#675 seems to show that more is required for Rails 7.1 support. Any ideas on the failing tests there?

#675 seems to show that more is required for Rails 7.1 support. Any ideas on the failing tests there?

Having a quick peek at the errors I see a lot of

undefined method `attribute' for #<Rpush::Client::Redis::Webpush::App:

I would guess that instead of thing.attribute(:name, "value") it's now obsolete and you need to refactor to thing.attributes(name: "value")
But I can't pinpoint in the code where this originates from right now.

In #675, dealing with upstream Rails breakage by ignoring 7.1.4 (assuming the fix which broke it will be in 7.1.5), there are no test failures.

I'd appreciate if you would test again at that version of rpush and if the problem still exists, help me understand why it's not causing any test failures or write a test which fails without this fix.

I'd love to see this fixed ASAP. Can I help with anything?

I'm actually already running Rpush (v7.0.1) on Rails 7.1.3.4 for a while now, I was not aware it was not supported. We are now noticing a lot of GCM DeprecatedApi errors and want to start using FCM with Rails 7.1 ASAP.

@WoutDev You can help by testing #675 to confirm if this PR is actually required or not.

Hey @WoutDev @benlangfeld I had a look at running the tests locally (WSL2) but unfortunately I ran out of time to get it to work.

Feel free to merge this PR into your more generic Rails 7.1 PR and close this one @benlangfeld 🇳🇱

@sjieg I need to figure out how to reproduce whatever failure you saw in a test to serve as justification for this change, TDD style.

My hypothesis is that this is actually a railties thing and Rpush does not use it, hence no failing tests.

The default of 'nil' is defined in railties:
https://github.com/rails/rails/blob/main/railties/lib/rails/application/configuration.rb#L289

I don't think this project uses railties, so if the default is not set then perhaps this default (YAML) is used instead:
https://github.com/rails/rails/blob/main/activerecord/lib/active_record/attribute_methods/serialization.rb#L20

And that would explain the passing tests.

The documentation mentioning the default of 'nil' linked in the original issue also hints at this:

[...] The default value depends on the config.load_defaults target version:

As load_defaults is a railties thing.

Makes sense, thanks @WoutDev. I wonder if we should simply set the default to nil in our test suite to simulate Rails' behaviour.

I included a variation of this in #675 and that is now on the master branch.