[rspec-users] Rollbacks, Sqlite3 bug

Scott Taylor mailing_lists at railsnewbie.com
Thu Oct 25 00:57:34 EDT 2007

On Oct 25, 2007, at 12:40 AM, David Chelimsky wrote:

> On 10/24/07, Scott Taylor <mailing_lists at railsnewbie.com> wrote:
>> Okay - so the sqlite bug reported a day or so ago on the list is real
>> bug.  I'm going to file something in the tracker for that...
>> I also learned that before(:all)...after(:all) is not wrapped in a
>> transaction, the way before(:each)...after(:each) is.  Is there some
>> reason behind this?  Can you not wrap transactions inside  
>> transactions?
> Only for DBs that support it. And it would add quite a bit of
> complexity to rspec I think.
> This is not really the intent of before(:all). It is more akin to
> TestCaseSetup in xUnit - i.e. a class level method that doesn't share
> state w/ the examples. So you might have some load script to get the
> DB into a known state before(:all), and then run each example in a
> transaction so it rolls back to that known state. I *think* that's how
> rails does it but I'd have to peruse the code to verify that.

Yeah, that's my guess, too.

> Accessing fixtures in before(:all) is something you can't do right now
> because before(:all) is not running in the same scope as the examples
> are - in fact, it runs before fixtures are loaded.

I'm still a little confused about fixture loading.  Isn't the test  
database populated with fixtures when running rake db:test:prepare?

> Even if we changed
> that, you still wouldn't want to access them before each, because
> you'd be keeping a handle on objects whose underlying data are getting
> rolled back after each example. Talk about unexpected
> consequences!!!!!

Well, we aren't using fixtures, but instead using the  
FixtureReplacement plugin (I'd be interested in your feedback, David,  
or anyone else, for that matter).

I'm a little worried that the plugin is a little too good (for  
readability, and for easy of writing tests): My coworker, who is new  
to rspec, would rather hit the database in controller specs than mock  
database access out.

Anyway, the situation was that he was hitting the database by  
inserting 40 records per test case.  Of course the thing came to a  
screeching halt, and he changed the before(:each) block to a before 
(:all).  The tests in that description didn't fail, but other ones  
did (because some of the records were still hanging around).   
Although the strangest thing was that running rake spec didn't fail  
the tests, but running them by hand did (and the same with Autotest).



More information about the rspec-users mailing list