Ruby on Rails

Comments and changes to this ticket

• 

  Rob Sterner June 20th, 2008 @ 04:23 PM

  The best way to describe the bug that we've run in to is to walk through the steps which lead to its discovery.

  When a database (acme_development) schema is loaded from migrations, a typical table definition:

  create_table :foo do |t|
    t.references :bar
  end
  

  causes a column, bar_id, to be created with a MySQL data type of int(11). Also, :primary_keys are hard-coded to be of type int(11).

  Starting (I believe) with Rails 2.1, when invoking a rake task that would (re)generate schema.rb the code examines the MySQL columns (in acme_development) to determine the :limit. If no default :limit was provided in the original migration, this will result in schema.rb being populated with the current acme_development size - 11, for all :integer columns.

  When acme_test is populated via schema.rb, the table definition becomes:

  create_table :foo do |t|
    t.integer :bar, :limit => 11
  end
  

  Specifying :limit => 11 to the column (the same limit as the default size hard-coded in mysql_adapter.rb) causes a different data type to be used: bigint(11). This is the root of the problem - specifying a :limit whose size is identical to Rails' default for :integer causes a different data type to be created when the database is loaded.

  We only stumbled across this when foreign keys were being inserted into our test database - MySQL can't create foreign keys across columns with different data types (int != bigint), so it fails.

  We'll be working on a patch, will update in a few days if we can't come up with an acceptable, generic solution.

• 

  John Trupiano June 24th, 2008 @ 11:29 AM

  Rob, any progress here? I was just about to start a patch....but if you're close to done or already done, I'd just wait it out.

  I think it should be fairly trivial. I'm looking into it as I type....I'm sitting in #rails-contrib (jtrupiano), so get in touch with me if you're in the middle of this.

• 

  Jeremy Kemper June 24th, 2008 @ 03:29 PM

  • → State changed from “new” to “open”
  • → Assigned user changed from “” to “Jeremy Kemper”
  • → Milestone changed from “” to “2.1.1”

  mysql integer limit should be ignored since it's really display width. Limit should be set as byte length according to tinyint/smallint/mediumint/int/bigint.

  Please do submit a patch + tests!

• 

  Rob Sterner June 24th, 2008 @ 03:36 PM

  Hey Jeremy,

  Just looked for you on IRC to talk about this. I was preparing my patch (patched against 2.1.0) when I ran it against edge (master) Rails, seems like your commits over the last 24 hours have fixed the issue.

  Now the only outstanding thing that I think the source/docs would benefit from is documentation explaining that the fundamental concept of what a :limit is has changed w/r/t Rails :-D

• 

  Jeremy Kemper June 24th, 2008 @ 03:51 PM

  • → State changed from “open” to “resolved”

  Yeah, should document that. There was no real concept before, just dumped it on the database's lap which led to odd results.

• 

  John Trupiano June 24th, 2008 @ 04:05 PM

  Hey Rob, Jeremy, other interested parties, I blogged about this whole saga here: http://blog.smartlogicsolutions....

  Jeremy, it does include a two-line patch for the 2.1 release. Not sure if that's useful at all...

  Thanks for working on this with me Rob.

• 

  Rob Sterner June 24th, 2008 @ 04:17 PM

  I figured that docs were TBD as this is a subtle but important change. Thanks Jeremy! Thanks again John, was fun trying to figure this out!