niceeval 0.14.1-canary.51 → 0.14.2

package/node_modules/undici/lib/handler/retry-handler.js

@@ -2,7 +2,7 @@
 const assert = require('node:assert')
 
 const { kRetryHandlerDefaultRetry } = require('../core/symbols')
-const { RequestRetryError } = require('../core/errors')
+const { RequestRetryError, RequestAbortedError } = require('../core/errors')
 const {
   isDisturbed,
   parseRangeHeader,
@@ -41,14 +41,26 @@ function validatePartialResponseContentLength (headers, range, statusCode, retry
 // new one: backpressure pauses the new connection's controller, but the
 // consumer's resume() targets the old one, so the resumed body stalls forever.
 // The proxy always forwards to the controller of the currently active connection.
+// An abort is additionally reported to the handler so it can cancel a pending
+// retry backoff instead of letting the request hang until the backoff elapses.
+// The notification is a private callback the handler hands over on construction,
+// so nothing outside the handler can trigger it.
 class RetryController {
-  constructor () {
+  #onAbort
+
+  constructor (onAbort) {
+    this.#onAbort = onAbort
     this.target = null
   }
 
   pause () { this.target?.pause() }
   resume () { this.target?.resume() }
-  abort (reason) { this.target?.abort(reason) }
+
+  abort (reason) {
+    this.target?.abort(reason)
+    this.#onAbort(reason)
+  }
+
   get paused () { return this.target?.paused ?? false }
   get aborted () { return this.target?.aborted ?? false }
   get reason () { return this.target?.reason ?? null }
@@ -112,7 +124,16 @@ class RetryHandler {
     this.etag = null
     this.statusCode = null
     this.headers = null
-    this.controllerProxy = new RetryController()
+    this.controllerProxy = new RetryController(reason => this.#onAbort(reason))
+    // A retry decision is in flight (the policy may be holding a backoff
+    // timer). While pending, a consumer abort cancels the wait.
+    this.retryPending = false
+    // Backoff timer returned by the retry policy, so #onAbort can cancel it.
+    // Null for custom policies that do not return their timer.
+    this.retryTimer = null
+    // Set once an abort during the backoff delivered the terminal error
+    // downstream; late policy callbacks and connection errors are then moot.
+    this.aborted = false
   }
 
   onResponseStartWithRetry (controller, statusCode, headers, statusMessage, err) {
@@ -135,6 +156,13 @@ class RetryHandler {
     }
 
     function shouldRetry (passedErr) {
+      if (this.aborted) {
+        // Aborted while the policy was deciding; the decision is moot.
+        return
+      }
+      this.retryPending = false
+      this.retryTimer = null
+
       if (passedErr) {
         this.headersSent = true
         this.handler.onResponseStart?.(this.controllerProxy, statusCode, headers, statusMessage)
@@ -154,14 +182,17 @@ class RetryHandler {
     // between, leaving this one paused forever -- the very stall the proxy exists
     // to prevent.
     controller.pause()
-    this.retryOpts.retry(
+    // The default policy returns its backoff timer so an abort can cancel it;
+    // a custom policy may return anything (or nothing), which is ignored.
+    this.retryPending = true
+    this.retryTimer = this.retryOpts.retry(
       err,
       {
         state: { counter: this.retryCount },
         opts: { retryOptions: this.retryOpts, ...this.opts }
       },
       shouldRetry.bind(this)
-    )
+    ) ?? null
   }
 
   onRequestStart (controller, context) {
@@ -176,6 +207,14 @@ class RetryHandler {
     }
   }
 
+  onBodySent (chunk) {
+    this.handler.onBodySent?.(chunk)
+  }
+
+  onRequestSent () {
+    this.handler.onRequestSent?.()
+  }
+
   onRequestUpgrade (_controller, statusCode, headers, socket) {
     this.handler.onRequestUpgrade?.(this.controllerProxy, statusCode, headers, socket)
   }
@@ -190,7 +229,8 @@ class RetryHandler {
       timeoutFactor,
       statusCodes,
       errorCodes,
-      methods
+      methods,
+      retryAfter
     } = retryOptions
     const { counter } = state
 
