnected = false

const ignoreEvents: RooCodeEventName[] = [
RooCodeEventName.Message,
RooCodeEventName.TaskTokenUsageUpdated,
RooCodeEventName.TaskAskResponded,
]
const ignoreEvents: Record<"broadcast" | "log", (RooCodeEventName | EvalEventName)[]> = {
broadcast: [RooCodeEventName.Message],
log: [RooCodeEventName.Message, RooCodeEventName.TaskTokenUsageUpdated, RooCodeEventName.TaskAskResponded],
}

client.on(IpcMessageType.TaskEvent, async (taskEvent) => {
const { eventName, payload } = taskEvent

if (taskEvent.eventName !== RooCodeEventName.Message) {
if (!ignoreEvents.broadcast.includes(eventName)) {
server.broadcast({
type: IpcMessageType.TaskEvent,
origin: IpcOrigin.Server,
Expand All @@ -227,7 +228,7 @@ const runExercise = async ({ run, task, server }: { run: Run; task: Task; server
})
}

if (!ignoreEvents.includes(eventName)) {
if (!ignoreEvents.log.includes(eventName)) {
console.log(
`${Date.now()} [cli#runExercise | ${language} / ${exercise}] taskEvent -> ${eventName}`,
payload,
Expand Down Expand Up @@ -320,11 +321,10 @@ const runExercise = async ({ run, task, server }: { run: Run; task: Task; server
data: { commandName: TaskCommandName.CancelTask, data: rooTaskId },
})

// Give the server some time to cancel the task.
// Allow some time for the task to cancel.
await new Promise((resolve) => setTimeout(resolve, 5_000))
}

// TODO: Notify clients that the task timed out.
await updateTask(task.id, { finishedAt: new Date() })
}

Expand All @@ -336,12 +336,15 @@ const runExercise = async ({ run, task, server }: { run: Run; task: Task; server
clientId: client.clientId!,
data: { commandName: TaskCommandName.CloseTask, data: rooTaskId },
})

// Allow some time for the window to close.
await new Promise((resolve) => setTimeout(resolve, 2_000))
}

client.disconnect()
}

return { success: !!taskFinishedAt, retry: false }
return { success: !!taskFinishedAt }
}

const runUnitTest = async ({ task }: { task: Task }) => {
Expand Down
10 changes: 5 additions & 5 deletions evals/apps/web/src/hooks/use-run-status.ts
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
import { useState, useCallback, useRef } from "react"
import { useQuery, keepPreviousData } from "@tanstack/react-query"

import { RooCodeEventName, taskEventSchema, TokenUsage } from "@evals/types"
import { TokenUsage, taskEventSchema, RooCodeEventName, EvalEventName } from "@evals/types"
import { Run } from "@evals/db"

import { getTasks } from "@/lib/server/tasks"
Expand Down Expand Up @@ -51,17 +51,17 @@ export const useRunStatus = (run: Run) => {
case RooCodeEventName.TaskStarted:
startTimes.current.set(taskId, Date.now())
break
case RooCodeEventName.TaskCompleted:
case RooCodeEventName.TaskAborted:
setTasksUpdatedAt(Date.now())
break
case RooCodeEventName.TaskTokenUsageUpdated: {
const startTime = startTimes.current.get(taskId)
const duration = startTime ? Date.now() - startTime : undefined
tokenUsage.current.set(taskId, { ...payload[1], duration })
setUsageUpdatedAt(Date.now())
break
}
case EvalEventName.Pass:
case EvalEventName.Fail:
setTasksUpdatedAt(Date.now())
break
}
}, [])

Expand Down
21 changes: 16 additions & 5 deletions evals/packages/types/src/ipc.ts
Original file line number Diff line number Diff line change
Expand Up @@ -50,12 +50,12 @@ export type TaskCommand = z.infer<typeof taskCommandSchema>
* TaskEvent
*/

export enum EvalEventName {
Pass = "pass",
Fail = "fail",
}

export const taskEventSchema = z.discriminatedUnion("eventName", [
z.object({
eventName: z.literal(RooCodeEventName.Connect),
payload: z.unknown(),
taskId: z.number(),
}),
z.object({
eventName: z.literal(RooCodeEventName.Message),
payload: rooCodeEventsSchema.shape[RooCodeEventName.Message],
Expand Down Expand Up @@ -111,6 +111,16 @@ export const taskEventSchema = z.discriminatedUnion("eventName", [
payload: rooCodeEventsSchema.shape[RooCodeEventName.TaskTokenUsageUpdated],
taskId: z.number().optional(),
}),
z.object({
eventName: z.literal(EvalEventName.Pass),
payload: z.undefined(),
taskId: z.number(),
}),
z.object({
eventName: z.literal(EvalEventName.Fail),
payload: z.undefined(),
taskId: z.number(),
}),
])

export type TaskEvent = z.infer<typeof taskEventSchema>
Expand All @@ -125,6 +135,7 @@ export enum IpcMessageType {
Ack = "Ack",
TaskCommand = "TaskCommand",
TaskEvent = "TaskEvent",
EvalEvent = "EvalEvent",
}

export enum IpcOrigin {
Expand Down