Issue with the latest upgrade 0.9.1

[kapso]

After the upgrade render json: { ok: true } is returning this {"sessions"=>[["ok", true]]} where sessions is the name of the controller. Its weird.

adding a to_json fixes it render json: { ok: true }.to_json ??

[masda70]

I confirm this, happens after upgrading to 0.9.1 from 0.9.0.

EDIT: After some debugging, this behavior was most likely introduced after the following change
90343ce

The change affects serializing behavior for Enumerables, though its effect on Hash might be undesirable.

[shwoodard]

Hi, I am experiencing the same problem,

render json: @this.returns_a_hash, root: false # hash {a: b}; output: [['a', 'b']]

Is this intended new behavior? Is it fixed in master? Can we bump priority on this and release? Locking to 0.9.0 in the interim.

Another question to consider is, if the json being rendered is a Hash and not an object that includes ActiveModel::Serialization or other applicable ActiveModel modules, should ams intervene at all?

[steveklabnik]

No, this is a bug.

[shwoodard]

Lemme see if I can fix :)

[shwoodard]

I created a pull request within my fork, shwoodard#1, with a failing spec. I made it from my repo to my repo instead to rails-api/ams because it doesn't work. However, I time boxed this and the fix wasn't apparent. I'm happy to make another pull request against rails-api/ams with just the failing spec, if desired.

[ashanbrown]

@andrusha pointed this out in the discussion at the end of #665 but unfortunately it appears to have been included in the v0.9.1. It's not clear to me why @andreychernih wanted this change. If the desired payload is the kind of Enumerable that wants to be rendered as an array, then maybe the payload should just define to_ary.

[mgonzalez-msights]

+1 completely broke my app

[magg]

+1

[woohoou]

+1

[msaint]

+1

[tadman]

Just discovered this problem and it pretty much trashes any render(json: ...) calls in an application. Shouldn't 0.9.1 be yanked until this is fixed?

[steveklabnik]

Yanking is always the wrong thing to do. I would like a fix, however.

[tadman]

It is a bit drastic, but this is a pretty severe bug. I've found that several people I work with have encountered it independently and have had to roll back to 0.9.0. This is included as a dependency in popular gems like ember-rails so a lot of people are probably tripping over this without even realizing why.

Republishing 0.9.0 as 0.9.2 would not be a bad idea until this can be properly resolved.

[steveklabnik]

Republishing 0.9.0 as 0.9.2 would not be a bad idea until this can be properly resolved.

I like this idea. I'll do this later today. :(

[steveklabnik]

I have re-released 0.9.0 as 0.9.2 until we can get a fix.

[ashanbrown]

Bummer about all this. Could you let us know what the release plan is? Does everyone need to resubmit their pull requests against v0.9.0 or should the owners of their respective features submit fixes for model-less json serialization bug and the namespace lookup bug? Thanks.

[steveklabnik]

Keep submitting to master. I just made https://github.com/rails-api/active_model_serializers/tree/0-9-2, which as you can see is just a version bump commit off of https://github.com/rails-api/active_model_serializers/releases/tag/v0.9.0

[tadman]

Thanks for fixing that. Helps keep things up to date.

[zamith]

With 0.9.2 I get an error that no serializer is used. I have this in the controller:

render json: Member.all

And I have a serializers/member_serializer.rb. It works perfectly fine with 0.9.1, but it just sends the entire object across, with 0.9.2.

[zamith]

Figured out that it was because of how rails v4.2.0-rc3 handles the hooking of renderers. It is not using _render_option_json anymore, it is still using _render_with_renderer_json though.

Since AMS 0.8.x defines both it works, but in 0.9.x _render_with_renderer_json stopped being defined.

[shwoodard]

@zamith IMHO these last 2 comments are not related to #746. Maybe open a separate issue?

[zamith]

@shwoodard I thought it was related, but it turned out not to be. It's not an issue for me anymore, you can just disregard it.

[steveklabnik]

0.9.2 is 0.9.0, so you'll get the same bugs you had in it.

[zamith]

@steveklabnik The bug is only present in 0.9.0 and 0.9.2, in 0.9.1 it has been fixed in 447d969

[steveklabnik]

Exactly.

[kurko]

We lost many, many good commits (almost 4 months) doing this. 0.9.1 was good except for 1 commit (#665 ?). We had support for Rails 4.2 and a bunch of other fixes.

Could someone kindly pick the 0.9.1 tag, revert just that one commit and open a PR? I really can't do this because I don't use 0.9 :( but your work will help tons of people.

[nateberkopec]

@kurko #775 but the merge isn't clean

EDIT: #776

[kurko]

#776, not #775.

[kurko]

Could everyone here please test #776? It's basically the 0.9.1 (full of fixes) minus the problematic commit that breaks the universe. If #776 works on your apps, then it could potentially become 0.9.3.

[recurrence]

Wouldn't this be better than 0.9.2 even if there is a small regression or two? Seems to be working fine for me if this naive and completely unscientific data point helps any :) Ship It! :)

[kurko]

Released https://rubygems.org/gems/active_model_serializers/versions/0.9.3. It's basically 0.9.1 (tons of bugfixes) minus the commit that caused tons of regressions. Please test and let us know how it goes 😁