Some browser tests do not succeed anymore

[lmcbout]

Bug Description:

It looks like the test don't run anymore. While publishing Theia version 1.3.0, I had to by-pass the browser test.

• On the other hand, Travis seems to pass all tests.
• On my local computer, cleaning and executing the tests fail

Note: removing the $HOME/.theia solved some errors but not all
Do we have dependencies on the home folder?

The tests have been running a few times and we get the same result

Steps to Reproduce:

1. git clean -ffdx
2. yarn
3. yarn test
   -> some tests don't succeed

Additional Information

Errors:

1. Find and Replace
   Replace in the active editor:
   TypeError: Cannot read property 'widgets' of null
   at ViewContainer.../../packages/core/lib/browser/view-container.js.ViewContainer.getParts (bundle.js:128607:37)

root INFO 2) Menus
should toggle 'Explorer' view:
Error: Timeout of 2000ms exceeded. For async tests and hooks, ensure "done()" is called; if returning a Promise, ensure it resolves.

root INFO 3) Menus
reveal more context menu in the explorer view container toolbar:
TypeError: Cannot read property 'widgets' of null
at ViewContainer.../../packages/core/lib/browser/view-container.js.ViewContainer.getParts (bundle.js:128607:37)
at Context. (/home/lmcbout/theia/stable/theia/examples/api-tests/src/menus.spec.js:88:42)

• Operating System: Ubuntu 18.04
• Theia Version: Latest

[vince-fugnitto]

The view-spec.js work when executed by themselves, but during the suite it fails locally.
I think it might be in a weird state due to other tests which are executed before.

[akosyakov]

Could we trace back to which PR started triggering a weird state?

[vince-fugnitto]

Could we trace back to which PR started triggering a weird state?

I tested with the v1.2.0 tag and the api integration test cases passed successfully, I'll try to pinpoint the regression.

[vince-fugnitto]

I identified the breaking commit as fdd7c41 using git bisect.

[vince-fugnitto]

I identified the breaking commit as fdd7c41 using git bisect.

I identified the commit (where the tests started to fail locally), however I was unable to determine how to fix them.
I tried temporarily to skip the failing tests but it still causes the entire suite to fail. I'm wondering if the tests are interdependent on eachother causing them to be in a weird state.

I noticed that:

• they fail locally and not on travis
• they do not fail if they are run individually (using describe.only)

[akosyakov]

I will have a look later this week. Tests are run together in the same app.

[akosyakov]

It is reproducible when menus and views tests are enabled.

[akosyakov]

It is timing issue, fdd7c41 makes sure that VS Code extensions views are properly registered in the explorer (shell). It slows down the creation of the navigator.