Investigate failing sync service test on stable27

[max-nextcloud]

Describe the bug
For a while the sync.spec.js has been failing on the stable27 branch:

Looks like it may have been introduced by #5087 or at the same time. The cypress run in the PR passed though.

To Reproduce
Steps to reproduce the behavior:

1. Go to cypress action run overview for stable27 branch
2. See the failed runs
3. Scroll down till you see green runs again.
4. Inspect individual failed runs to see most of them fail in sync.spec.js.

Expected behavior
Cypress should pass on all integration branches.

[max-nextcloud]

Server log is probably unrelated - maybe from a retry:

{
  "reqId": "XNUpWmg1ATmxO2rATg7u",
  "level": 3,
  "time": "2023-12-21T05:51:19+00:00",
  "remoteAddr": "127.0.0.1",
  "user": "admin",
  "app": "ocs_api",
  "method": "POST",
  "url": "/ocs/v2.php/cloud/users?format=json",
  "message": "Failed addUser attempt: User already exists.",
  "userAgent": "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Cypress/13.6.1 Chrome/114.0.5735.289 Electron/25.8.4 Safari/537.36",
  "version": "27.1.5.1",
  "data": {
    "app": "ocs_api"
  }
}

[max-nextcloud]

running the last passing commit of text with the latest commit of server stable27 passes: https://github.com/nextcloud/text/actions/runs/7104754058/job/19905039693
At the same time the current state of text with the last passing commit of server still fails: https://github.com/nextcloud/text/actions/runs/7303886691/job/19905235351

So looks like this somehow broke on the text side of things rather than the server side.

[mejo-]

I tried to understand what happens here, but failed so far.

Two tests in sync.spec.js fail reliably when running them with cypress run locally - even though they pass reliably with cypress open:

• saves the actual file and document state
• saves on close

Both tests load a document, edit the document, download the file and see whether the edits are in the downloaded file. The first one does a manual save before download, the second closes the document before download.

In the screenshots it's obvious that the file got saved. I suspected a race condition, but adding cy.wait() at different places in the test doesn't change the situation. The downloaded file always stays the one initially uploaded, without the editing changes.

[mejo-]

Ok, found a reliable fix now: #5601

[mejo-]

Fixed by #5604