bundle exec bin/bench_regression a5eaf6cd7a7fed42d9e64777753a1762e187eadc 1033b711c7d7c231bb5b832e7dfe7f99389f22c4 --pattern bm_caching
["perf/only_calc_associations_once", "a5eaf6cd7a7fed42d9e64777753a1762e187eadc", "1033b711c7d7c231bb5b832e7dfe7f99389f22c4", "a5eaf6c"]
"version": "0.10.0.rc5",
"rails_version": "4.2.6",
"benchmark_run[environment]": "2.2.2p95",
Note: checking out 'a5eaf6cd7a7fed42d9e64777753a1762e187eadc'.
HEAD is now at a5eaf6c... Nicer debug; compare caching by serializer, grouped by caching on/off
caching on: caching serializers: gc off 783.6956866669746/ips; 1355 objects
caching on: non-caching serializers: gc off 798.8629770532652/ips; 1257 objects
caching off: caching serializers: gc off 682.3661326140281/ips; 1355 objects
caching off: non-caching serializers: gc off 721.2175067555897/ips; 1257 objects
HEAD is now at 1033b71... Merge pull request #1638 from bf4/caching_redux
caching on: caching serializers: gc off 570.6905948477781/ips; 1803 objects
caching on: non-caching serializers: gc off 822.8418206976623/ips; 1257 objects
caching off: caching serializers: gc off 523.4174806572001/ips; 1803 objects
caching off: non-caching serializers: gc off 747.6026493097758/ips; 1257 objects
failing test
use try for when the assoc_data is possibly nil
rubocop test/action_controller/json_api/deserialization_test.rb -a
attempt to work on rails-master
account for rails/master having instead of nil for assoc_data
added changelog
This will prevent objects PORO objects that don't have updated_at defined, from throwing an error.
Not as big a deal now that PORO objects can inherit ActiveModelSerializers::Model, but still necessary if it's not inherited for whatever reason.
Add the Adapter type to the cache key.
This prevents incorrect results when the same object is serialized with different adapters.
BF:
Cherry-pick of
040a97b9e9
which was a squash of
f89ed71058
from pr 1346
It seems that fecthing from memory_store returns a reference to the
object and not a copy. Since the Attributes adapter applies #merge! on
the Hash that is returned from the memory_store, the value in the cache
is also modified.
Status quo in test app:
In Rails
ActionController::Base.cache_store = :memory_store
and then AMS railtie does:
ActiveModelSerializers.config.cache_store = config.action_controller.cache_store
then, in the Railtie
1. ActiveSupport.on_load(:action_controller) fires
- ActiveModelSerializers.config.cache_store #=> nil
- ActionController::Base.cache_store #=> #<ActiveSupport::Cache::FileStore:0x007fe319256760...]
2. After set_configs fires
- ActiveModelSerializers.config.cache_store #+> #<ActiveSupport::Cache::FileStore:0x007fe319256760 ,
3. Tests pass, but notice that we're using the FileStore, not memory store
When we change the config to the test app:
ActionController::Base.cache_store = :memory_store
config = Rails.configuration
config.action_controller.cache_store = :memory_store
then, in the Railtie:
1. ActiveSupport.on_load(:action_controller) fires
- ActiveModelSerializers.config.cache_store #=> nil
- ActionController::Base.cache_store #=> #ActiveSupport::Cache::MemoryStore entries=0, size=0, options={}>]
2. After set_configs fires
- ActiveModelSerializers.config.cache_store #=> :memory_store
3. And we get a lot of failures:
NoMethodError: undefined method `fetch' for :memory_store:Symbol
So, we see that when we set the ActionController::Base.cache_store
directly in our test app, we could set
ActiveModelSerializers.config.cache_store in the :action_controller load
hook, but that would never use the Rails config.
