Previously, we always formed groups consisting of a single test.
Now, we group tests that share `beforeAll`/`afterAll` hooks into
`config.workers` equally-sized groups.
This patch:
- starts using directory of `package.json` to resolve default
output directory path
- starts using either `package.json` directory or configuration
directory to resolve all relative paths
References #12970
Resolves#11318.
* Adds `TestConfig.attachments` public API. (We opted to not implement an analog to the async `TestInfo.attach(…)` API.)
* Adds `TestConfig.attachments` to common reporters.
* Dogfoods some git and CI-info inference to generate useful atttachments
* Updates HTML Reporter to include a side bar to present a pre-defined set of attachments (a.k.a git/commit context sidebar)
Here's what it looks like:
<img width="1738" alt="Screen Shot 2022-03-21 at 3 23 28 PM" src="https://user-images.githubusercontent.com/11915034/159373291-8b937d30-fba3-472a-853a-766018f6b3e2.png">
See `tests/playwright-test/reporter-html.spec.ts` for an example of usage (for dogfood-ing only). In the future, if this becomes user-facing, there the Global Setup bit would likely become unnecessary (as would interaction with attachments array); there would likely just be a nice top-level config and/or CLI flag to enable collecting of info.
This patch aligns the strategies that are used to generate new
screnshot expectations and to compare screenshot expectations against
baseline.
With this patch, `toHaveScreenshot` will:
- when generating a new expectation: will wait for 2 consecutive
screenshots to match and accept the last one as expectation.
- when given an expectation:
* will compare first screenshot against expectation. If matches,
resolve successfully
* if first screenshot doesn't match, then wait for 2 consecutive
screenshots to match and then compare last screenshot with the
expectation.
An example of a new detailed call log:
```
1) a.spec.ts:3:1 › should work ===================================================================
Error: Screenshot comparison failed:
20000 pixels (ratio 0.03 of all image pixels) are different
Call log:
- expect.toHaveScreenshot with timeout 5000ms
- verifying given screenshot expectation
- fast-path: checking first screenshot to match expectation
- taking page screenshot
- disabled all CSS animations
- waiting for fonts to load...
- fonts in all frames are loaded
- fast-path failed: first screenshot did not match expectation - 20000 pixels (ratio 0.03 of all image pixels) are different
- waiting for 2 consecutive screenshots to match
- waiting 100ms before taking screenshot
- taking page screenshot
- disabled all CSS animations
- waiting for fonts to load...
- fonts in all frames are loaded
- 2 consecutive screenshots matched
- final screenshot did not match expectation - 20000 pixels (ratio 0.03 of all image pixels) are different
- 20000 pixels (ratio 0.03 of all image pixels) are different
Expected: /Users/andreylushnikov/tmp/test-results/a-should-work/should-work-1-expected.png
Received: /Users/andreylushnikov/tmp/test-results/a-should-work/should-work-1-actual.png
Diff: /Users/andreylushnikov/tmp/test-results/a-should-work/should-work-1-diff.png
3 | test('should work', async ({ page }) => {
4 | await page.goto('file:///Users/andreylushnikov/prog/playwright/tests/assets/rotate-z.html');
> 5 | await expect(page).toHaveScreenshot();
| ^
6 | });
7 |
```
By default, fixtures share timeout with the test they are instantiated for.
However, for more heavy fixtures, especially worker-scoped ones, it makes
sense to have a separate timeout.
This introduces `{ timeout: number }` option to the list of fixture options
that opts the fixture into a dedicated timeout rather than sharing it
with the test.
Turns out relying on PWTRAP in stack is not reliable: depending on the
call structure, the stack might be cut unpredictably by Node.js.
This patch removes PWTRAP and instead plumbs explicit stack and
pre-set `apiName` all the way down to `wrapApiCall`.
