Otherwise, we might split the `beforeAll`-grouped test group into
`workers` parts instead of `shard.total` parts as the user would expect.
Fixes#33077.
Closes https://github.com/microsoft/playwright/issues/32960
If the socket is reused, the connect and DNS timings are set to -1,
because that timing doesn't apply to the current request. The time
between request start and the socket being free is counted as `blocked`.
The diff is now shown inline in the errors list.
There are 2 possible failures of toHaveScreenshot
* Previous and actual snapshot mismatch. In this case html report will
show diff between Actual/Previous and have Expected as a separate
screenshot.
* Actual/Previous are equal but they differ from the expected. In this
case html report only contains Actual/Expected images and the diff.
Reference: https://github.com/microsoft/playwright/issues/32341
<img width="1039" alt="image"
src="https://github.com/user-attachments/assets/b458f986-cc25-4721-862c-0cc2c1b01a42">
This reapplies what we reverted in
https://github.com/microsoft/playwright/pull/32989.
Max and me debugged this, and found that the test failures come from
SOCKS proxy now preferring IPv6 over IPv4. We've updated the tests and
made sure that this doesn't mask any breaking change.
I'm enabling CQ1 to make sure we don't oversee any other CI failures.
Fixes a bug discovered in
https://github.com/microsoft/playwright/pull/32647. When using http
proxy, the `connect` event isn't emitted so we don't populate
`tcpConnectionAt`. The updated version of `https-proxy-agent` emits a
`proxyConnect` as a replacement, so this PR updates and listens to that
event.
For socks proxies, the `on("socket")` event is emitted once the SOCKS
connection is established, which is the equivalent of having a TCP
connection available.
---------
Signed-off-by: Simon Knott <info@simonknott.de>
Co-authored-by: Max Schmitt <max@schmitt.mx>
This includes all actions that perform locator handler check.
Note this makes it impossible to interact with the page while a main
frame navigation is ongoing. This was already the case for Chromium, but
now WebKit and Firefox align with it.
Setting `PLAYWRIGHT_SKIP_NAVIGATION_CHECK` environment variable disables
this behavior.
When the list of highlighted elements changes over time, we should
update the elements marked as `__playwright_target__` in the snapshot.
A good example is an `expect(locator).toHaveText([...])` where the list
of elements changes from 4 items to 3 after clicking a "Delete" button.
Closes https://github.com/microsoft/playwright/issues/32853
Vite turns the shorthand fragment notation `<></>` into `import {
Fragment } from "react"; <Fragment></Fragment>`. On the Node.js side of
things, this `react` import resolves to our mock version of React, which
currently mocks `Fragment` as `{}`. Currently, we pass that straight to
`React.createElement`, which throws an error.
The fix is to make our `Fragment` mock detectable with a tag, and when
we render it replace it with the real `__pwReact.Fragment`.
This is fixing a case where the test failed with strict mode violation,
but all the matched elements are not highlighted in the trace.
For example, all the buttons will be highlighted when the following line
fails due to strict mode violation:
```ts
await page.locator('button').click();
```
To achieve this, we mark elements during `querySelector` phase instead
of inside `onBeforeInputAction`. This allows us to only mark from inside
the `InjectedScript` and remove the other way of marking from inside the
`Snapshotter`.
- Renamed to `page.requestGC`.
- Added a useful snippet to the docs.
References #32278.
---------
Signed-off-by: Dmitry Gozman <dgozman@gmail.com>
Co-authored-by: Max Schmitt <max@schmitt.mx>
Closes https://github.com/microsoft/playwright/issues/32653.
Adds some test coverage to ensure we respect minimal mode for API
Requests in HAR tracing.
| omit setting | result |
| - | - |
| `omitCookies` | added test for it |
| `omitTiming` | already covered |
| `omitSecurityDetails` | not recorded yet |
| `omitServerIP` | we don't record it yet, so no action here. gonna open
a separate issue |
| `omitPages` | not relevant to API requests |
| `omitSizes` | added test for it |
| `omitScripts` | not relevant to API requests |
Fixes https://github.com/microsoft/playwright/issues/30160
### Description:
This pull request introduces the ability to specify custom locations for
test steps in Playwright. By enabling the provision of arbitrary
locations to the test.step method, it resolves the limitation where
helper methods obfuscate the original call site, providing more accurate
and meaningful location data in test reports.
### Motivation:
To enhance the utility and clarity of test reports in Playwright.
Specifically, it addresses the need to trace test steps back to their
precise location in the code, which is especially important when steps
are abstracted in helper functions. This feature is crucial for
maintaining accurate documentation and facilitating debugging processes.
### Changes:
Added functionality to pass a custom location object to test.step.
### Expected Outcome:
This PR is expected to significantly improve the precision and
usefulness of diagnostic data in test reports by allowing specific
locations within helper functions to be accurately documented. It
facilitates better tracking of test executions and simplifies the
debugging process, making it easier for developers to understand and
address issues within complex tests.
### References:
Closes https://github.com/microsoft/playwright/issues/30160 -
"[Feature]: allow to pass arbitrary location to test.step"
**Code Check**
I conducted tests on this new feature by integrating it into some
existing test codes, and it worked well. I will attach the code used for
testing and a screenshot showing the successful outcome.
