Repository navigation
chore: version packages - #1487
Conversation
563da43 to
c2df740
Compare
8b700c1 to
a4d1b88
Compare
🟢 e2e web: 91 passedAll 91 tests in 41 files
Tested by e2e 0.23.0 · 1m 2s · web · run artifacts |
🟢 e2e web agent: 33 passed34 agent steps · 26 replayed from cache · 20 model calls · 87.8k tokens (94% cached) · $0.0036 All 33 tests in 5 files
Tested by e2e 0.23.0 · 1m 57s · web · run artifacts |
🟢 e2e android agent: 20 passed24 agent steps · 15 replayed from cache · 43 model calls · 243.3k tokens (87% cached) · $0.0138 All 20 tests in 1 file
Tested by e2e 0.23.0 · 5m 4s · android-emulator · run artifacts |
🟢 e2e ios: 1 flaky, 54 passed, 1 skipped
|
| Test | Time | |
|---|---|---|
| tests/flows-lists.e2e.ts · 1 flaky, 6 passed | 3m 8s | |
| 🟢 | infinite scroll reaches the target item | 25.0s |
| 🟢 | product catalog adds exactly two of the right variant | 9.9s |
| 🟢 | gestures: long-press, then swipe left | 9.2s |
| sticky chrome: accept below the fold, then continue (1 failed attempt first) | 9.9s | |
| 🟢 | async states: load, pull to refresh, claim through the toast | 12.0s |
| 🟢 | huge virtualized list reaches row 512 | 1m 50s |
| 🟢 | web view coupon form applies the code shown on the page | 11.6s |
| ⏭️ | tests/controls.e2e.ts · 23 passed, 1 skipped | 3m 15s |
| 🟢 | control inventory › a label query resolves a labeled text field | 13.5s |
| 🟢 | control inventory › a labeled field answers name, display value, value reads, fill, and clear | 16.4s |
| 🟢 | control inventory › a placeholder query and attribute reads find the field | 6.0s |
| 🟢 | control inventory › a secure field takes a fill and denies value reads | 9.9s |
| 🟢 | control inventory › press sends a character and Enter through the soft keyboard | 9.7s |
| 🟢 | control inventory › a field autofocuses on mount and focus() moves the focus | 12.0s |
| 🟢 | control inventory › toBeFocused follows the focus | 12.3s |
| 🟢 | control inventory › a switch checks, unchecks, and reports its state | 13.4s |
| 🟢 | control inventory › checkbox and radio roles answer getByRole, and a checkbox toggles by tap | 10.7s |
| 🟢 | control inventory › checkbox and radio roles check, uncheck, and report their state | 13.2s |
| 🟢 | control inventory › a radio group reports the checked option | 10.0s |
| 🟢 | control inventory › buttons report disabled, enabled, and a state that arrives later | 7.8s |
| 🟢 | control inventory › a display: none element is absent rather than hidden, its visible twin attached | 7.1s |
| 🟢 | control inventory › a header answers its accessible name | 4.1s |
| 🟢 | control inventory › a tab row answers count, all, texts, filters, positions, and switching | 6.1s |
| 🟢 | control inventory › a list answers counts, list-form text, reads, and boxes in order | 6.4s |
| ⏭️ | control inventory › a row contains its texts (skipped: React Native flattens a plain View out of the accessibility tree on both platforms: XCTest and the Android helper report a View with a testID as a leaf beside its children, so filter({ has }) and a query scoped to it match nothing; only Sc…) | |
| 🟢 | control inventory › a long press lands as one | 8.5s |
| 🟢 | control inventory › pointer verbs land at the point asked for | 9.7s |
| 🟢 | control inventory › an ambiguous locator fails at once | 5.6s |
| 🟢 | control inventory › screenshot, restart, and clearState reopen the app on its home list | 11.0s |
| 🟢 | refusals › verbs a device never declares are refused before the engine | 3ms |
| 🟢 | refusals › keys the soft keyboard cannot send are refused by the engine | 565ms |
| 🟢 | refusals › a missing locator fails as not found once its timeout passes | 1.1s |
| 🟢 | tests/device-fixture.e2e.ts · 11 passed | 1m 47s |
| 🟢 | device fixture › setAppearance flips the color scheme the app reads | 13.8s |
| 🟢 | device fixture › setOrientation rotates the window the app measures | 8.7s |
| 🟢 | device fixture › the clipboard reads back what was written | 7.1s |
| 🟢 | device fixture › home leaves the app, openApp brings it back where it was | 8.0s |
