`test_edit_textanswer_redirect` is flaky

[geromequa]

test_edit_textanswer_redirect in evap/staff/tests/test_live.py seems to be flaky. The test fails on a few CI runs for unrelated PRs e.g. 22320022227, 62417088524 or 63018645758.

Also: https://github.com/e-valuation/EvaP/actions/runs/22586267335/job/65431926176

[richardebeling]

Code in question:

EvaP/evap/staff/tests/test_live.py

Lines 233 to 238 in 9b14de1

	with self.enter_staff_mode():
	self.selenium.get(
	self.reverse("staff:evaluation_textanswers", query={"view": "quick"}, args=[evaluation.pk])
	)
	next_textanswer_btn = self.selenium.find_element(By.XPATH, "//span[@data-slide='right']")

The last line fails. self.selenium.get() waits for onload to fire. The span is part of the HTML template and not created by javascript, so it should exist when onload fires. Selenium suggest three common error cases, I think we're looking for the correct selector at the right time, so perhaps something went wrong with the previous page load.

[richardebeling]

Playing around with our frontend tests, I found various problems:

• self.selenium.get() is documented to wait for the HTML load event, but the specification says that this should fire as soon as resources have loaded. It seems that this can fire before javascript has actually executed, causing the test to run too early.
• Selenium has an "implicit wait" feature for any find_element calls. By default, this should be disabled, but it seems that some part of our infrastructure actually enables it. Removing it makes various tests fail in situations where they actually depend on waiting behavior (getting a tomselect input, wait for bootstrap carousel element, wait for anything created/removed by our javascript, wait until our click handlers are registered before clicking something, ...)
• The test_edit_textanswer_redirect test here was flaky on one hand because of such a timing flakiness, but on the other hand, it randomly also runs into assertion errors inside the view logic returning a 500 error page from the server. The test framework of LiveServerTestCase doesn't assert the return code, so we just get selenium operating with the 500 error page. The assertion we trigger is in the following stack:

  evap/staff/views.py:1670 in evaluation_textanswers:
  evaluation_sections, contributor_sections = get_evaluation_and_contributor_textanswer_sections(
  
  evap/staff/views.py:1631 in get_evaluation_and_contributor_textanswer_sections:
  text_results.append(TextResult(querstion=question, answers=answers))
  
  evap/results/tools.py:137 in TextResult.__init__
  assert question.can_have_textanswers
  

  The root cause here is the same thing as we observed in Fix crash when deleting a heading question #2665: By chance, the question type will be a HEADING question, and then we trigger this assert.
  I'll try to see if we can make self.selenium.get somehow cause a test assertion failure immediately when a 500 is returned.
• (Various tests have inefficient waits. For example, we have a assert no element with this selector exists thingy at one point which waits for 10 seconds, trying to find an element, before it times out. Not what we want)