Check for inverse_of in manually defined associations

[lowjoel]

Rails 4 added the inverse_of option when defining has_one and has_many relationships. It then does heuristics to determine the inverse of the relation, but does not warn when the inverse cannot be found.

inverse_of is necessary for proper behaviour in the following scenario: Assuming two models A and B, with instances a and b, and code like a.b.a.b.a would to be identical to a. We tend to expect this but if inverses are not properly set we will get 3 different A instances created.

In my opinion, adding a warning when ActiveRecord fails to find an association would be sufficient, but more adventurous contributors might want to add a rake task or some sort of validation that can be run in a CI build.

Of course, @ronen and I think that using the schema_associations gem is the best since that defines the inverses automatically, but for the rest of us who do not...

[lowjoel]

As a matter of reporting, we can see what the bullet gem does, and do something similar.

[lowjoel]

@ronen I've started some work with schema_association_inverses (name suggestions welcome...). That requires a PR from schema_monkey, but is essentially just raising an error when an inverse cannot be detected for now.

I absolutely cannot write specs for it because my development machines are all running Windows -- any way to get the build matrix working on it?

[ronen]

@lowjoel cool

schema_association_inverses (name suggestions welcome...)

maybe something like...?

• schema_plus_association_inverses? make it be part of "the family"?
• schema_plus_assoicaions -- be a general holder for basic association cleanups?

just raising an error when an inverse cannot be detected for now.

Is that likely to break things for core rails or 3rd-party gems that rely on lack of inverse failing quietly? Should this add support for something like inverse: :none to suppress the error if you know that there isn't one?

That requires a PR from schema_monkey

Looks good, go ahead and issue the PR! BTW see also SchemaPlus/schema_monkey#1

I absolutely cannot write specs for it because my development machines are all running Windows -- any way to get the build matrix working on it?

Hmm, I've never done any ruby development on Windows. I have no idea what it would take to get schema_dev rspec working on Windows. Is that something you can try to do? Unfortunately, AFAIK travis-ci doesn't yet have support for Windows, which will make it difficult for me to maintain it.

...or can you run linux in a VirtualBox on Windows and do your development in that...?

[lowjoel]

I think schema_plus_association_inverses should be good. schema_plus_associations sounds like the schema_associations gem.

Is that likely to break things for core rails or 3rd-party gems that rely on lack of inverse failing quietly? Should this add support for something like inverse: :none to suppress the error if you know that there isn't one?

I think we can preliminarily implement throwing an error first, once we have all the boilerplate up, we can start implementing nicer error displays (like how Bullet does it)

I have no idea what it would take to get schema_dev rspec working on Windows. Is that something you can try to do?

I could, but I'm afraid I'll have to push the stuff out without specs for now. RVM on Windows is a royal pain. I have a VM for things like these, but it's annoying to maintain separate environments... I'll try to see if I can get the test infra working on Windows then.

Unfortunately, AFAIK travis-ci doesn't yet have support for Windows, which will make it difficult for me to maintain it.

Yup. As a whole Ruby on Windows is fine. What works on Linux tends to work just fine on Windows, unless a native extension is involved...

[ronen]

I have no idea what it would take to get schema_dev rspec working on Windows. Is that something you can try to do?

I could, but I'm afraid I'll have to push the stuff out without specs for now. RVM on Windows is a royal pain.

Actually you can run tests without schema_dev and without RVM (or rbenv or chruby). You just need to do

$ BUNDLE_GEMFILE=gemfiles/rails-4.2/Gemfile.postgresql bundle exec rspec

...or whatever the equivalent is in windows?

I've been thinking of adding an option to schema_dev so that if the current ruby version is the same as the matrix version, it'll skip the RVM/rbenv/chruby stuff and just issue the above for each item in the matrix.

[lowjoel]

@ronen sorry I've been too busy lately. I just upgraded my gem to use the schema_plus_core gem (since that's where all the hooks are defined).

I'm going to give a shot at making the tests run.

[lowjoel]

@ronen So to this end, I've submitted SchemaPlus/schema_plus_core#1 to implement the hook needed for this gem. I've manually set up schema_dev in lowjoel/schema_plus_association_inverses@d5fdff3b2c and got the first test up and running.

Let me know what you think.

[lowjoel]

Guys, I've released lowjoel/schema_plus_association_inverses@5574f33