[IslamElsayed]

Fixes #1618.

build_stubbed picks the id from an integer sequence unless the primary key is a uuid, and it decided that by comparing the column's sql_type to "uuid":

column.respond_to?(:sql_type) && column.sql_type == "uuid"

That is PostgreSQL's name for the type. SQL Server calls the same thing uniqueidentifier, so the check failed, next_id fell through to @@next_id += 1, and assigning an Integer to a uuid attribute left id nil — the symptom in the issue.

The adapters disagree on the name but agree on the Active Record type. From the SQL Server adapter's own source:

# lib/active_record/connection_adapters/sqlserver/type/uuid.rb
class Uuid < String
  def type
    :uuid
  end

  def sqlserver_type
    "uniqueidentifier"
  end
end

# sqlserver_adapter.rb
m.register_type "uniqueidentifier", SQLServer::Type::Uuid.new

So this asks the column for type first and keeps the sql_type comparison as a fallback for anything that reports the name without mapping the type.

Testing

Three examples in spec/factory_bot/strategy/stub_spec.rb, driven by a column double so they need no SQL Server:

• type :uuid / sql_type "uniqueidentifier" → uuid id (the reported case)
• sql_type "uuid" with no type → uuid id (the existing PostgreSQL path)
• type :integer → still an Integer id

I checked they are not vacuous by breaking the fix both ways: reverting to the sql_type-only check fails the first, and dropping the sql_type branch fails the second. The existing spec/acceptance/stub_spec.rb uuid examples still pass unchanged.

Full suite: 777 examples, 0 failures. standardrb clean on both files. Run under Ruby 3.3.5 rather than the 4.0.1 in .tool-versions, since the gemspec allows >= 3.0.0.

I do not have SQL Server to hand, so the adapter behaviour above is read from its source rather than executed — correct me if uniqueidentifier reaches column_for_attribute differently in practice.

Disclosure: I use AI assistance when working on patches, and I check and test everything before submitting.

[IslamElsayed]

Friendly ping on this one. CI hasn't run yet because it's waiting for first-time contributor approval. Could a maintainer approve the workflows when there's a moment? Happy to make any changes. (I used an AI assistant while preparing this PR, and I reviewed and tested the change myself.)