This patch:
- adds call logs to track screenshot timeouts, e.g. due to
waiting for web fonts
- makes sure all snapshot expectations have `.png` extension
- throws a polite error when given a buffer or a string instead of a
page or a locator
- removes stray NL between error description and call log
- makes sure `apiName` is always correct (and adds a test for it)
- Line reporter now shows stats in addition to the test name:
```log
[chromium] › page/page-click-react.spec.ts:108:1 › should not retarget when element changes on hover
23% [21/93] Passed: 17 Flaky: 0 Failed: 0 Skipped: 4 (7s)
```
- When connected to a TTY or with `env.PLAYWRIGHT_LIVE_TERMINAL`
set to anything but `'0'` or `'false'`, line reporter updates in place.
- When not connected to a TTY, line reporter prints an update
after each ~1% of tests done, so it never prints more than 100 lines.
- Updated tests to the golden style.
Use a top-level .env file to control the internal testing setup.
This allows for easy manipulation of environment variables regardless
of your setup (VSCode Extension, CLI, etc.).
This patch adds support to multiple diffs. These are possible
due to soft assertions.
Drive-by: rename second screenshot in `toHaveScreenshot` failure when
re-generating expectations from "expected" to "previous".
This patch:
- Enables configuration of certain defaults for some options of `expect.toHaveScreenshot` method via `TestProject.expect.toHaveScreenshot` property
- Sets sensible defaults for these options:
* `fonts: "ready"`
* `animations: "disabled"`
* `size: "css"`
This patch reverts 2 commits that removed the feature from the method:
- "fix: explicitly ignore maxDiffPixels in toMatchSnapshot (#12570)"
commit b8af8458d6.
- "chore: remove `maxDiffPixels` from toMatchSnapshot (#12539)"
commit a3dff45974.
The `screenshotsDir` option controls the expectation storage
for `toHaveScreenshot()` function.
The new expectation management for screenshots has the following
key properties:
- All screenshots are stored in a single folder called `screenshotsDir`.
- Screenshot names **do not** respect `snapshotDir` and `snapshotSuffix`
configurations.
- `screenshotsDir` is configurable per project. This way a "smoke tests"
project can re-use screenshots from "all tests" project.
- Host platform is a top-level folder.
For example, given the following config:
```js
// playwright.config.ts
module.exports = {
projects: [
{ name: 'Mobile Safari' },
{ name: 'Desktop Chrome' },
],
};
```
And the following test structure:
```
smoke-tests/
└── basic.spec.ts
```
Will result in the following screenshots folder structure by default:
```
__screenshots__/
└── darwin/
├── Mobile Safari/
│ └── smoke-tests/
│ └── basic.spec.ts/
│ └── screenshot-expectation.png
└── Desktop Chrome/
└── smoke-tests/
└── basic.spec.ts/
└── screenshot-expectation.png
```
Previously, we used to skip all the tests from the same file when
any `beforeAll` fails in the file.
Now, we only skip the rest of the tests affected by this particular
`beforeAll` and continue with other tests in the new worker.
Reland: worker.stop() before worker.run() was hanging because
`_runFinished` promise was not initially resolved.
---
This moves `beforeAll`, `afterAll` and some modifiers from running
as a separate entity into running inside a test.
Pros:
- All errors are reported as test errors.
- All artifacts are collected as test artifacts.
- Reporters support this out of the box.
Details:
- Each test computes the necessary hooks to run and runs them.
- Teardown is usually performed during the test (on test failure or worker stop).
- `skipRemaining` is added to `DonePayload` to preserve the behavior
where `beforeAll` hook failure skips subsequent tests.
This behavior can now be improved to only target tests affected by this hook.
* Revert "fix(hooks): separate test timeout from beforeAll/afterAll timeouts (#12413)"
This reverts commit 73dee69558.
* Revert "fix(test-runner): rely on test title paths instead of ordinal (#12414)"
This reverts commit d744a87aee.
* Revert "chore(test runner): run hooks/modifiers as a part of the test (#12329)"
This reverts commit 47045ba48d.
chore(test runner): run hooks/modifiers as a part of the test
This moves `beforeAll`, `afterAll` and some modifiers from running
as a separate entity into running inside a test.
Pros:
- All errors are reported as test errors.
- All artifacts are collected as test artifacts.
- Reporters support this out of the box.