| 🟢 | device fixture › closeApp then openApp relaunches on the home list | 8.9s |
| 🟢 | device fixture › closeApp leaves the next commands on the same device | 14.6s |
| 🟢 | device fixture › foregroundApp after closeApp reports the closed session as APP_NOT_OPEN | 6.4s |
| 🟢 | device fixture › device.back pops the scenario | 7.1s |
| 🟢 | device fixture › unstable_raw reaches the test session through agent-device | 5.6s |
| 🟢 | device fixture › setLocation and clearLocation change what the app reads | 14.2s |
| 🟢 | device fixture › enrollBiometrics and setBiometrics answer a biometric prompt | 12.5s |
| 🟢 | tests/device.e2e.ts · 3 passed | 1m 7s |
| 🟢 | sequential onboarding echoes the exact values | 41.5s |
| 🟢 | onboarding rejects a malformed email before advancing | 13.9s |
| 🟢 | dismissing the keyboard over choice controls leaves them as they were | 11.7s |
| 🟢 | tests/flows.e2e.ts · 7 passed | 1m 9s |
| 🟢 | text inputs unlock the submit button | 13.3s |
| 🟢 | filled text inputs report their values | 10.2s |
| 🟢 | modal flow confirms through the native alert | 9.9s |
| 🟢 | bottom tabs act inside the last tab | 7.3s |
| 🟢 | error recovery retries, then confirms the delete | 9.2s |
| 🟢 | choice controls place the exact order | 7.4s |
| 🟢 | debounced search selects the target once results arrive | 11.3s |
| 🟢 | tests/login-form.e2e.ts · 3 passed | 48.0s |
| 🟢 | login form › rejects a malformed email | 13.7s |
| 🟢 | login form › rejects an unknown account | 13.9s |
| 🟢 | login form › signs in and logs out | 20.4s |
| 🟢 | tests/smoke.e2e.ts · 1 passed | 7.5s |
| 🟢 | home lists the scenarios and opens one | 7.5s |
Tested by e2e 0.23.0 · 6m 17s · ios-simulator · run artifacts
🟢 e2e android: 57 passed, 1 skippedAll 58 tests in 7 files
Tested by e2e 0.23.0 · 8m 16s · android-emulator · run artifacts |
🟢 e2e ios agent: 21 passed, 1 skipped25 agent steps · 12 replayed from cache · 68 model calls · 449.6k tokens (83% cached) · $0.0291 All 22 tests in 1 file
Tested by e2e 0.23.0 · 9m 15s · ios-simulator · run artifacts |
This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.
Releases
e2e@0.23.0
Minor Changes
d2556bcThanks @okwasniewski! - Breaking:inputValue()on a node with no value, such as a heading or a button, andisChecked()on a node with no checked state throwACTION_FAILED, as in Playwright, instead of returning""andfalse. A field with no value still reads"", and a checkable role with no state reported still readsfalse.uncheck()on an ARIArole="radio"now clicks it, as in Playwright; a native radio still refuses.Patch Changes
#1524
cba5e0eThanks @okwasniewski! - The built-in agent can resize the viewport with the newresize_viewporttool, soagent.actande2e explorecan check layouts at desktop, tablet, and phone widths. The tool is offered when the target's engine implements the new optionalsession.resizehook. Custom executors callctx.actions.resize({ width, height }), and the replay cache records and replays it.#1479
d15a98eThanks @okwasniewski! -app.baseUrlno longer ends in a slash, so${app.baseUrl}/api/usersrequests/api/usersinstead of//api/users. Withtestfrom@e2e-dev/web,app.baseUrlis typedstring, since aweb()target always declaresapp.url. A test that relied on the trailing slash, such as${app.baseUrl}api/usersornew URL('users', app.baseUrl)on a URL with a path, adds the slash itself.#1491
108bc60Thanks @okwasniewski! - Bump @agentclientprotocol/sdk to 2.2.0, @ai-sdk/provider to 4.0.35, and oxc to 0.156.0.#1516
0410fc9Thanks @okwasniewski! - Bump agent-device to 0.21.36, @agentclientprotocol/sdk to 2.4.0, @vitest/expect to 5.1.0, and oxc to 0.157.0.#1573
3d5f4eaThanks @okwasniewski! - Bump @ai-sdk/provider to 4.0.36.#1451
b4dd4bdThanks @amoscicki! - The run summary'sCacheline counts the steps of passing and flaky tests whose recording was not saved (step.cache.writeofunconfirmed,no-change, ornot-writtenin the attempt that passed), such as4 missed · 1 not saved.#1493
