Domain: Prevent constructing domain with non-unique variable names

[VesnaT]

Is your feature request related to a problem? Please describe.
We have an agreement to construct domains with unique variable names, but there are no checks for that in the code.

Describe the solution you'd like
Raise an error when constructing a domain with duplicated variable names.

Describe alternatives you've considered
Create unique variable names under the hood when constructing the domain.

[janezd]

I support raising an exception because we should prefer failing early. The cost of this will be unhandled exceptions at various places, but this is better than unreported errors in various places (and miscomputations later).

I don't like the under-the-hood idea.

[VesnaT]

Before we accept this enhancement, it is necessary to fix all widgets that may construct a new domain (which seems impossible).

[AndrejaKovacic]

Problematic widgets

Data widgets

• File allows changing of names indiscriminately (fixed in [FIX] Ensure unique var names in file #4431)
• Paint data (fixed in [FIX] Owpaintdata, owpivot: ensure unique domain #4578)
• Pivot (fixed in [FIX] Owpaintdata, owpivot: ensure unique domain #4578)
• Concatenate
• Preprocess (for PCA)
• Feature constructor

Visualize

• Venn (fixed in [FIX] Unique domain checks #4760)
• Silhouette (fixed in [FIX] Unique domain checks #4760)

Unsupervized

• T-SNE (fixed in [FIX] Unique domain checks #4760)
• DBScan (fixed in [FIX] Unique domain checks #4760)
• Manifold Learning (fixed in [FIX] Unique domain checks #4760)

Evaluate

• Confusion Matrix (fixed in [FIX] Unique domain checks #4760)
• Predictions (fixed in [FIX] Unique domain checks #4760)
• Test and Score (fixed in [FIX] Unique domain checks #4760)

[VesnaT]

Another proposal: Prevent constructing domains from tuples and lists. Create custom collection e.g. SortedSet. The collection is returned by get_unique_names. The purpose is to force the user to always use the get_unique_names when constructing the domain.

[janezd]

Clever idea. But it would be backwards incompatible. Plus ... think about tests with hardcoded names - which are obviously unique. I'd hate having to call get_unique_names there. Plus if everybody just called get_unique_names, this is not much different from Domain calling it itself, I guess.

[janezd]

I usually remember to edit "Partially fixes ..." to "Fixes (partially) ..." to prevent closing of an issue. In #4578 I forgot. :)