@@ -222,7 +262,7 @@ class RetryHandler {
       return
     }
 
-    let retryAfterHeader = headers?.['retry-after']
+    let retryAfterHeader = retryAfter === false ? undefined : headers?.['retry-after']
     if (retryAfterHeader) {
       retryAfterHeader = Number(retryAfterHeader)
       retryAfterHeader = Number.isNaN(retryAfterHeader)
@@ -237,7 +277,9 @@ class RetryHandler {
           ? Math.min(retryAfterHeader, maxTimeout)
           : Math.min(minTimeout * timeoutFactor ** (counter - 1), maxTimeout)
 
-    setTimeout(() => cb(null), retryTimeout)
+    // Return the backoff timer so the handler can cancel it when the
+    // consumer aborts while the retry decision is pending.
+    return setTimeout(() => cb(null), retryTimeout)
   }
 
   onResponseStart (controller, statusCode, headers, statusMessage) {
@@ -435,6 +477,12 @@ class RetryHandler {
   }
 
   onResponseError (controller, err) {
+    if (this.aborted) {
+      // #onAbort already delivered the terminal error downstream; the late
+      // error of the torn-down connection must not be forwarded twice.
+      return
+    }
+
     // controller is THIS failed connection (not the proxy): we inspect whether
     // the consumer aborted it to decide retry-vs-propagate.
     if (controller?.aborted || isDisturbed(this.opts.body)) {
@@ -443,6 +491,13 @@ class RetryHandler {
     }
 
     function shouldRetry (returnedErr) {
+      if (this.aborted) {
+        // Aborted while the policy was deciding; the decision is moot.
+        return
+      }
+      this.retryPending = false
+      this.retryTimer = null
+
       if (!returnedErr) {
         this.retry()
         return
@@ -462,14 +517,31 @@ class RetryHandler {
       this.retryCount += 1
     }
 
-    this.retryOpts.retry(
+    this.retryPending = true
+    this.retryTimer = this.retryOpts.retry(
       err,
       {
         state: { counter: this.retryCount },
         opts: { retryOptions: this.retryOpts, ...this.opts }
       },
       shouldRetry.bind(this)
-    )
+    ) ?? null
+  }
+
+  #onAbort (reason) {
+    // A consumer abort lands on the controller proxy. If the retry policy is
+    // still deciding (typically holding a backoff timer), cancel the wait and
+    // surface the abort immediately instead of letting the request hang until
+    // the backoff elapses.
+    if (!this.retryPending) {
+      return
+    }
+
+    this.aborted = true
+    this.retryPending = false
+    clearTimeout(this.retryTimer)
+    this.retryTimer = null
+    this.handler.onResponseError?.(this.controllerProxy, reason ?? new RequestAbortedError())
   }
 }
 

package/node_modules/undici/lib/interceptor/dump.js

@@ -84,7 +84,7 @@ class DumpHandler extends DecoratorHandler {
       return
     }
 
-    if (this.#controller.aborted === true) {
+    if (this.aborted === true) {
       super.onResponseError(controller, this.reason)
       return
     }

package/node_modules/undici/lib/mock/mock-agent.js

@@ -73,7 +73,7 @@ class MockAgent extends Dispatcher {
     opts.origin = normalizeOrigin(opts.origin)
 
     // Call MockAgent.get to perform additional setup before dispatching as normal
-    this.get(opts.origin)
+    const mockDispatcher = this.get(opts.origin)
 
     this[kMockAgentAddCallHistoryLog](opts)
 
@@ -81,6 +81,18 @@ class MockAgent extends Dispatcher {
 
     const dispatchOpts = { ...opts }
 
+    // Agent keeps HTTP/1.1-only dispatchers under a separate key. Legacy
+    // global dispatcher consumers use that path, so mirror the mock dispatches
+    // before delegating to the internal Agent.
+    if (dispatchOpts.allowH2 === false) {
+      const http1OnlyKey = `${dispatchOpts.origin}#http1-only`
+      if (!this[kClients].has(http1OnlyKey)) {
+        const http1OnlyDispatcher = this[kFactory](dispatchOpts.origin)
+        http1OnlyDispatcher[kDispatches] = mockDispatcher[kDispatches]
+        this[kMockAgentSet](http1OnlyKey, http1OnlyDispatcher)
+      }
+    }
+
     if (acceptNonStandardSearchParameters && dispatchOpts.path) {
       const [path, searchParams] = dispatchOpts.path.split('?')
       const normalizedSearchParams = normalizeSearchParams(searchParams, acceptNonStandardSearchParameters)

package/node_modules/undici/lib/mock/mock-utils.js

@@ -333,8 +333,7 @@ function mockDispatch (opts, handler) {
             handler.onResponseError(null, new InvalidArgumentError('reply options callback must return an object'))
             return
           }
-          mockDispatch.data = { ...responseDefaults, ...resolvedData }
-          dispatchMockReply(mockDispatches, mockDispatch, key, opts, handler)
+          dispatchMockReply(mockDispatches, mockDispatch, key, opts, handler, { ...responseDefaults, ...resolvedData })
         },
         (error) => {
           handler.onResponseError(null, error)
@@ -347,7 +346,7 @@ function mockDispatch (opts, handler) {
       throw new InvalidArgumentError('reply options callback must return an object')
     }
 