To fix the Rails config, we change the config to the test app:
config = Rails.configuration
config.action_controller.cache_store = :memory_store
and then AMS railtie does:
ActiveModelSerializers.config.cache_store = ActiveSupport::Cache.lookup_store(config.action_controller.cache_store
ActiveSupport.on_load(:action_controller) do
::ActiveModelSerializers.config.cache_store = cache_store
end
then
1. After set_configs fires
- ActiveModelSerializers.config.cache_store #=> <#ActiveSupport::Cache::MemoryStore, object_id 70207113611740
2. ActiveSupport.on_load(:action_controller) fires
- ActionController::Base.cache_store #=> <#ActiveSupport::Cache::MemoryStore, object_id 70207106279660
- ActiveModelSerializers.config.cache_store #=> <#ActiveSupport::Cache::MemoryStore, object_id 70207106279660
(notice the object_id changed)
3. And we get a failure:
1) Failure:
ActiveModelSerializers::CacheTest#test_associations_cache_when_updated
[active_model_serializers/test/cache_test.rb:141]:
--- expected
+++ actual
@@ -1 +1 @@
-{:id=>"post", :title=>"New Post", :body=>"Body"}
+{:id=>"post", :title=>"New Post", :body=>"Body", :comments=>[{:id=>2, :body=>"ZOMG A NEW COMMENT"}], :blog=>{:id=>999, :name=>"Custom blog"}, :author=>{:id=>"author", :name=>"Joao M. D. Moura"}}
If we take out the on_load(:action_controller) hook, we get a ton of
failures. So clearly, our code expects the controller cache to be the
same as the serializer cache.
So, we make sure we use an on_load(:action_controller) hook that runs
after set_configs
And look at the test and see it is filled with direct calls to ActionController::Base.cache_store
assert_equal(new_comment_serializer.attributes, ActionController::Base.cache_store.fetch(new_comment.cache_key))
assert_equal(@post_serializer.attributes, ActionController::Base.cache_store.fetch(@post.cache_key))
But that's not a problem in this case, since they're the same object.
For now, let's remove the :memory_store setting and use the default FileStore
The JSON API adapater dasherizes every key, but the deserializer left the keys
unaltered. Thus, the client had to send underscored keys in the request body in
order for Rails to properly match sent values to model attributes.
This commit adds automatic key transformation on deserialization. Per default the
deserializer transforms the keys to underscore, but this behaviour can also be
changed by including `key_transform` in the deserializer options.
NoMethodError is current_user is nil, so nil.admin?
NameError is a superclass of NoMethodError (which Rails 4.0 won't allow)
and means current_user might not be defined
- Setup dummy app files in `test/dummy`
- Setup dummy test server `bin/serve_dummy
- Note: Serializer caching can be completely disabled by passing in
`CACHE_ON=off bin/serve_dummy start` since Serializer#_cache is only
set at boot.
- run with
- ./bin/bench
- `bin/bench` etc adapted from ruby-bench-suite
- target files are `test/dummy/bm_*.rb`. Just add another to run it.
- benchmark cache/no cache
- remove rake dependency that loads unnecessary files
- remove git gem dependency
- Running over revisions to be added in subsequent PR
Adding a benchmak test structure to help contributors to keep track
of how their PR will impact overall performance.
It enables developers to create test inside of tests/benchmark.
This implementation adds a rake task: ```rake benchmark``` that checkout
one commit before, run the test of tests/benchmark, then mover back to
the last commit and run it again. By comparing the benchmark results between
both commits the contributor will notice if and how much his contribution
will impact overall performance.
- The removed classes and modules were added back with deprecation
warning and deprecation test were added for them.
- One test was renamed because it contained `__`.
- Some tests were refactored.
- The ActiveModelSerializers::Deserialization module is now called
Adapter instead of ActiveModelSerializers::Adapter.
- The changelog was added for #1535
Idea per remear (Ben Mills) in the slack:
https://amserializers.slack.com/archives/general/p1455140474000171
remear:
just so i understand, the adapter in `render json: resource, status: 422, adapter: 'json_api/error',
serializer: ActiveModel::Serializer::ErrorSerializer` is a different one than, say what i’ve
specified in a base serializer with `ActiveModel::Serializer.config.adapter = :json_api`. correct?
and a followup question of, why not same adapter but different serializer?
me:
With the way the code is written now, it might be possible to not require a special jsonapi adapter.