91bd614Thanks @okwasniewski! - Adde2e doctor: checks thate2e runcan start without starting anything. It covers the Node.js runtime, the config and its targets, each engine's own checks, and the model each agent resolved to.--jsonprints thedoctor-1report for a host,--check-modelcalls each model once, and the command exits 2 when a check failed. Engines can report their checks through the new optionaldoctorhook.#1575
7d7d0e1Thanks @kacperkapusciak! -e2e initande2e --helpshow the new e2e logo, its six rows of bars drawn in full blocks, one row per bar. Ine2e initthe logo shades in from ░ to █, top rows first, in about half a second. The logo needs a terminal at least 52 columns wide, and prints without motion on one shorter than 7 rows.#1478
2208056Thanks @okwasniewski! -CONFIG_NOT_FOUNDfor an explicit config path (--config <path>, the MCPopen_sessionconfiginput, oropenHost({ config: '<path>' })) now names the absolute path it checked and, for a relative path, the working directory it resolved against.#1572
0c04696Thanks @okwasniewski! - Documentation links in CLI help,e2e initoutput, and the MCP server listing now point to https://e2e.com/docs.#1232
3f43eddThanks @okwasniewski! - Add the engine protocol:connectEngine,serveEngine, and transports ine2e/enginerun an engine in another process, on another machine, or in another language over LSP-style JSON-RPC. Every engine built withdefineEnginenow runs through the protocol in process, so the harness drives every engine the same way.Engine.disposenow runs only onceinitwas called, even wheninitfailed. A worker that never initialized no longer disposes its engine.#1522
73734c6Thanks @okwasniewski! -e2e exploredigs deeper once the main flows are covered: it returns to them with edge-case inputs, combined filters and sorting, reloads, and other accounts, finishes early only when nothing new is reachable, and checks other screens that offer a control it found broken.#1523
f4be1c4Thanks @okwasniewski! -e2e exploreruns longer:--max-stepsgoes up to 100 and--timeoutup to 7200000 (2 hours).#1480
723c1afThanks @okwasniewski! - A failed test's block in the terminal and insummary.mdquotes the first 3 errors the app logged (console errors, uncaught exceptions, failed requests, 5xx responses), so an app error behind a locator failure shows without openingtrace.md.#1499
1d671f6Thanks @okwasniewski! - The markdown report footer (summary.md, the@e2e-dev/githubpull request comment and job summary) now readsTested by e2e <version> · ..., withe2elinking to https://e2e.com.#1464
57f9ed9Thanks @okwasniewski! -locator.waitFor,expecton a locator, andtoHaveScreenshotgive their last read extra time past the timeout, up to their slowest read so far and within the test timeout, so a screen that shows up during a slow last read still passes.#1576
33a1541Thanks @kdy1! - Start faster by loading heavy dependencies when a command or session needs them. The CLI no longer loads the MCP server, the prompt library, or the schema library untilmcp,init,login,logout,models,explore,upgrade,doctor, or--watchruns, soe2e --versionand--helptake about 25% less time.@e2e-dev/webimports Playwright when a browser is first launched or looked up, so a process that only reads the config skips it:e2e liston a web project takes about 44% less time, and importing the package half as long. A worker loads Playwright for its first browser instead of before its first test, so that test's reported duration grows by about 0.14 s while the run takes no longer.e2e/runnerno longer loads the explorer when imported.@e2e-dev/acploads the ACP and MCP SDKs when its first session starts, and@e2e-dev/limrunloads@limrun/apiwhen its first device is leased.#1434
50d8063Thanks @DeryFerd! - Later agent steps now see a fully replayed step's summary labeledrecorded in an earlier runinstead ofobserved.#1482
d94593bThanks @okwasniewski! - Thelistreporter prints each test as it finishes when stdout is not a terminal (piped, redirected, most CI), instead of only once its whole file finishes, and ends every run with aFailedrow per failed test (file:line, the error's first line, the trace page when one was written), soe2e run | tailstill names what broke.#1416
