[ENH] File: explicit file format choice

[markotoplak]

Issue

Resolves #5724

Description of changes

For files, the reader used was already stored in .file_format, which could be either a reader's qualified name or None (which means that reader is detected depending on extension and priority).

All available readers are shown in a combo box.

For URLs, the reader selection is disabled. That part would have to be refactored a bit so that the chosen reader could be saved.

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

Merging #5736 (9586906) into master (06d7f24) will increase coverage by 0.00%.
The diff coverage is 98.78%.

@@           Coverage Diff           @@
##           master    #5736   +/-   ##
=======================================
  Coverage   86.16%   86.16%           
=======================================
  Files         315      315           
  Lines       66195    66258   +63     
=======================================
+ Hits        57035    57092   +57     
- Misses       9160     9166    +6     

[borondics]

I think this is great, but I have two comments:

The error message if the selected reader is not the right one is still not too informative:

Then, the widget forgets the selected reader for subsequent files. I think it would be nice to save this in the widget settings because one could expect that the same user works with the same type of TXT files. The best solution here would be to first try to autodetect and then fall back to the last selected for that file extension in my opinion. Might be difficult to store all this data though...

[markotoplak]

@borondics, I do not want to do any application of the currently selected reader onto a new file into this PR. Added automatics like this can lead to confusion (remember disappearing settings is Select Columns or Feature Constructor) so I would not add it unless absolutely sure. Furthermore, your proposed change would change the current behavior, while this PR aims to keep it exactly as it was, it only introduces the possibility of explicit modifications.

The message is very precise (= not useful for non-developers), just click/mouseover on it. :)

[borondics]

OK, I understand that this PR is aimed only for the new placement of the file loader selector. I hope though that in another PR the saving will be also possible.

For the error: yeah, I know that there is a longer message on moouse-over but something for the normal user wouldn't hurt either. I guess there are more users than developers, so it is not absurd to also let them clearly know what is happening. :)

[markotoplak]

Pylint has a fun complaint: there are too many public methods (=tests) in the test class. :)

[ajdapretnar]

I checked and this PR breaks nothing in Text. 👍

A small comment. It would be nice(r) if add-on readers would be listed separately. I.e.
.tab
.csv
.pkl
.basket (is this even core?)
-----
.HDF5
.dat
.gsf....

So readers from each add-on would be in a separate "category".

[VesnaT]

If I try to open the Iris dataset with an Excel reader, I get a wrong error message:

[markotoplak]

I tried to address comments by @ajdapretnar and @VesnaT: error reporting is now fixed and the list is sorted per package name.