Details:
- Each test computes the necessary hooks to run and runs them.
- Teardown is usually performed during the test (on test failure or worker stop).
- `skipRemaining` is added to `DonePayload` to preserve the behavior
where `beforeAll` hook failure skips subsequent tests.
This behavior can now be improved to only target tests affected by this hook.
This uses `Module._resolveFilename` to intercept module resolution and
check `tsconfig.paths` similarly to pirates usage ot `Module._compile`.
Previously, we resolved during compilation that required reproducible
resolution due to caching. Now we can resolve as we go and support
all `tsconfig.paths`.
- `stdout.isTTY` controls whether list reporter updates lines or just adds them;
- `env.CI` is used in a few places to affect the defaults:
- whether to open interactive html;
- default reporter dot/line;
- default terminal reporter added to non-terminal reporters;
- `env.PWTEST_SKIP_TEST_OUTPUT` is removed;
- `env.PW_TEST_DEBUG_REPORTERS` is introduced specifically for tests.
This way we control the timeout error message from the runner,
so that later on we can differentiate between test timeout, fixture
timeout and hook timeout.
This patch prepares for the `toHaveScreenshot` implementation
by splitting common parts from `toMatchSnapshot`.
Drive-by: fix default extension generation from `.bin` to `.dat`
for unknown buffers.
This patch adds additional options to `toMatchSnapshot` method:
- `pixelCount` - acceptable number of pixels that differ to still
consider images equal. Unset by default.
- `pixelRatio` - acceptable ratio of all image pixels (from 0 to 1) that differ to still
consider images equal. Unset by default.
Fixes#12167, #10219
In experimental ESM mode a child process is forked in order to run the tests. Currently the exit code of this child process is not propagated to the exit code of the parent process, which means that the process exits with a status code of `0` even if some of the tests failed.
This makes it difficult to use Playwright in CI in experimental mode, as the CI pipeline as a whole will pass despite the test failures.
This change addresses this by propagating the exit code in the case where it is non-zero.
The new (as of 1.18) `async testInfo.attach(…)` API handles this
gracefully (and is part of the reason for the new API's existence).
However, for the foreseeable future, it's still possible to manually
push onto the attachments array where we can't validate the contents
until it's too late, so this change ensures more graceful handling in
that case.
Fixes#11565
Soft expects will still fail the test, but will not abort it's execution. As a consequence of this:
- `TestResult` now might have multiple errors, which is reflected with a new `testResult.erros: TestError[]` field.
- `TestInfo` now might have multiple errors as well, which is reflected with a new `testInfo.errors: TestError[]` field.
Fixes#7819
In several of the Playwright APIs, falsey values were not handled correctly. This changeset adds tests (and some fixes):
- route.continue: If options.postData was the empty string, the continue failed to override the post data.
- page.post (application/json with options.data: false|''|0|null): Raw falsey values were getting dropped (i.e. you can't do the equivalent of curl --header application/json … -d 'false'). This has been fixed with most values across all browsers, but an additional fix is needed for 'null' which the channel serializer treats extra specially.
- testInfo.attach: This didn't get reported as an error when options.path was the empty string, but should have been.
#11413 (and its fix#11414) inspired this search as they are the same
class of bug.
- Use file path, not content to calculate the attachment hash.
- Always cleanup fixture from the list on teardown, to avoid reporting
teardown error multiple times: from the test, and from the cleanup.
Previously, reporter would look for a stack frame directly in the test file.
Often times, that is not a top stack frame, especially when the test uses
some helper functions.
This changes error snippets and locations to use the top frame. When top
frame does not match the test file, we additionally show the location
to avoid confusion:
```
1) a.spec.ts:7:7 › foobar ========================================================================
Error: oh my
at helper.ts:5
3 |
4 | export function ohMy() {
> 5 | throw new Error('oh my');
| ^
6 | }
7 |
at ohMy (.../reporter-base-should-print-codeframe-from-a-helper/helper.ts:5:15)
at .../reporter-base-should-print-codeframe-from-a-helper/a.spec.ts:8:9
at FixtureRunner.resolveParametersAndRunHookOrTest (.../src/fixtures.ts:281:12)
```
Textual snapshot diffs were previously broken in the HTML Report. The strikethrough'd text extended beyond the intended region.
