* don't set nil when size not called
recently, #16570 worked around an issue where aggregated siblings
wasn't present, by only calling size if present.
Unfortunately, this causes a comparison of integer (new json_data
aggregated siblings count) with nil (result of the safe navigation for dig()&.size
assigned nil to previous_siblings_size, when we expected it to be
0). The initial error on the same data moved farther down the controller.
If old_json_data is present, but does not have an array in
reaction.aggregated_siblings, we want to have previous size be zero.
Remove guard clause and previous assignment
The issue was not that old_json_data was nil (it came from an existing
notification) but that there were missing keys (or rather that the
json data for some notifications didn't have a reaction key at all).
Remove the guard clause and either set to the size, or zero if there
is none. I don't understand what the guard clause was for, so replace the
safety by adding a nil safe call to dig. I don't _believe_ this is
possible but in case it was we can shorten the check here.
* Add test to cover missing json_data['reaction'] key
* Remove unneeded temporary variable
* Convert symbol hash keys to strings when calling .perform_async
Fixes a warning from Sidekiq 6.4.0+ about perform_async arguments
which are not equal when passed to perform (`JSON.parse(JSON.dump(arg))`
should equal arg).
This is a safety measure to prevent passing objects (like classes, or
model instances) rather than their representations (like a class name,
or a model's attributes hash).
This warning will be an error in sidekiq 7
* Turn warning into an error in non-production environments
* Use string keys for reaction notification and article fetched
Missed these two on the first pass
* Update example argument hashes for #enqueues_on_correct_queue
Since this calls perform_async under the hood we need to pass json
safe hashes in the test cases as well.
https://github.com/forem/forem/blob/main/spec/workers/shared_examples/enqueues_on_correct_queue.rb
* Convert keys from FollowData#to_h to string before perform_async
I'm not sure enough where else (outside of notification) to_h is being
called, so I'm converting here when building args, rather than in
FollowData#to_h, which might be my next step.
* Let FollowData#to_h return a hash with string keys
Update spec to use string keys as well.
* Make to_h return string keys for ReactionData
Like FollowData, the #to_h method is only used to call
notifications (this is used to enqueue sidekiq jobs).
* Remove a key that was in the hash
Since reaction_data calls to_h, it gets string and not symbol,
keys. Call Hash#except with a key that was actually there.
I saw the following error in Honeybadger:
```
[PROJECT_ROOT]/app/services/notifications/reactions/send.rb:61 :in `call`
old_json_data = notification.json_data
previous_siblings_size = notification.json_data["reaction"]["aggregated_siblings"].size if old_json_data
notification.json_data = json_data
[PROJECT_ROOT]/app/services/notifications/reactions/send.rb:20 :in `call`
[PROJECT_ROOT]/app/workers/notifications/new_reaction_worker.rb:16 :in `perform`
[PROJECT_ROOT]/app/models/notification.rb:84 :in `send_reaction_notification_without_delay`
[PROJECT_ROOT]/app/controllers/reactions_controller.rb:136 :in `destroy_reaction`
[PROJECT_ROOT]/app/controllers/reactions_controller.rb:168 :in `handle_existing_reaction`
[PROJECT_ROOT]/app/controllers/reactions_controller.rb:77 :in `create`
```
As I was exploring the error, I saw that we were making assumptions
about the data coercion. These are valid and somewhat stable
assumptions, but I wanted to look a little deeper into the situation.
This refactor helps consistently negotiate two situations where we're
transforming request-cycle models into lightweight data structures. It
also begins to show a path towards a generalizable "macro" for this behavior.
Related to #2122, #9534
* Make notifications reactions specs more confident about time
* Protect AnalyticsService specs if run on International Date Line West time zone
* Display local datetime to make it easier to reproduce time zone bugs
* Display date in RFC 3339
* Freeze time and confront without nanoseconds
