Reify on associations fails if using Single Table Inheritance

[ksornberger]

It appears as though if Single Table Inheritance is being used on an association, if you try to reify with has_many: true or has_one: true, it will not find and build the previous version of the association and instead always use the current version of the association when browsing previous versions.

I believe this is happening because if STI is used, the item_type is always stored as the base class. But when calling the method reify_has_many_directly, it is using the main class of the association. This builds a WHERE clause that is looking for the wrong value in the item_type column.

ex:

class Document < ActiveRecord::Base
  has_paper_trail
  has_one :authorship
end

class Assignment < ActiveRecord::Base
  belongs_to :document 
end

class Authorship < Assignment
  has_paper_trail
end

@document.reify(has_many: true, has_one: true)

The last line there will produce a SQL query like this:

SELECT  `versions`.* FROM `versions` 
INNER JOIN `version_associations` ON `version_associations`.`version_id` = `versions`.`id` 
WHERE (version_associations.foreign_key_name = 'document_id') 
AND (version_associations.foreign_key_id = 1) 
AND (versions.item_type = 'Authorship') 
AND (created_at >= '2015-08-26 20:07:28.000000' OR transaction_id = 933)  
ORDER BY versions.id ASC LIMIT 1

But item_type will never be 'Authorship', it is always stored as Assignment. I'm still working on figuring out a solution to this if I can, but this seems to be what is happening from my understanding.

Modifying the database manually and changing the item_type column from Assignment to Authorship results in being able to view the correct versions of the association as I browse through the document history.

[jaredbeck]

Sounds like you're onto something. Thanks for the bug report.

I'm still working on figuring out a solution to this ..

Any luck? Even a failing test would be welcome, thanks.

[jaredbeck]

Still working on this, Kevin? Were you able to write a failing test? Thanks.

[ksornberger]

Hey, sorry.
I'm still planning on looping back to take another look at this, just haven't had the time to get my head into the right space to try to sort this out again. I'll try to make it a priority soon and keep you updated!

[jaredbeck]

I recently adapted the AR bug report template for use in PaperTrail. Maybe it will help you write your failing test? Please check it out: https://github.com/airblade/paper_trail/blob/master/doc/bug_report_template.rb

I hope it helps!

[jaredbeck]

Kevin, I think I've reproduced your issue: https://gist.github.com/jaredbeck/4664e45abe3aa09416ae

Does that look like the issue you're seeing?

[daneshd]

I worked with Kevin on this, your gist does reproduce the issue. Here's another one too:

https://gist.github.com/daneshd/f0e61744075e1617dbcb

I was working on a fix for this issue but I haven't had the time to finish it. One thing I noticed in your report is you'd like the item_type to be Authorship. I went down that path when patching paper_trail but it didn't work out. I eventually found this

Using polymorphic associations in combination with single table inheritance (STI) is a little tricky. In order for the associations to work as expected, ensure that you store the base model for the STI models in the type column of the polymorphic association.

Saving item_type to the child model instead of the base model caused issues with how it was loaded.

I tried to change the way the lookup was done to grab all the versions whose class is the base model here. I only started on the reify_has_ones.

- where("#{version_table_name}.item_type = ?", assoc.class_name).
+ where("#{version_table_name}.item_type = ?", assoc.klass.base_class.name).

This isn't enough though. You could have more than 1 record saved with the base class if your model uses other has_one associations of the same STI base model (e.g. the sponsorship model in my bug report). Say I have a user that has an authorship and a sponsorship, then there would be 2 version records whose base class is Assignment and whose foreign key is my user's id.

I tried iterating through the found versions to lookup and filter these by their type. I wasn't sure about this approach because I had to force PT to never skip/ignore the type attributes, and it meant serializing the object data to figure out which to use.

Our implementation for versioning things caused some other issues for me so I had to put this fix on hold to figure that out. I never got around to reify_has_manys.

[jaredbeck]

This just came up again recently. What do you think about ecd1df7 as a recommendation?

[jaredbeck]

What do you think about ecd1df7 as a recommendation?

Nevermind, I forgot about 23ffbdc. Also, I don't think ecd1df7 would help with reifying associations, would it?

[daneshd]

The STI related documentation from ecd1df7 looks good to me. Reifying the polymorphic associations are currently bugging out though. I think it may be better to say that it's a known issue instead. What do you think?

[jaredbeck]

I think it may be better to say that it's a known issue instead.

https://github.com/airblade/paper_trail#5a-single-table-inheritance-sti now mentions this issue (#594) as a known issue.