HTML Report Before:
<img width="693" alt="Screen Shot 2021-12-27 at 4 43 35 PM" src="https://user-images.githubusercontent.com/11915034/147518750-a60f9002-6eed-48a1-a412-20fabd076fa6.png">
HTML Report After:
<img width="206" alt="Screen Shot 2021-12-27 at 4 48 37 PM" src="https://user-images.githubusercontent.com/11915034/147518762-19a4c8f9-ccc3-4a3c-a962-5a42edc6fc5d.png">
This now matches what's expected and shown in the terminal (which has always been correct):
<img width="1384" alt="Screen Shot 2021-12-27 at 4 36 29 PM" src="https://user-images.githubusercontent.com/11915034/147518799-f538259e-5a45-4d6f-916c-a12ccb620c5b.png">
NB: This MR is a workaround, but not a root cause fix. It works, but I never fully got to the root cause so a bug upstream may be required. It's unclear whether it's (1) in [`colors`](https://www.npmjs.com/package/colors), (2) in [`ansi-to-html`](https://www.npmjs.com/package/ansi-to-html), or (3) Playwright's use of the two. Since the terminal output is correct, I suspect it is in `ansi-to-html`. For example:
```js
const colors = require("colors");
const Convert = require('ansi-to-html');
const convert = new Convert();
// original (strike incorrectly wraps everything in the HTML)
console.log(convert.toHtml(colors.strikethrough("crossed out") + ' ' + colors.red("red")))
// prints: <strike>crossed out <span style="color:#A00">red<span style="color:#FFF"></span></span></strike>
// workaround
console.log(convert.toHtml(colors.reset(colors.strikethrough("crossed out")) + ' ' + colors.red("red")))
// prints: <strike>crossed out</strike> <span style="color:#A00">red<span style="color:#FFF"></span></span>
```
Fixes#11116
This prepares for beforeAll/afterAll hooks to be handled in the same way.
Since we do not know in advance whether a hook will run, we must create
TestResults lazily.
feat(api): add explicit async testInfo.attach
We add an explicit async API for attaching file paths (and Buffers) to
tests that can be awaited to help users ensure they are attaching files
that actually exist at both the time of the invocation and later when
reporters (like the HTML Reporter) run and package up test artifacts.
This is intended to help surface attachment issues as soon as possible
so you aren't silently left with a missing attachment
minutes/days/months later when you go to debug a suddenly breaking test
expecting an attachment to be there.
NB: The current implemntation incurs an extra file copy compared to
manipulating the raw attachments array. If users encounter performance
issues because of this, we can consider an option parameter that uses
rename under the hood instead of copy. However, that would need to be
used with care if the file were to be accessed later in the test.
1. Fixtures defined in test.extend() can now have `{ option: true }` configuration that makes them overridable in the config. Options support all other properties of fixtures - value/function, scope, auto.
```
const test = base.extend<MyOptions>({
foo: ['default', { option: true }],
});
```
2. test.declare() and project.define are removed.
3. project.use applies overrides to default option values and nothing else. Any test.extend() and test.use() calls take priority over config options.
Required user changes: if someone used to define fixture options with test.extend(), overriding them in config will stop working. The solution is to add `{ option: true }`.
```
// Old code
export const test = base.extend<{ myOption: number, myFixture: number }>({
myOption: 123,
myFixture: ({ myOption }, use) => use(2 * myOption),
});
// New code
export const test = base.extend<{ myOption: number, myFixture: number }>({
myOption: [123, { option: true }],
myFixture: ({ myOption }, use) => use(2 * myOption),
});
```
Makes it easier to understand that expect does indeed have a separate timeout.