79e2e13Thanks @DeryFerd! - Filtered-locator failures can list how many candidates each filter step left, when the post-failure diagnostic pass can read the chain, so a step that narrowed too much or too little names itself. The message says the counts were read then.#1520
50fa7afThanks @okwasniewski! -e2e mcphas arun_teststool: it runs a test file, one test (tests/checkout.e2e.ts:12), or the whole suite ase2e rundoes and returns what the run printed, so a coding agent can run tests without a shell.run_tests {files: ["tests/checkout.e2e.ts:12"]}#1498
4b56473Thanks @21prnv! - A replay no longer fails its end check on a date written with an ordinal suffix, such asOctober 26thor3rd October, when the day differs from the recording.#1571
4c44e44Thanks @kdy1! - The agent compares screenshots with a native SHA-256 instead of a byte-by-byte loop in script, so a settle poll or screen update that carries pixels no longer spends about 8 ms of the event loop per megabyte of image.#1493
91bd614Thanks @okwasniewski! - A model call without a response schema that the provider ends on anerrorfinish now fails asMODEL_PROVIDER_FAILEDinstead ofMODEL_OUTPUT_INVALID.#1500
7b1b8b2Thanks @bheemreddy-samsara! - Add coordinates toagent.act'sonActionevent: narrow onevent.actionto readpointfortapAtandhoverAt, orfromandtofordragAt, replay included. Points are in the same space asscreen.tapAtandboundingBox(), rounded and clamped to the viewport.ActionEventis now a union keyed byaction, typedActionNameortool:<name>instead ofstring. NewActionNameandActionDetailstypes.#1474
d10655fThanks @samanyudas! -e2e runwithout thelistreporter prints each run error on stderr asCODE: message. Before,--reporter junitormarkdownwith a test file that failed to collect, a config that failed to load, or no model for the agent exited 2 and printed nothing.#1481
f078364Thanks @okwasniewski! - Thelistsummary names the first 10app.screenshotfiles in aScreenshotsrow, then counts the rest. The MCPscreenshottool takes an optionalpath(a.png, relative to the project root): it saves the masked full-resolution screenshot there and returns the absolute path.#1470
b8485baThanks @samanyudas! -run.summary.executednow counts a serial member whose body ran, the same way it counts an ordinary test. Before, it left out a member that calledtest.skip()in its body, and a member that ran in one group attempt before a retry skipped it behind a failed predecessor. The run's telemetry cache tiers count them the same way.#1483
0b84d35Thanks @okwasniewski! -open_sessionnow names the app command it started: its executable, the URL it answers on, the port a free-porturlwas assigned, and theenvandinheritEnvnames it got. It prints noargsand noenvvalues. It also says the config reads the MCP server's environment, not the agent's shell.e2e runprints the URL on the app command's setup line, and theappsetup event carries it asurl.#1546
a8ed7f7Thanks @okwasniewski! -locator.setInputFiles([])clears a file input and firesinputandchange, as in Playwright. It used to throwINVALID_ARGUMENT. A blank path is still refused.#1567
65ebc7dThanks @NathanWalker! - The bundled e2e skill now names NativeScript'stestIDamong the idsgetByTestIdreads on a device.#1538
c7cbcefThanks @kacperkapusciak! - The bundled skills declare what they need to run (nodeandnpx, and the optionalAI_GATEWAY_API_KEYfor the e2e skill) inmetadata.openclaw, the frontmatter ClawHub reads. All three are now on ClawHub under@e2e.#1547
4dce5a7Thanks @okwasniewski! -tap({ position })measures the offset from inside the node's border, as Playwright'sclick({ position })does. Before, a bordered element shifted the tap by its border width. Engines can report that point as the new optionalSemanticNode.origin.#1539
732ea8aThanks @okwasniewski! - Addtest.failto mark a test that is expected to fail:test.fail(title, fn),test.fail(condition, reason)inside a body, or thefailtest option. A marked test that fails with a test failure passes, and the report keeps the reason and the original error underexpectedFailure. A marked test that passes fails withUNEXPECTED_PASS, so a fixed gap gets noticed. Infrastructure errors such asENGINE_FAILUREand blocked agent verdicts are not covered.#1505