-    mockDispatch.data = { ...responseDefaults, ...callbackResult }
+    return dispatchMockReply(mockDispatches, mockDispatch, key, opts, handler, { ...responseDefaults, ...callbackResult })
   }
 
   return dispatchMockReply(mockDispatches, mockDispatch, key, opts, handler)
@@ -356,9 +355,13 @@ function mockDispatch (opts, handler) {
 /**
  * Replies to a request once the mock dispatch data is fully resolved
  */
-function dispatchMockReply (mockDispatches, mockDispatch, key, opts, handler) {
-  // Parse mockDispatch data
-  const { data: response, delay } = mockDispatch
+function dispatchMockReply (mockDispatches, mockDispatch, key, opts, handler, resolvedResponse) {
+  // Parse mockDispatch data. When a reply callback has already been resolved
+  // in mockDispatch() (i.e. no body lifecycle hooks are involved), the resolved
+  // response is passed in here, leaving mockDispatch.data untouched so the
+  // callback can be re-invoked for persistent / times() replies.
+  const { data: responseData, delay } = mockDispatch
+  const response = resolvedResponse ?? responseData
 
   // If specified, trigger dispatch error
   if (response.error !== null) {
@@ -454,8 +457,7 @@ function dispatchMockReply (mockDispatches, mockDispatch, key, opts, handler) {
               handler.onResponseError(null, new InvalidArgumentError('reply options callback must return an object'))
               return
             }
-            mockDispatch.data = { ...responseDefaults, ...resolvedData }
-            handleReply(dispatches, mockDispatch.data)
+            handleReply(dispatches, { ...responseDefaults, ...resolvedData })
           },
           (err) => {
             handler.onResponseError(null, err)
@@ -468,8 +470,7 @@ function dispatchMockReply (mockDispatches, mockDispatch, key, opts, handler) {
         throw new InvalidArgumentError('reply options callback must return an object')
       }
 
-      mockDispatch.data = { ...responseDefaults, ...callbackResult }
-      handleReply(dispatches, mockDispatch.data)
+      handleReply(dispatches, { ...responseDefaults, ...callbackResult })
       return
     }
 

package/node_modules/undici/lib/web/cookies/parse.js

@@ -189,14 +189,16 @@ function parseUnparsedAttributes (unparsedAttributes, cookieAttributeList = {})
     // 1. If the first character of the attribute-value is not a DIGIT or a
     //    "-" character, ignore the cookie-av.
     const charCode = attributeValue.charCodeAt(0)
+    const startsWithDigit = charCode >= 48 && charCode <= 57
+    const startsWithSignedDigit = attributeValue[0] === '-' && attributeValue.length > 1
 
-    if ((charCode < 48 || charCode > 57) && attributeValue[0] !== '-') {
+    if (!startsWithDigit && !startsWithSignedDigit) {
       return parseUnparsedAttributes(unparsedAttributes, cookieAttributeList)
     }
 
     // 2. If the remainder of attribute-value contains a non-DIGIT
     //    character, ignore the cookie-av.
-    if (!/^\d+$/.test(attributeValue)) {
+    if (/[^\d]/.test(attributeValue.slice(1))) {
       return parseUnparsedAttributes(unparsedAttributes, cookieAttributeList)
     }
 

package/node_modules/undici/lib/web/eventsource/eventsource-stream.js

@@ -1,4 +1,5 @@
 'use strict'
+const buffer = require('node:buffer')
 const { Transform } = require('node:stream')
 const { isASCIINumber, isValidLastEventId } = require('./util')
 
@@ -23,6 +24,8 @@ const COLON = 0x3A
  */
 const SPACE = 0x20
 
+const defaultMaxEventSize = buffer.kStringMaxLength
+
 const DATA = Buffer.from('data')
 const EVENT = Buffer.from('event')
 const ID = Buffer.from('id')
@@ -66,6 +69,12 @@ function isFieldName (line, length, field) {
   return true
 }
 
