Don't generate dashboards for non-constant models

[c-lliope]

Fixes #145 and #219

Problem:

When the same has_and_belongs_to_many relationship is defined twice,
Rails generates an unnamed subclass of ActiveRecord::Base.
This is probably a bug in Rails.

The unnamed subclass shows up like this:

> ActiveRecord::Base.descendants.last
 #<Class:0x007fddea1bdf08> (call '#<Class:0x007fddea1bdf08>.connection' to establish a connection)

When the install generator tries to populate the DashboardManifest
with all of the ActiveRecord models,
it trips up on unnamed models,
and generates a manifest with invalid syntax.

This results in a NameError for the user when they run the
administrate:install generator:

/gems/administrate-0.1.0/lib/generators/administrate/dashboard/dashboard_generator.rb:87:
in `const_get': wrong constant name #<class:0x007fcf52bfe248> (NameError)

This error most often pops up in applications
that are using the spring gem.

Solution:

• Add a test case that reproduces the error
  by defining duplicate has_many associations.
• In the install generator, ignore any models that are not constants.
  We determine constants by checking that the model's #name
  matches its #to_s.
• Print a warning for the ignored models.

[jakecraige]

Is there something that could be linked to the user to help them debug/fix this?

[c-lliope]

@jakecraige I... don't know.

I'm about 85% sure this is a bug in Rails, and I'm going to open a PR there to try to fix it. On the user's end, they're probably not aware they even have a non-constant class in their application, so it'll be strange for them to see a warning.

I'm considering removing this warning altogether, or changing it to something less scary-sounding, like:

NOTICE: Skipping dynamically generated model `#<Class:asdfasdf>`.

Thoughts?

[sgrif]

I spent a bit thinking about this. Does this even need to be a warning? I feel like we can pretty certainly say that if a class isn't assigned to a constant, you can ignore it. (Also you should look at the implementation of descendants if you haven't in the past, because it's hilarious)

[c-lliope]

Ping @sgrif - figured out the issue where ActiveRecord is creating unnamed classes. Looks like it's a bug in ActiveRecord - is that right?

I have the bandwidth to open up a PR for it, if you'd like.

[c-lliope]

Talked to @sgrif - since calling has_and_belongs_to_many multiple times for the same relationship isn't a supported use case in Rails, the bug actually lives in Spring. For some reason, it's reloading the relationship when it shouldn't be.

[sgrif]

For some reason, it's reloading the relationship when it shouldn't be.

More likely it's retaining the constant that should be getting released as part of the general constant reloading process.

[c-lliope]

yep, good catch. Thanks!

[Rodrigora]

Cool @Graysonwright. 💯

Just letting you know that I do not have duplicated HABTM relationships. These are my models:

class Feature < ActiveRecord::Base
  has_and_belongs_to_many :units
end

class Unit < ActiveRecord::Base
  has_and_belongs_to_many :features
end

[c-lliope]

@Rodrigora yep, I think there may be a bug in spring that is causing the equivalent effect of a duplicated HABTM relationship.

Are you using spring in your app?

[Rodrigora]

Yes @Graysonwright, I'm using it.

You're right, certainly it is a Spring bug. Thanks!