The situation

I want an "id" column, but I don't want it to be a primary key, and I want full control over its population.

Steps to reproduce

create_table :people, :id => false do |t|
      t.integer :id
      t.timestamps
end

>> Person.primary_key
=> "id"

Problem #1 - Person should not have any primary key, clearly, from the schema dump:

CREATE TABLE "people" ("id" integer, "created_at" datetime, "updated_at" datetime);

That leads to Problem #2:

>> Person.create(:id => 123)
=> #<Person id: 1, created_at: "2009-01-22 06:31:43", updated_at: "2009-01-22 06:31:43">

Using constructor syntax #create or #new, #attributes_from_column_definition is used and the "id" column is filtered out because it's thought of (incorrectly) as the primary key.

If you use the block creation syntax like:

person = Person.new do |p|
p.id = 123
end
person.save

Then you get id correctly set to 123. So the behavior is inconsistent.

Conclusion

If possible, ActiveRecord should not consider a column as being a primary key unless the DB layer says so. Currently it falls back on the string "id".

I'm not sure if you can "set_primary_key null" in the ActiveRecord model, but that might be a valid workaround if we can't rely on the DB layer. In the case of a workaround, it should be documented alongside AR::CA::SS#create_table to let people know they should explicitly set the primary key to null (or whatever) if they expect to use a column called "id".