bdf769bThanks @okwasniewski! - A fixture call that times out names what it waited for:browser.waitForResponse('foo.css')fails withoperation timed out in browser.waitForResponse "foo.css". Steps forwaitForURL,waitForResponse,route, andunrouteshow a string pattern in quotes and a RegExp as its literal.#1554
d2953bbThanks @okwasniewski! - A call'stimeout: 0now means no deadline of its own, as in Playwright: locator actions,waitFor, matchers,toHaveScreenshot,expect.poll, contributed fixture calls, andbrowser.goto,waitForResponse, andwaitForDownloadwait until the test timeout (cleanupTimeoutinafterEach) instead of failing at once or being refused. A negative,NaN, or infinitetimeoutisINVALID_ARGUMENTon every call, locator actions included (INVALID_CONFIGonexpect.poll, as before).expect.pollwithtimeout: 0outside a test isINVALID_CONFIG. Engine authors getstepTimeoutfrome2e/engineto give a contributed call'stimeoutoption the same rules.#1471
80c2fdaThanks @samanyudas! - A string a test or the config writes toprocess.stdoutorprocess.stderrwith an encoding, such ashex,base64, orlatin1, now prints decoded from a worker, as Node prints it. The runner process decoded onlyhex,base64, andbase64url, and now decodes every encoding Node accepts too. Bytes that end inside a UTF-8 character no longer print after the string written next.@e2e-dev/acp@0.1.1
Patch Changes
#1491
108bc60Thanks @okwasniewski! - Bump @agentclientprotocol/sdk to 2.2.0, @ai-sdk/provider to 4.0.35, and oxc to 0.156.0.#1516
0410fc9Thanks @okwasniewski! - Bump agent-device to 0.21.36, @agentclientprotocol/sdk to 2.4.0, @vitest/expect to 5.1.0, and oxc to 0.157.0.#1576
33a1541Thanks @kdy1! - Start faster by loading heavy dependencies when a command or session needs them. The CLI no longer loads the MCP server, the prompt library, or the schema library untilmcp,init,login,logout,models,explore,upgrade,doctor, or--watchruns, soe2e --versionand--helptake about 25% less time.@e2e-dev/webimports Playwright when a browser is first launched or looked up, so a process that only reads the config skips it:e2e liston a web project takes about 44% less time, and importing the package half as long. A worker loads Playwright for its first browser instead of before its first test, so that test's reported duration grows by about 0.14 s while the run takes no longer.e2e/runnerno longer loads the explorer when imported.@e2e-dev/acploads the ACP and MCP SDKs when its first session starts, and@e2e-dev/limrunloads@limrun/apiwhen its first device is leased.@e2e-dev/decision@0.3.2
Patch Changes
db70c36Thanks @okwasniewski! - On a page with 255 or more elements, the decision executor no longer asks the model a target question with 256 choices, one over the TypeSafe limit of 255. Each question now lists at most 254 targets plusnone. Elements past the cap are left out as before, and the model can scroll to reach them.@e2e-dev/eas@0.3.3
Patch Changes
#1518
1293b78Thanks @kacperkapusciak! - Errors and logs now name Expo instead of EAS, following Expo's service names: an error from Expo's API starts withExpo:instead ofEAS:, and a queued session logsstill queued at Expo after two minutes.#1517
c59156eThanks @kacperkapusciak! -easSimulators()is nowexpoSimulators(), matching the service's name, Expo Simulators.easSimulators()andEasSimulatorsOptionsstill work as deprecated aliases ofexpoSimulators()andExpoSimulatorsOptions. The provider now logs asexpo-simulatorsinstead ofeas-simulators, and its config errors start withexpoSimulators.@e2e-dev/limrun@0.2.4
Patch Changes
#1516
0410fc9Thanks @okwasniewski! - Bump agent-device to 0.21.36, @agentclientprotocol/sdk to 2.4.0, @vitest/expect to 5.1.0, and oxc to 0.157.0.#1576
