E2E Tests: Intermittent failure in Links tests

[aduth]

Examples:

• https://travis-ci.com/github/WordPress/gutenberg/jobs/328478197
• https://travis-ci.com/github/WordPress/gutenberg/jobs/329172043
• https://travis-ci.com/github/WordPress/gutenberg/jobs/329142173
• https://travis-ci.com/github/WordPress/gutenberg/jobs/329142173
• https://travis-ci.com/github/WordPress/gutenberg/jobs/329142177

Error:

  ● RichText › should handle Home and End keys
    expect(received).toMatchSnapshot()
    Snapshot name: `RichText should handle Home and End keys 1`
    - Snapshot  - 1
    + Received  + 1
      <!-- wp:paragraph -->
    - <p>-<strong>12</strong>+</p>
    + <p>-12+</p>
      <!-- /wp:paragraph -->
      266 | 		await page.keyboard.type( '+' );
      267 | 
    > 268 | 		expect( await getEditedPostContent() ).toMatchSnapshot();
          | 		                                       ^
      269 | 	} );
      270 | 
      271 | 	it( 'should update internal selection after fresh focus', async () => {
      at Object.<anonymous> (specs/editor/various/rich-text.test.js:268:42)
          at runMicrotasks (<anonymous>)

cc @ellatrix

[ellatrix]

I've seen it too, but I'm not sure what the problem is. Seems to have been introduced by #22108. Unsure how this is relevant to the bold shortcut though.

[ellatrix]

I'm starting to think we should revert that PR for the stability of the e2e tests. It shouldn't matter, but it seems like the block toolbar not rendering instantly also triggering some intermittent failures for other navigation tests. Still not sure how this could affect the bold shortcut though. This shortcut is available even without the block toolbar rendered.

[aduth]

I'm not sure how much we were expecting to benefit from #22108 to know if it makes sense to revert, or how we'd plan to "properly" work around this. I think as we start to make more things asynchronous for the sake of performance, we'll continue to run into issues with end-to-end test stability, where we assume things can run nicely in sequence, biased by our powerful computers. I know in the past @youknowriad has resisted ideas to turn off asynchronous behaviors in testing in order to more accurate emulate how the user feels. I do sort of agree with this, at least from the perspective of: If things fail in tests because they're run "too fast", inevitably someone could encounter this in the real-world as well.

But it feels like a problem we're going to continue to run into over time, and it really degrades the confidence in our tests, and probably in the actual quality of the application 😕

[sainthkh]

I fixed it in #22202. I thought there was a racing condition. I solved it by waiting for loading the rich text editor.

[aduth]

Fixed by #22202.