Feature Request: `unique=True` column constraint

[gab23r]

Currently, enforcing uniqueness on a column requires either:

1. primary_key=True - but when multiple columns have this, they form a composite key (uniqueness checked on the combination, not individually)
2. A custom @dy.check function - works, but is verbose for such a common constraint

There's no convenient inline way to express "this column must contain unique values" independently.

Example

import dataframely as dy
import polars as pl

class Users(dy.Schema):
    id = dy.Int64(primary_key=True)
    email = dy.String(primary_key=True)  # Want unique emails, but this creates composite key (id, email)
    username = dy.String()

# This passes validation, but shouldn't:
df = pl.DataFrame({
    "id": [1, 2],
    "email": ["a@x.com", "a@x.com"],  # Duplicate email!
    "username": ["alice", "alex"], 
})
Users.validate(df)  # No error - composite key (1, "a@x.com") != (2, "a@x.com")

Proposed Solution

Add a unique=True parameter to column definitions:

class Users(dy.Schema):
    id = dy.Int64(primary_key=True)
    email = dy.String(unique=True)     # Must be unique independently
    username = dy.String(unique=True)  # Must be unique independently

This would check each column's uniqueness separately from the primary key constraint, without requiring a custom check function.

[delsner]

Thanks, while it might be more ergonomic to add a unique parameter, I believe this statement is not true:

There's no convenient inline way to express "this column must contain unique values" independently.

Here's an example for an inline uniqueness check:

class S(dy.Schema):
    a = dy.Integer(check=lambda x: x.is_unique())

Also, your code example seems off, there's no dy.check decorator function exposed 👀
If the issue text was AI-generated, please make sure to (at least) sanity-check code examples.

[gab23r]

You're right you can do this with a custom check, but then you loose the ability to use sample.

class S(dy.Schema):
    a = dy.Integer(check=lambda x: x.is_unique())
S.sample(10)
# ValueError: Samples cannot be generated for columns with custom checks.

Apologies for the completely off example, I fixed it.

[borchero]

Slightly off-topic but is there a good reason to disallow sampling with custom checks given that we do fuzzy sampling anyways Daniel Elsner (@delsner) Andreas Albert (@AndreasAlbertQC) ? 🤔

[delsner]

I don't see a reason why we shouldn't allow it for column-level rules/checks as well.

[gab23r]

I still think that checking the uniqueness on a column can deserve its own parameter. The main reason is that it could allow to have a nicer error message with maybe example of duplicated values. I can start working on the implementation if accepted