Colin Law wrote in post #1091609:
Yes, but really don't do it unless you are working with a legacy db
that you cannot change (absolutely definitely impossible to change).
Otherwise use the default id for the primary key and add compound
indexes for the compounded fields if necessary. That will be /much/
First, I agree with this completely. I just wanted to add a bit of
clarification as to why I agree with this.
An important aspect of working with Rails (or any modern Object
Relational Mapping framework) is mapping an object's identity with a
database table row. Rails by default uses a simple integer value, which
it manages itself, to ensure that a model object always maps to a single
database table row in the database. This simple integer value is then
used as the primary key for the table.
As Colin says, it really is best to just let Rails have its identity
column. There is no reason you cannot have your identifying column(s) on
that table as well, generated by whatever means you wish. Just add a
unique index to your column(s) and treat that as your key, but don't
make that the table's primary key. Let Rails manage that using its own
built-in identity mapping.
This technique also makes your database design less fragile. I've run
into many, many cases where some business rule changes forcing all
identifying columns to have to change for one reason or another. If that
column(s) is used for relational mapping then multiple tables have to be
updated in order to facilitate the change. Yet when surrogate primary
keys are used for mapping relationships (as well as the ORM identities)
this sort of change is isolated to the single table where the natural