33a1541Thanks @kdy1! - Start faster by loading heavy dependencies when a command or session needs them. The CLI no longer loads the MCP server, the prompt library, or the schema library untilmcp,init,login,logout,models,explore,upgrade,doctor, or--watchruns, soe2e --versionand--helptake about 25% less time.@e2e-dev/webimports Playwright when a browser is first launched or looked up, so a process that only reads the config skips it:e2e liston a web project takes about 44% less time, and importing the package half as long. A worker loads Playwright for its first browser instead of before its first test, so that test's reported duration grows by about 0.14 s while the run takes no longer.e2e/runnerno longer loads the explorer when imported.@e2e-dev/acploads the ACP and MCP SDKs when its first session starts, and@e2e-dev/limrunloads@limrun/apiwhen its first device is leased.#1513
383c0bcThanks @thymikee! - Device-cloud leases are allocated and renewed under a named agent-device session, so a newer agent-device that refuses a provider lease to an unnamed client keeps working with BrowserStack, TestMu AI, and Limrun.@e2e-dev/mobile@0.12.6
Patch Changes
#1516
0410fc9Thanks @okwasniewski! - Bump agent-device to 0.21.36, @agentclientprotocol/sdk to 2.4.0, @vitest/expect to 5.1.0, and oxc to 0.157.0.#1493
91bd614Thanks @okwasniewski! -e2e doctorchecks a mobile target: agent-device, the platform toolchain, each configured device, an installedbundleId, and theappPathbuild. Requires e2e 0.22.1 or newer.mobile({ device })now rejects a blank device name at config load instead of failing the run's boot.#1464
57f9ed9Thanks @okwasniewski! - Snapshots and screenshots now run within the timeout of the call that takes them (an action'stimeout, anexpecttimeout), so a device that stops answering fails the step instead of holding it until the test timeout. Requirese2e0.22.0 or later.#1559
8f425b1Thanks @shamilovtim! - Share a pending screen-recording stop with cleanup after a caller cancels its wait, so finalization is not repeated and a confirmed stop clears the recording state.#1513
383c0bcThanks @thymikee! - Device-cloud leases are allocated and renewed under a named agent-device session, so a newer agent-device that refuses a provider lease to an unnamed client keeps working with BrowserStack, TestMu AI, and Limrun.@e2e-dev/testmu@0.0.4
Patch Changes
0410fc9Thanks @okwasniewski! - Bump agent-device to 0.21.36, @agentclientprotocol/sdk to 2.4.0, @vitest/expect to 5.1.0, and oxc to 0.157.0.@e2e-dev/web@0.15.5
Patch Changes
#1479
d15a98eThanks @okwasniewski! -app.baseUrlno longer ends in a slash, so${app.baseUrl}/api/usersrequests/api/usersinstead of//api/users. Withtestfrom@e2e-dev/web,app.baseUrlis typedstring, since aweb()target always declaresapp.url. A test that relied on the trailing slash, such as${app.baseUrl}api/usersornew URL('users', app.baseUrl)on a URL with a path, adds the slash itself.#1553
9267248Thanks @okwasniewski! -browser.evaluatereturns its result when the evaluated code deletes or replaces page globals such asMaporObject, and on pages that deleteArrayorJSON. When a page replaces a built-in the engine's in-page code needs, such asMap, later evaluates and locators fail withEVALUATE_FAILEDinstead ofENGINE_FAILURE.browser.evaluatenow returns plain JSON only, orundefinedwhen the code returns nothing. A class instance such as aDOMRector an element fails withINVALID_ARGUMENTinstead of arriving as a partial object; returnrect.toJSON()or the fields you need.#1549
08b102aThanks @okwasniewski! - An element inside a hidden<iframe>now reads as hidden, cross-origin frames included:isVisible()returns false,toBeHidden()passes,getByRoleand{ visible: true }skip it, and actions wait until the frame shows.#1576
33a1541Thanks @kdy1! - Start faster by loading heavy dependencies when a command or session needs them. The CLI no longer loads the MCP server, the prompt library, or the schema library untilmcp,init,login,logout,models,explore,upgrade,doctor, or--watchruns, soe2e --versionand--helptake about 25% less time.@e2e-dev/webimports Playwright when a browser is first launched or looked up, so a process that only reads the config skips it:e2e liston a web project takes about 44% less time, and importing the package half as long. A worker loads Playwright for its first browser instead of before its first test, so that test's reported duration grows by about 0.14 s while the run takes no longer.e2e/runnerno longer loads the explorer when imported.@e2e-dev/acploads the ACP and MCP SDKs when its first session starts, and@e2e-dev/limrunloads@limrun/apiwhen its first device is leased.#1571
