incorrect sql from has_many :through

I am not able to reproduce this issue in rails edge.

-1 not reproducible

This ticket is not fully coherent to me. What are you wanting to get with FleetAgent.count(:include=> [:position_mappings])
which would be any different from a plain FleetAgent.count ?

+1 verified in current master.

Here are my models:

class User < ActiveRecord::Base
  has_many :position_mappings
end

class PositionMapping < ActiveRecord::Base
  belongs_to :user
end

class FleetAgent < ActiveRecord::Base
  belongs_to :user
  has_many :position_mappings, :through => :user
end

And here's the error:

ruby-1.8.7-p249 > FleetAgent.count(:include => [:position_mappings])
ActiveRecord::StatementInvalid: Mysql::Error: Unknown column 'users.user_id' in 'on clause': SELECT     COUNT(DISTINCT fleet_agents.id) AS count_id FROM       fleet_agents LEFT OUTER JOIN users ON fleet_agents.id = users.user_id LEFT OUTER JOIN position_mappings ON position_mappings.user_id = users.id

But, strangely, the error doesn't happen when calling for actual models instead of counting:
(Note: I didn't actually write any records into the DB)

ruby-1.8.7-p249 > FleetAgent.all(:include => [:position_mappings])
 => []

But why would one want to count FleetAgents and include the position_mappings relation in it? To create conditions based on the position_mappings for the count?

Yes, haven't repro'd this with latest rails, but I think the original issue only occurred once you tried to set any kind of conditions on the relation, it's not the relation itself that see's the problem. So .count is just one way to repro it, but also doing any kind of filter I think should also cause the same issue?

I think if you glance at the code I referenced in the original error report (and presumably factor in that it might have moved line since), then the problem was fairly easy to see - there are several code paths through that section and quite clearly you can see that several types of association should be handled, but there is code only for I think "has_many" associations?

I think if you look at the code the problem should be quite apparent? The challenge to code a fix is that I wasn't sure what other code paths enter this hunk and hence what the full and proper fix should look like. It's somewhat abstract at this point what the various variables point to in terms of different types of relationships and I think it would take someone a little more familiar with what else uses this code to do a full fix (ie we should cover all types of relationship here). I don't think the change is hard or a big change, it's simply that the code appears to be used by several other (unknown) code paths and the solution looks a bit subtle?

From memory it's only a problem when you apply a relationship which is probably why not so many people bang into this?

Just realised the file name of that patch says 2461. This is a mistake, but rest assured the patch is definitely concerned with this ticket and not a different one.