<details>
<summary>toggle dropdown</summary>
<div markdown="1">
```
import type { Location } from '../../../packages/playwright/types/testReporter'
...
test('should respect the back button', async ({ page }) => {
await page.locator('.todo-list li .toggle').nth(1).check();
await checkNumberOfCompletedTodosInLocalStorage(page, 1);
...
await test.step('Showing active items', async () => {
await page.getByRole('link', { name: 'Active' }).click();
}, {location});
```
<img width="1109" alt="image"
src="https://github.com/user-attachments/assets/359feafa-0949-4c71-9426-46debef21bdd">
</div>
</details>
Related to https://github.com/microsoft/playwright/issues/19621
Adds some instrumentation to collect timings for `APIRequestContext`
requests and adds them to the HAR trace. Doesn't yet expose them via an
API, but makes our `Duration` field in the trace viewer show a nice
duration:
<img width="1392" alt="Screenshot 2024-09-14 at 11 46 04"
src="https://github.com/user-attachments/assets/8020382d-9494-4634-9cfd-22b6f4a1d770">
I'm gonna add it to our API in a separate PR.
---------
Signed-off-by: Simon Knott <info@simonknott.de>
Co-authored-by: Dmitry Gozman <dgozman@gmail.com>
Closes https://github.com/microsoft/playwright/issues/32578.
Adds a buffer mode that can be toggled by pressing <kbd>b</kbd>. When
engaged, changed test files are collected and shown on screen. The test
run is then kicked off by pressing <kbd>Enter</kbd>.
This changes the signal to start a test run from <kbd>Cmd+s</kbd> to a
<kbd>Enter</kbd> press in the test terminal. It should help users with
auto-save and make it easier to run on long-running tests. It feels very
similar to running `npx playwright test`, but without having to write a
filter.
https://github.com/user-attachments/assets/71e16139-9427-4e90-b523-8d218f09ed9d
Closes https://github.com/microsoft/playwright/issues/32632. A side
effect of Remix's hydration implementation is that it throws away the
entire DOM. This is broadly discussed in
https://github.com/remix-run/remix/issues/4822 - there might be a fix in
coming React versions, but who knows.
Besides breaking browser extensions, this also deletes our toolbar!
This PR fixes it by periodically checking in on `x-pw-glass`, and
remounting it if it was unmounted. Hacky but effective!
Changes `expect.extend` behaviour so that it doesn't mutate the global
instance and behaves closer to what users expect. This is formally a
breaking change, and I had to remove a test that asserts the breaking
behaviour.
TODO:
- [x] decide wether this is a separate method or a flag for
`expect.extend`
- [x] figure out if we need to change docs
Based on the expectations the tests that are expected to timeout or fail
will be skipped to save resources. The expectations can be manually
updated when corresponding feature is fixed.
What was happening?
- When we use CT, we go over the test files, look at the imports using
`tsxTransform.ts` and store them inside a map, these we feed into the
import registry which we build using Vite and have access inside the
browser
- In case of an inline component in the same file as where the test file
is, this is not happening.
- jsx-runtime via babel kicks in, transforms every JSX component in
something like that:
```
{
__pw_type: 'jsx',
type: [Function: MyInlineComponent],
props: { value: 'Max' },
key: undefined
}
```
this then gets passed into `wrapObject` which maps any function from the
Node.js side into expose function calls so they work inside the browser.
The assumption for `wrapObject` was to do it mostly for callbacks. So it
does for `type` - which is actually our component. We then pass this to
the React render function, which calls back the exposed function but we
never return anything, so it mounts `undefined`.
---
While there have been experiments from certain vendors to get the
'client only' code inside a server side file, we should throw for now to
not confuse users. We might revisit this in the future since Babel / TSX
doesn't support it outside of the box.
Fixes https://github.com/microsoft/playwright/issues/32167
We now hopefully align with `moduleResolution: bundler` tsconfig option,
allowing directory imports in every scenario, and allowing proper module
imports when not going through the type mapping.
This regressed in #32078. Fixes#32480, fixes#31811.
In this test, the trace recording goes super fast. Sometimes, this means
that the recording is finished before the screen recorder got a chance
to take a screenshot. If that happens, the tests fail because we never
show a screenshot.
This PR fixes the flakiness by delaying the trace recording so that
there's always a screenshot taken.
Closes https://github.com/microsoft/playwright/issues/23964.
Trace snapshots are a best-effort snapshots of the browser DOM, but we
can't guarantee them to be exactly what the browser showed. One example
of this is `canvas` elements, where you just can't see their contents.
That makes snapshots useful, but not perfect.
For those cases where the snapshot doesn't show everything, this PR
introduces a new setting to show a screenshot instead. You won't be able
to scroll or inspect the DOM or select a locator anymore. But if the
snapshot was missing something, or displaying something wrong, you can
now check the screenshot instead.
Instead of plumbing it through a custom unspecified config field, make
it a part of plugin interface.
Additionally, use task runner for starting/stopping dev server.
This pull request introduces initial support for the WebDriver BiDi
protocol in Playwright. The primary goal of this PR is not to fully
implement BiDi but to experiment with the current state of the
specification and its implementation. We aim to identify the biggest
gaps and challenges that need to be addressed before considering BiDi as
the main protocol for Playwright.
