I don’t have a slam dunk solution, but one useful suggestion is to catch AR::StatementInvalid, and then in your rescue block check for a duplicate record in the database. If you find a duplicate record, chances are that’s the reason for the exception and you can handle it accordingly, whereas if you don’t find a duplicate, then the AR::StatementInvalid must be for some other reason and my inclination would be to reraise.

@Robin Ward: Thanks for the note. We’re doing our best to be PostgreSQL-compatible (harder than it sounds), so we do want to avoid MySQL-specific error handling at this point. But you’re right, that comes at a cost, since the exception we catch is quite general. Regarding the error-stashing: we’ve found that most duplicates are inadvertent, so we prefer to ignore them silently, but your approach is certainly valid.

Catching StatementInvalid alone will work if you violate an index, but it will also fire if somehow you have cocked up your SQL.

Unfortunately, it seems each database has its own version of the error that occurs when you violate an index. But if you know you are going to use MySQL, you can do something like this:

rescue ActiveRecord::StatementInvalid => error
if error.to_s.include?(“Mysql::Error: Duplicate entry”)
# raise your own exception

Note that a more graceful way than re-directing would be to stash something in the errors collection. That’s how I do it anyway :)

Instead, this is what I’m doing:

1. database-level uniqueness constraint to ensure data integrity

2. JavaScript to mitigate the issue

I should probably also catch ActiveRecord::StatementInvalid (if the user doesn’t have JS) and do something useful like redirect them to the next step of the registration process for the user with the details they just resubmitted.

By the way, I’ve seen this issue from a user using IE7, but I can’t reproduce with Firefox 3. Possibly Firefox queues up the HTTP requests, waiting for one (the first submit) to complete before starting the next. So instead of an database level error, I would get a Rails-level validation error about the e-mail already being in use. My JS mitigates that as well.