STI cast new instances to default type on initialize.

[kuldeepaggarwal]

fixes #17121

[kuldeepaggarwal]

any update?

[rafaelfranca]

Please don't ping in the issue tracker. It will be reviewed when someone have time and want to review.

[kuldeepaggarwal]

Apology for creating noise in the issue tracker but its been 3 months. So, it was just a gentle reminder. I will take care of it.

[rafaelfranca]

No problem.

Yeah, we are a lot behind the PRs schedule 😢. I'm trying to catch up, so expect this PR to being reviewed 🔜 😄

[kuldeepaggarwal]

Thanks. 👍

[ryanrperez]

I can verify that the added tests are correct and this is a fix for the issue.

[jormon]

Workaround: in the class at the top of the inheritance hierarchy:

after_initialize do
    self.type = "Company" if self.class == Company
end

(for those too lazy to run off a patch) :/

[senny]

@kuldeepaggarwal can we try to make a copy of attrs with inheritance_column => inheritance_column_default if inheritance_column is not already in attrs. We could possible do that here: https://github.com/kuldeepaggarwal/rails/blob/fix-STI-default-type/activerecord/lib/active_record/inheritance.rb#L53 and then subclass_from_attributes? and subclass_from_attributes can be left as is.

[kuldeepaggarwal]

@senny Sure. will do it asap.

Thanks for the hint.

[kuldeepaggarwal]

Updated the PR.

[senny]

@kuldeepaggarwal shouldn't this happen before if subclass_from_attributes?(attrs)? Otherwise subclass_from_attributes?(attrs) will not see the default.

[kuldeepaggarwal]

We are not actually seeing the defaults in subclass_from_attributes?(attrs),

def subclass_from_attributes?(attrs)
  attribute_names.include?(inheritance_column) && attrs.is_a?(Hash)
end

[senny]

👍 sorry, my bad. Misread attribute_names as attrs

[senny]

@kuldeepaggarwal sorry for the long delay 😓. Finally got around to cleaning up my inbox. This doesn't apply cleanly on master. Can you submit a rebased version?

[kuldeepaggarwal]

@senny: Sure, give me sometime.

[kuldeepaggarwal]

@senny Updated the PR.. 😄

[senny]

@kuldeepaggarwal do we still need that ||. Previous code was calling with_indifferent_access directly, so that should never be nil, no?

[kuldeepaggarwal]

No, actually in my previous commit I was initializing attrs ||= {} so that's why I called with_indifferent_access directly, which is equivalent to current implementation.

[senny]

ah, I see. It's because we modify subclass_from_attributes? in a weird way. Why not keep subclass_from_attributes? as is but modify this part:

def new

        if subclass_from_attributes?(attrs)
          subclass = subclass_from_attributes(attrs)
        else
          subclass = columns_hash[inheritance_column].default
        end

This way the methods still do what their name suggests and it should be obvious at a glance how the subclass is determined.

[senny]

We'd still have to do the find_sti_class(subclass_name) though.

[kuldeepaggarwal]

Modified the implementation.

[senny]

@kuldeepaggarwal thank you 💛

I made some modifications in the merge commit so that we can get rid of some duplicate conditionals.

[kuldeepaggarwal]

Awesome..