+function createMaxEventSizeExceededError () {
+  const error = new Error('EventSource message size exceeded')
+  error.aborted = false
+  return error
+}
+
 /**
  * @typedef {object} EventSourceStreamEvent
  * @type {object}
@@ -114,6 +123,8 @@ class EventSourceStream extends Transform {
   pos = 0
   lineChunkIndex = 0
   linePos = 0
+  eventDataSize = 0
+  maxEventSize
 
   event = {
     data: undefined,
@@ -125,6 +136,7 @@ class EventSourceStream extends Transform {
   /**
    * @param {object} options
    * @param {boolean} [options.readableObjectMode]
+   * @param {number} [options.maxEventSize]
    * @param {eventSourceSettings} [options.eventSourceSettings]
    * @param {(chunk: any, encoding?: BufferEncoding | undefined) => boolean} [options.push]
    */
@@ -136,6 +148,7 @@ class EventSourceStream extends Transform {
     super(options)
 
     this.state = options.eventSourceSettings || {}
+    this.maxEventSize = options.maxEventSize ?? defaultMaxEventSize
     if (options.push) {
       this.push = options.push
     }
@@ -231,7 +244,12 @@ class EventSourceStream extends Transform {
 
         // In any case, we can process the line as we reached an
         // end-of-line character
-        this.parseLine(this.readLine(), this.event)
+        try {
+          this.parseLine(this.readLine(), this.event)
+        } catch (error) {
+          callback(error)
+          return
+        }
         this.consumeCurrentByte()
         // A line was processed and this could be the end of the event. We need
         // to check if the next line is empty to determine if the event is
@@ -282,6 +300,13 @@ class EventSourceStream extends Transform {
     }
 
     if (isFieldName(line, fieldLength, DATA)) {
+      const valueBytes = line.length - valueStart
+      const eventDataSize = this.eventDataSize + (event.data === undefined ? 0 : 1) + valueBytes
+
+      if (this.maxEventSize > 0 && eventDataSize > this.maxEventSize) {
+        throw createMaxEventSizeExceededError()
+      }
+
       const value = line.toString('utf8', valueStart)
 
       if (event.data === undefined) {
@@ -289,6 +314,7 @@ class EventSourceStream extends Transform {
       } else {
         event.data += `\n${value}`
       }
+      this.eventDataSize = eventDataSize
       return
     }
 
@@ -345,6 +371,7 @@ class EventSourceStream extends Transform {
     this.event.event = undefined
     this.event.id = undefined
     this.event.retry = undefined
+    this.eventDataSize = 0
   }
 
   hasPendingEvent () {

package/node_modules/undici/lib/web/eventsource/eventsource.js

@@ -10,6 +10,7 @@ const { isNetworkError } = require('../fetch/response')
 const { kEnumerableProperty } = require('../../core/util')
 const { environmentSettingsObject } = require('../fetch/util')
 const { createPotentialCORSRequest } = require('./util')
+const { getGlobalDispatcher } = require('../../global')
 
 let experimentalWarned = false
 
@@ -281,6 +282,7 @@ class EventSource extends EventTarget {
 
       const eventSourceStream = new EventSourceStream({
         eventSourceSettings: this.#state,
+        maxEventSize: this.#dispatcher.eventSourceOptions?.maxEventSize,
         push: (event) => {
           this.dispatchEvent(createFastMessageEvent(
             event.type,
@@ -465,7 +467,8 @@ webidl.converters.EventSourceInitDict = webidl.dictionaryConverter([
   },
   {
     key: 'dispatcher', // undici only
-    converter: webidl.converters.any
+    converter: webidl.converters.any,
+    defaultValue: () => getGlobalDispatcher()
   },
   {
     key: 'node', // undici only

package/node_modules/undici/lib/web/fetch/request.js

@@ -923,7 +923,7 @@ function makeRequest (init) {
     serviceWorkers: init.serviceWorkers ?? 'all',
     initiator: init.initiator ?? '',
     destination: init.destination ?? '',
-    priority: init.priority ?? null,
+    priority: init.priority ?? 'auto',
     origin: init.origin ?? 'client',
     policyContainer: init.policyContainer ?? 'client',
     referrer: init.referrer ?? 'client',
@@ -1129,8 +1129,7 @@ webidl.converters.RequestInit = webidl.dictionaryConverter([
   {
     key: 'priority',
     converter: webidl.converters.DOMString,
-    allowedValues: ['high', 'low', 'auto'],
-    defaultValue: () => 'auto'
+    allowedValues: ['high', 'low', 'auto']
   }
 ])
 

