Fix calls to set_domain

[janezd]

Issue

Fixes #5323, and the same in three other widgets.

This happenned because empty table is also considered false, hence data and data.domain resulted in passing a Table (instead of None or an instance of Domain) when the table was empty.

I'd prefer not to write tests for this, though they'd be trivial. I don't like tests that just replicate a bug. Of course one could say "but you should test that a widget does not crash on empty table", but then we should add such a test to all widgets, not just the four that happenned to have this problem.

Includes

• Code changes
• Tests

[codecov]

Codecov Report

Merging #5324 (0a809e4) into master (63ed27f) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #5324   +/-   ##
=======================================
  Coverage   86.29%   86.30%           
=======================================
  Files         301      301           
  Lines       61191    61193    +2     
=======================================
+ Hits        52806    52810    +4     
+ Misses       8385     8383    -2     

[ajdapretnar]

I have a couple of issues, not necessarily with this fix, but let's discuss what needs to be solved where.

1. Merge Data. The first issue is now solved, so when I connect from Pivot Table to Merge Data. But if I connect the second table to Merge Data, a new, but related issue appears.

Traceback (most recent call last):
  File "/Users/ajda/orange/orange-canvas-core/orangecanvas/scheme/signalmanager.py", line 1010, in __process_next
    if self.__process_next_helper(use_max_active=True):
  File "/Users/ajda/orange/orange-canvas-core/orangecanvas/scheme/signalmanager.py", line 1048, in __process_next_helper
    self.process_node(selected_node)
  File "/Users/ajda/orange/orange-canvas-core/orangecanvas/scheme/signalmanager.py", line 679, in process_node
    self.send_to_node(node, signals_in)
  File "/Users/ajda/opt/miniconda3/envs/o3/lib/python3.8/site-packages/orangewidget/workflow/widgetsscheme.py", line 792, in send_to_node
    self.process_signals_for_widget(node, widget, signals)
  File "/Users/ajda/opt/miniconda3/envs/o3/lib/python3.8/site-packages/orangewidget/workflow/widgetsscheme.py", line 833, in process_signals_for_widget
    widget.handleNewSignals()
  File "/Users/ajda/orange/orange3/Orange/widgets/data/owmergedata.py", line 369, in handleNewSignals
    self.openContext(self.data and self.data.domain,
  File "/Users/ajda/opt/miniconda3/envs/o3/lib/python3.8/site-packages/orangewidget/widget.py", line 1113, in openContext
    self.settingsHandler.open_context(self, *a)
  File "/Users/ajda/orange/orange3/Orange/widgets/data/owmergedata.py", line 191, in open_context
    self._encode_domain(domain1),
  File "/Users/ajda/orange/orange3/Orange/widgets/data/owmergedata.py", line 210, in _encode_domain
    all_vars = chain(domain.variables, domain.metas)
AttributeError: 'Table' object has no attribute 'variables'

2. Correlations. Now it reports 0 instances on the input and a warning "At least two instances are needed". Transpose reports None (-). We should decide on one option, IMO.
3. Create Class. Report None (-) and a warning "Data contains only numeric variables." Shouldn't widgets report "No data on input" or something?

[janezd]

1. Fixed. (The general DomainContextHandler accepts Domain or Table; MergeDataContextHandler didn't, but now does.)
2. Inconsistencies in summaries will be fixed by auto-summaries ([ENH] Automated and better summaries #5308).
3. I fixed the warning.

As for whether the widget should report that the data table is empty: I wouldn't, because in this case every widget would have such a warning (or even an error) with the corresponding additional code ... I think a "0" in the summary should suffice, unless some particular widget needs to show a warning -- for instance a widget that need at least two instance and wants to inform the user that a single instance is not enough.

[ajdapretnar]

Ok, let's go with a 0. It is just slightly confusing from a user's perspective, since we denote "empty" signals with a dashed line. Here, the line is full. But I agree, people should get used to using the status bar. ;)