However, the behavior is pretty different from the jsonapi adapter.
this first draft of the PR had it automatically set the adapter if there were errors. since that
requires more discussion, I took a step back and made it explicit for this PR
If I were to re-use the json api adapter and remove the error one, it think the serializable hash
method would look like
```
def serializable_hash(options = nil)
return { errors: JsonApi::Error.collection_errors } if serializer.is_a?(ErrorsSerializer)
return { errors: JsonApi::Error.resource_errors(serializer) } if serializer.is_a?(ErrorSerializer)
options ||= {}
```
I suppose it could be something more duckish like
```
def serializable_hash(options = nil)
if serializer.errors? # object.errors.any? || object.any? {|o| o.errors.any? }
JsonApi::Error.new(serializer).serializable_hash
else
# etc
```
Changed the namespace in adapters and folder to active_model_serializers from active_model::serializer
Changed namespace of adapters in serializers and other folders
Moved adapter_for_test file to active_model_serializers folder and changed namespace of adapter inside the test file
Require ActiveSupport's string/inflections
We depend on string/inflections to define String#underscore.
Refactor JsonApi adapter to avoid redundant computations.
Update readme.md to link to v0.10.0.rc4
changed namespace of adapter folder testcases
Changed all namespaces of adapter under active_moder_serializers
Namespaced IncludeTree which is from serializer module, so needed to namespace it properly
Fixed wrong namsepacing of fieldset
namespace change in deserializer json_api
Fixed the namespace for collection serializer when used inside adapter, changed namespace for adapter to new namespace which I had forgotten previously
Modified logging test and adapter test cases to make the testcases pass
Changed the yardoc links,as old links are not taking to documentation pages,proper links for 0.10,0.9 and 0.8 in rubydoc
Rubocop errors are fixed by underscore naming unused variables
Moved the require of adapter to serializable resource
Remoeved adapter dependency inside serializer and added warning to Serializer::adapter method
Fixed frament cache test which is calling Serializer.adapter
Changed the name of lookup_adapter_from_config to configured_adapter
Changed the docs which will show the new namespace of adapters
Rubocop fix
The ActiveModel::Serializer.type method now accepts symbol as paremeter:
class AuthorSerializer < ActiveModel::Serializer
type :profile
end
The test file for the type was also refactored.
Fix the "Calling deprecated ArraySerializer... Please use
CollectionSerializer" warning that appeared when running the specs. This
happened because on of the test called ArraySerializer instead of
CollectionSerializer.
When using the relationship link with a block, calling "meta" without "href", e.g.
has_one :bio do
link :related do
meta id: 1
end
end
results in in a nil "href", e.g.
{ links: { posts: { related: { href: nil, meta: { id: 1 } } } } }.
According to JSONAPI, we should be able to use meta without href
(http://jsonapi.org/format/#document-links).
When using the relationships DSL with a block e.g.
has_one :bio do
link :self, "some_link"
end
the "data" field would be rendered with a nil value even though the bio
is not nil. This happened because the block return value was set to nil
but used as a value for the "data" field.
PR #1454 was merged with some missing fixes. All these fixes are
addressed by this commit:
- Rename ActiveModel::Serializer::Adapter::JsonApi::Association to
ActiveModel::Serializer::Adapter::JsonApi::Relationship
- Move ActiveModel::Serializer::Adapter:: JsonApi::Relationship and
ActiveModel::Serializer::Adapter::JsonApi::ResourceIdentifier to
ActiveModel::Serializer::Adapter::JsonApi::ApiObjects module
- Add unit test for
ActiveModel::Serializer::Adapter::JsonApi::Relationship
- Add unit test for
ActiveModel::Serializer::Adapter::JsonApi::ResourceIdentifier
We were not previously cloning the type setting into the dynamically
generated cached/non-cached serializers for a given fragment-cached
serializer. This led to the type generated for JsonApi having the wrong
value when fragment caching is enabled by adding either :except or :only
options to cache.
This pulls the type setting from the fragment-cached serializer forward
onto the dynamic caching classes so it is preserved in the output.
