* fix obsolute configuration in generated .rubocop.yml
* fix style of `expect` with block in generated ruby client's test code
* update sample of ruby client
* add test cases to cover different collection format
* add space params to retrofit 1.x
* add space params to retrofit 2.x
* rename url to localVarUrl
* fix exception in haskell servant
* Introduce GeneratorSettings + cleanup
GeneratorSettings is an immutable settings object, intended to limit the
manipulation of generator settings.
To move to GeneratorSettings, lots of modification was done to
CodegenConfigurator. The goal here is that CodegenConfigurator
would create the contextual information required to initiate a
generator run:
* GeneratorSettings
* Workflow related settings
* Configuring "system" GeneratorProperties (ThreadLocal properties)
* Deserializing from file to config object
* Input spec document (OpenAPI, intending to target others)
ClientOpts was generally unused, and the few places it was being used
have been updated to pass the properties to
codegen.additionalProperties.
* Add sanity to system properties
The -D argument for the generate command is an application argument
which is easily confused for Java System Properties. This isn't the
case, as setting values here doesn't update the configuration in
System.getProperties().
This adds a warning and deprecation to that option, as defining these
values as system properties will also continue to work as expected. This
makes the -D application argument redundant and confusing.
* Contextualize generator/workflow settings
This splits settings relevant to generator configuration (the what) and
workflow configuration (the how) in an attempt to make configuration
easier to conceptualize.
* Update Gradle task w/ CodegenConfigurator setters
* Remove -D usage in scripts
* Add -p option for additional properties
* Regnerate samples
`URI.encode` is obsolete. `CGI.escape`, `URI.encode_www_form` or
`URI.encode_www_form_component` are recommended instead.
https://ruby-doc.org/stdlib-2.6/libdoc/uri/rdoc/URI/Escape.html#method-i-escape
URI.encode has different behaviour to CGI.escape:
```ruby
URI.encode('hello/world?test%string')
=> "hello/world?test%25string"
CGI.escape('hello/world?test%string')
=> "hello%2Fworld%3Ftest%25string"
```
I recently raised pull request #3039
201cbdce29cc6cdbbbe9efcb1afb250a05bc2ffd
That pull request escapes path items at insertion.
Before either pull request, the path item 'hello?world' would go into
the URL as 'hello?world'. That behaviour was insecure as if an attacker
could control the path item value, they could change the URL the
application connected to.
After #3039 'hello?world' would go in as 'hello%253Fworld'. This was
safer than before, but it's still not correct.
If I'd realised at the time, I would have made it correct at the time.
What this pull request does is make it go in as 'hello%35world', which
is correct.
ApiClient::build_request_url was URI.encoding the whole path.
This wasn't protecting against all undesirable characters in the path
items, but was escaping % characters a 2nd time which was unhelpful.
I have additionally removed URI.encode from Configuration::base_url as I
can't see any benefit it could be bringing.
There is no justification for it in the commit where it was originally
added: 47c8597d36a9bc0983ba5c40e2489bb094f9f076
