Hierarchical Clustering: smaller ControlArea

[ajdapretnar]

What's wrong?
Hierarchical Clustering has a very wide ControlArea, depending on the initial input data.
ControlArea could be smaller (narrower).

How can we reproduce the problem?

1. File (iris) - Distances - Hierarchical Clustering. Everything ok.
2. New Orange instance. Datasets (HDI) - Distances - Hierarchical Clustering. ControlArea very wide.
   Note that when I open HDI after iris, ControlArea is ok. So it depends on initialization.

Compare below:
Iris

HDI

What's your environment?

• Operating system: OSX
• Orange version: 3.38.1
• How you installed Orange: pip

[janezd]

Note to whomever will be resolving this: I think that the width of Annotations and Color by combos should be reduced by setting their horizontal size policy to Ignored. This seems to make them occupy all available space but not more.

The problem with this solution -- or any other that limits the width of the (closed) combo -- is that the popup will be just as wide as combo. The popup seems to ignore its size policy and even directly setting its width doesn't work. The only place where the width can be extended is in the showPopup method. Setting a different width at https://github.com/biolab/orange-widget-base/blob/master/orangewidget/utils/combobox.py#L276 works. To determine the proper width we could check the size policy and -- if needed -- observe the width provided by sizeHintFromColumn(0).