package/node_modules/undici/package.json

@@ -1,6 +1,6 @@
 {
   "name": "undici",
-  "version": "8.10.0",
+  "version": "8.10.1",
   "description": "An HTTP/1.1 client, written from scratch for Node.js",
   "homepage": "https://undici.nodejs.org",
   "bugs": {
@@ -97,7 +97,7 @@
     "test:websocket:autobahn": "node test/autobahn/client.js",
     "test:websocket:autobahn:report": "node test/autobahn/report.js",
     "test:wpt:setup": "node test/web-platform-tests/wpt-runner.mjs setup",
-    "test:wpt": "npm run test:wpt:setup && node test/web-platform-tests/wpt-runner.mjs run /fetch /mimesniff /websockets /serviceWorkers /eventsource",
+    "test:wpt": "npm run test:wpt:setup && node test/web-platform-tests/wpt-runner.mjs run /fetch /mimesniff /xhr /websockets /eventsource",
     "test:cache-tests": "node test/cache-interceptor/cache-tests.mjs --ci",
     "coverage": "npm run coverage:clean && cross-env NODE_V8_COVERAGE=./coverage/tmp npm run test:javascript && npm run coverage:report",
     "coverage:ci": "npm run coverage:clean && cross-env NODE_V8_COVERAGE=./coverage/tmp npm run test:javascript && npm run coverage:report:ci",
@@ -109,7 +109,7 @@
     "prepare": "husky && node ./scripts/platform-shell.js"
   },
   "devDependencies": {
-    "@fastify/busboy": "3.2.0",
+    "@fastify/busboy": "3.2.2",
     "@matteo.collina/tspl": "^0.2.0",
     "@metcoder95/https-pem": "^1.0.0",
     "@sinonjs/fake-timers": "^12.0.0",

package/node_modules/undici/types/client.d.ts

@@ -80,6 +80,8 @@ export declare namespace Client {
     maxResponseSize?: number;
     /** WebSocket-specific options */
     webSocket?: Client.WebSocketOptions;
+    /** EventSource-specific options */
+    eventSource?: Client.EventSourceOptions;
     /** Enables a family autodetection algorithm that loosely implements section 5 of RFC 8305. */
     autoSelectFamily?: boolean;
     /** The amount of time in milliseconds to wait for a connection attempt to finish before trying the next address when using the `autoSelectFamily` option. */
@@ -169,6 +171,14 @@ export declare namespace Client {
      */
     settings?: Omit<SessionOptions['settings'], 'enablePush' | 'maxConcurrentStreams' | 'enableConnectProtocol'>
   }
+  export interface EventSourceOptions {
+    /**
+     * Maximum allowed event size in bytes for EventSource messages.
+     * Set to 0 to disable the limit.
+     * @default buffer.kStringMaxLength
+     */
+    maxEventSize?: number;
+  }
 }
 
 export default Client

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "niceeval",
-  "version": "0.14.1-canary.51",
+  "version": "0.14.2",
   "description": "Agent-native eval tool — eval agents, services, functions, and coding-agent fixtures",
   "type": "module",
   "license": "MIT",
@@ -206,7 +206,7 @@
     "ws"
   ],
   "dependencies": {
-    "autoevals": "0.0.132",
+    "autoevals": "0.3.0",
     "effect": "4.0.0-rc.112",
     "jiti": "2.7.0",
     "openai": "^6.49.0",
@@ -214,7 +214,7 @@
     "yaml": "^2.9.0"
   },
   "devDependencies": {
-    "@ai-sdk/otel": "^1.0.9",
+    "@ai-sdk/otel": "^1.0.83",
     "@effect/platform-node": "4.0.0-rc.112",
     "@effect/platform-node-shared": "4.0.0-rc.112",
     "@opentelemetry/exporter-trace-otlp-http": "^0.219.0",
@@ -251,7 +251,7 @@
     "recharts": "3.10.1",
     "tailwindcss": "4.3.3",
     "typescript": "npm:@typescript/typescript6@^6.0.2",
-    "undici": "8.10.0",
+    "undici": "8.10.1",
     "vite": "^8.1.0",
     "vitest": "^4.1.11",
     "ws": "8.21.3"