4c44e44Thanks @kdy1! - An observation no longer creates and releases an element handle for every node it lists. The reader keeps its elements in one in-page array per document, and an action makes the handle for its node the first time it reaches for it. Reading a 3000-node page takes about 80 ms instead of about 360 ms, and the engine stops blocking its event loop for around 100 ms releasing the previous observation's handles. References still name the element the walk read.#1552
d2556bcThanks @okwasniewski! - Breaking:inputValue()on a node with no value, such as a heading or a button, andisChecked()on a node with no checked state throwACTION_FAILED, as in Playwright, instead of returning""andfalse. A field with no value still reads"", and a checkable role with no state reported still readsfalse.uncheck()on an ARIArole="radio"now clicks it, as in Playwright; a native radio still refuses.#1544
8f17ad3Thanks @okwasniewski! - TwotoHaveScreenshotcalls (or other screen reads) running at once on one page no longer fail withLOCATOR_NOT_FOUND: node became stale. The engine now takes concurrent observations in turn.#1548
40826e4Thanks @okwasniewski! -route.fulfilltakes aBufferorUint8Arraybody and sends the bytes unchanged, and sets every cookie of aset-cookieheader that joins them with\n. Aroute.continueheader set toundefinedis dropped, also on redirects.WebResponseaddsbody(), which returns the bytes, soroute.fetch()andwaitForResponsecan hand binary bodies on.#1551
e99a6faThanks @okwasniewski! - In Chromium, screenshots,toHaveScreenshot, and observations no longer wait on an iframe whose first navigation has not committed, such as one whose request a route handler holds. Before, they timed out with no screenshot.#1547
4dce5a7Thanks @okwasniewski! -tap({ position })measures the offset from inside the node's border, as Playwright'sclick({ position })does. Before, a bordered element shifted the tap by its border width. Engines can report that point as the new optionalSemanticNode.origin.#1505
bdf769bThanks @okwasniewski! - A fixture call that times out names what it waited for:browser.waitForResponse('foo.css')fails withoperation timed out in browser.waitForResponse "foo.css". Steps forwaitForURL,waitForResponse,route, andunrouteshow a string pattern in quotes and a RegExp as its literal.#1554
d2953bbThanks @okwasniewski! - A call'stimeout: 0now means no deadline of its own, as in Playwright: locator actions,waitFor, matchers,toHaveScreenshot,expect.poll, contributed fixture calls, andbrowser.goto,waitForResponse, andwaitForDownloadwait until the test timeout (cleanupTimeoutinafterEach) instead of failing at once or being refused. A negative,NaN, or infinitetimeoutisINVALID_ARGUMENTon every call, locator actions included (INVALID_CONFIGonexpect.poll, as before).expect.pollwithtimeout: 0outside a test isINVALID_CONFIG. Engine authors getstepTimeoutfrome2e/engineto give a contributed call'stimeoutoption the same rules.#1493
91bd614Thanks @okwasniewski! -e2e doctorreports whether the browser a run launches is installed, or that the first run will download it. Requires e2e 0.22.1 or newer.#1550
4e7e8b8Thanks @okwasniewski! - Two failures caused by the test or the page no longer report asENGINE_FAILURE:browser.gotowhose request a route aborted fails withACTION_FAILED. When the route aborted it because the handler threw, such as aroute.fulfillwith an invalid header, thegotofails with the handler's error.app.screenshot()taken while the page navigates captures the document showing, masked.A closed page or browser is still
ENGINE_FAILURE.#1524
cba5e0eThanks @okwasniewski! - The web engine implementssession.resize, so the agent'sresize_viewporttool is available on web targets. Requires e2e 0.22.1 or newer.