```
Error: expect(received).toHaveCount(expected) // deep equality
Expected: 0
Received: 1
Call log:
- expect.toHaveCount with timeout 500ms
- waiting for selector "span"
- selector resolved to 1 element
- unexpected value "1"
- selector resolved to 1 element
- unexpected value "1"
- selector resolved to 1 element
- unexpected value "1"
```
This patch:
- rolls stable-test-runner to Nov 2, 2021 tip-of-tree
- introduces a new npm script, `npm run vtest`, to run Visual Regression
Tests for our HTML reporter
Two bug fixes:
- Do not use the worker that is being shutdown for a new job.
- Report unhandled errors during "expected to fail" tests as
fatal errors.
In some circumstances, dispatcher was waiting for all exisitng jobs
to finish before scheduling a new one. This leads to unneded stalls.
Instead, we can schedule jobs right away, if we have a worker
available.
This prevents future retries from passing because of the actual
snapshot being written.
In theory, we can avoid running the retry since it should fail anyway.
However, this brings problems, for example in the `describe.serial` mode
where running a test also has some side effects and so it should not be
skipped. Since running a test without a snapshot is rare, it should be
fine to retry it.
When the test fails (usually with timeout), we wait until all hooks are run
and worker scope is teared down before reporting test end result.
This allows us to collect any error details populated by teardown
in addition to the "timed out" message.
This matches when each expected item from the array
is matched to one of the resolved elements, in order.
Note this performs both "sub-array" and "substring" matching.
Drive-by: documentation fixes.
Drive-by: added "selector resolved to 3 elements" log line
when expecting arrays.
When fixture value `R` is a function, TypeScript sometimes confuses
function `R` and function `async ({}, use) => {}`. This leads to
`any` types in the latter because it could be either of the functions
as TS thinks.
The solution is to only accept the second syntax, assuming that noone
passes fixture value that is a function as is:
```js
// This will stop working.
test.extend<{ foo: (x: number) => number }>({
foo: x => 2 * x,
});
// This will get inferred types and autocomplete.
test.extend<{ foo: (x: number) => number }>({
foo: async ({}, use) => {
await use(x => 2 * x);
},
});
```
Instead of multiple `system-out` entries we produce a single one
with concatenated content. This is compatible with various junit xml
parsers in the wild.
We used to not report fatal error and hang forever because worker
did not run any tests but also did not report any errors.
Also properly show stack-less errors.
This makes `test.fail` tests considered as passing when they actually fail:
- Stop restarting the worker.
- Retry when it passes instead of a fail.
- Behaves similar to regular tests in a `describe.serial` suite.
fix(test runner): avoid internal error for step end without begin
Consider the following scenario:
- Test finishes and starts tearing down fixtures.
- Fixture teardown starts a step S and then times out.
- We declare the test finished (with timeout).
- Dispatcher shuts down the worker and spins a new one for a retry.
Additionally, it clears steps information for the test to be
ready for the new retry. Step S information is lost.
- Meanwhile, during worker teardown, the step S does
actually finish (usually with an error), and we send stepEnd for S.
- Dispatcher does not know what to do with step S end and
prints an internal error.
The fix is to ignore certain messages from the shutting down worker that failed.
When sharing a context between tests and using `'on-first-retry'` we
could end up with tracing still running in non-retried tests. That's
extra overhead without a reason.
Using a worker fixture forces a new worker. This might be unexpected
when part of the test file runs in one worker, and another runs
in another worker. Top-level use of worker fixtures is still fine.
- Uses some auto fixtures to set default options and instrumentation on BrowserType.
- Moves screenshot, trace and video to worker-scoped fixtures.
- Throws in page/context when used from beforeAll/afterAll.
- Plumbs around BrowserType to be accessible from Browser and BrowserContext.
Each hook gets its own test scope. This is not too useful for
object fixtures like `page` (although one can use a page in
`beforeAll` to save storage state), but much more useful for option
fixtures like `viewport`.
We used to sort based on workerHash, and that changes depending on
the exact worker fixtures list. Now we replace workerHash with
an ordinal when constructing the TestGroup list to preserve the
natural order.
- `TestResult.startTime`
- `Suite.location` is optional now
- `Test.status()` renamed to `Test.outcome()` to differentiate against a
`Test.expectedStatus` and `TestResult.status` of the different type.
