rspec users - Apr 2007 - pontential bug in rspec_on_rails and ActiveRecord transactions

Hello rspec-users,

I''ve encountered a strange bug in rspec, but it may be me who is wrong.

I''m running latest version from the trunk (r1678) of both rspec and 
rspec_on_rails.

Please consider the following model:

class Url < ActiveRecord::Base
   def after_save
     Contact.create!(:primary_email => ''user at
example.com'')
     raise ActiveRecord::RecordInvalid.new(self)
   end
end

Basically what it does is that in after_save it will insert a new 
contact into database but after the exception has been raised, the Rails 
will make a rollback on transaction.

Running in console:

Url.create!(:url => ''http://example.com'')

Will yield following SQL (taken from the log)

  BEGIN
  INSERT INTO urls (`icon_url`, `domain`, `url`, `cached_version`, 
`description`, `created_at`) VALUES(NULL, NULL, 
''http://www.example.com/'', 0, NULL, ''2007-04-04
15:36:08'')
  INSERT INTO contacts (`internal_id`, `name`, `updated_at`, 
`primary_email`, `notes`, `user_id`, `created_at`) VALUES(NULL, NULL, 
''2007-04-04 15:36:08'', ''user at
example.com'', NULL, NULL, ''2007-04-04
15:36:08'')
  ROLLBACK

However, when I wrote the specification that looked something like this:

context "Url" do
   setup do
     Contact.delete_all
     Url.delete_all
   end

   specify do
     lambda { Url.create(:url => ''http://www.example.com/'')
}.should
raise_error
     Url.count.should == 0
     Contact.count.should == 0
   end
end

It failed!

The log output was:

  BEGIN
  DELETE FROM contacts
  DELETE FROM urls
  INSERT INTO urls (`icon_url`, `domain`, `url`, `cached_version`, 
`description`, `created_at`) VALUES(NULL, NULL, 
''http://www.example.com/'', 0, NULL, ''2007-04-04
15:35:46'')
  INSERT INTO contacts (`internal_id`, `name`, `updated_at`, 
`primary_email`, `notes`, `user_id`, `created_at`) VALUES(NULL, NULL, 
''2007-04-04 15:35:46'', ''user at
example.com'', NULL, NULL, ''2007-04-04
15:35:46'')
  SELECT count(*) AS count_all FROM urls
  ROLLBACK

How did count statement got into transaction? Is rspec doing some magic 
with ActiveRecord that is causing this to happen?

Thanks for your help,

Esad

* Esad Hajdarevic <esad at esse.at> [20070404
18:57]:
> The log output was:
> 
>   BEGIN
>   DELETE FROM contacts
>   DELETE FROM urls
>   INSERT INTO urls (`icon_url`, `domain`, `url`, `cached_version`, 
> `description`, `created_at`) VALUES(NULL, NULL, 
> ''http://www.example.com/'', 0, NULL, ''2007-04-04
15:35:46'')
>   INSERT INTO contacts (`internal_id`, `name`, `updated_at`, 
> `primary_email`, `notes`, `user_id`, `created_at`) VALUES(NULL, NULL, 
> ''2007-04-04 15:35:46'', ''user at
example.com'', NULL, NULL, ''2007-04-04
> 15:35:46'')
>   SELECT count(*) AS count_all FROM urls
>   ROLLBACK
> 
> How did count statement got into transaction? Is rspec doing some
> magic with ActiveRecord that is causing this to happen?
RSpec does the same as the Test::Unit framework, it will wrap each
test run in a transaction and roll it back at the end, if the
configuration

use_transactional_fixtures = true

is set (in spec_helper.rb). If you want to test behaviour that is
depending on db transactions then I think it is necessary to set this
to false.


Til