As discussed in https://github.com/OpenAPITools/openapi-generator/pull/7415#discussion_r1113274416, it seems unlikely the code was correct.
server_operation_index is a hash table. In Ruby, `hash[key]` will return the value associated with `key`. If key is absent, `nil` is returned. Because that is sometimes undesirable, there is also `hash.fetch(key)`, which raises an error if the key is absent. It also allows you to specify a default to fall back on: `hash.fetch(key, default)` will return `default` if the key is absent.
So, since not all users will specify a 'server per operation' (or at least: I'm not), the old code would usually set `index` to the `server_index`, which is initialized to 0. The subsequent `if index == nil` will usually return false (`0 != nil` in Ruby), after which the `server_url` call on line 177 constructs the url based on the `server_operation_variables` and `operation_server_settings`, assuming we are dealing with the case where a server per operation is configured. The case where the url should be constructed from `scheme`, `host`, etc. is only called if either `server_index` is explicitly set to `nil` or the key `operation` is explicitly associated with the value `nil` in the `server_operation_index` hash table, both of which seem inappropriate.
* Enable the ruby client to support refreshing access tokens
- The client can now be configured with an access token getter proc
- The proc overrides the the static access token if it is set
* Run generators
* Add args to reserved words
* arg and args to _arg and _args in templates
* Corrections
* Test added
* Corrections
* Use arg and args as defined properties
* Removed unnecessary assertion
* Suggested change
* fix for issue #13722: send body for application/x-www-form-urlencoded data
* fix python test_application_x_www_form_urlencoded_serialization
* x-www-form-urlencoded data needs to be percent encoded
* add verification endpoint test for x-www-form-urlencoded data
Co-authored-by: David Chaiken <dchaiken@pinterest.com>
* I feel the issue is due to the creation of self._var_name_to_model_instances while doing the deserialization of the data.
Earlier the Python SDK code was using get_var_name_to_model_instances function which was adding var name to model instances that contain it. So <class 'openapi_client.model.stream_options_all_of'> will not part of mapping in self._var_name_to_model_instances for variable name stream_options.
Now with the latest Python SDK code following is the way through which var_name_to_model_instances is created:
for prop_name in model_args:
if prop_name not in discarded_args:
var_name_to_model_instances[prop_name] = [self] + composed_instances
Now as we can see that the var_name_to_model_instances is populated with self and composed_instance which will also contain stream_options_all_of as a composed instance and there will be no check that if stream_options is present in composed_instances or not.
As there is no attribute_mapping found for stream_options in stream_options_all_of, the type for stream_options will be treated as dict for mapping stream_options_all_of as mentioned by @Chekov2k.
So what I suggest is the following code:
for prop_name in model_args:
if prop_name not in discarded_args:
var_name_to_model_instances[prop_name] = [self] + list(
filter(
lambda x: prop_name in x.openapi_types, composed_instances))
This way we can check if the property name is present in that composed instance or not. If it's okay for @spacether I can raise a PR for this.
* [get_item_all_of_bug]
Added samples, test cases to validate all_of schema.
* [getiem_all_of_bug]
Updated docs and samples.
* [getiem_all_of_bug]
Updated test cases, docs and samples.
