duck-type values responding to #to_hash or #to_ary

[notEthan]

I work with JSON data wrapped in classes which behave like Hash and Array, which respond to #to_hash / #to_ary to support implicit conversion (as described in https://docs.ruby-lang.org/en/master/doc/implicit_conversion_rdoc.html )

this PR lets jmespath be used with these implicitly convertible objects. it is in draft because I have updated no tests or documentation, and may be incomplete in other ways. if an active maintainer can tell me it might be merged, I will bring it to completion.

[mullermp]

Thanks for opening up a draft PR. I think what you're trying to do makes sense. What is the experience without any changes? - i.e. for your JSON data wrapped classes, are you able to call to_a or to_h on them and pass those raw converted values into jmespath?

[alextwoods]

Overall I think this is a reasonable change, but see comment re: Struct.

[notEthan]

What is the experience without any changes? - i.e. for your JSON data wrapped classes, are you able to call to_a or to_h on them and pass those raw converted values into jmespath?

I can pass the JSON that my classes wrap to jmespath, yes, but then lose the functionality that is added by the wrapper I have implemented. my goal is to search the wrapped document and get wrapped children out as the search result(s).

the wrapping I work with is associating JSON data with JSON schemas (per the specification at https://json-schema.org/) describing those data. so if I search within a document described by a schema, the resulting children get returned associated with subschemas that describe them. the library implementing this is here, in case that is of any interest: https://github.com/notEthan/jsi

if this is merged and released I'll add a method something like #jmespath_search(expression) to the wrapper class, to help with searching documents.

[notEthan]

I think I have the code in a good state for review, and have added a couple tests

[mullermp]

Thanks. We'll review this shortly.

[alextwoods]

I've added CI back to this project and merged those changes in. I've also ported the compliance tests into the implicit conversion test cases - there were a few failures in functions which I've fixed.

There are a few other failures in the compliance tests w/ implicit conversion that I'm not positive how to address (they may be issues in my spec code as well) - could you take a look at those cases?

[notEthan]

thanks for the additions to the tests, that was definitely needed. I added and fixed a couple more to_arys, and added a Util.as_json for recursive comparison (this made redundant the unwrap method you'd added to the tests, so I removed that). added some test cases to the compliance json to cover code changes.
also rebased against main.

[alextwoods]

LGTM - great contribution and thanks for the extra testing!

[notEthan]

glad to hear. thanks very much for your reviews and additions. I am quite happy to find active maintainers here, the project looked very quiet. also, aside, this is among the best-written libraries I've worked with or contributed to.
looking forward to integrating support when this is merged and gets released.

[alextwoods]

@notEthan - Thanks again for a great contribution. I've released this as version 1.5.0.

[brianalexander]

For people searching google:

Any developers using the AWS SDK for Ruby v2 that has a dependency on jmespath updating to 1.5.0 or greater, Aws::EC2::Instance.block_device_mappings returns an array of InstanceBlockDeviceMapping that are now converted from Structs to Hashes implicitly.

To correct errors, update dot operator to bracket notation for value access.

Old way:

block_devices = Aws::EC2::Instance.new(instance_id, client: EC2_CLIENT_OBJECT).block_device_mappings
data_block_device = block_devices.find { |bd| bd.device_name == "/dev/xvdb" }
data_block_device.ebs.volume_id

New way:

block_devices = Aws::EC2::Instance.new(instance_id, client: EC2_CLIENT_OBJECT).block_device_mappings
data_block_device = block_devices.find { |bd| bd[:device_name] == "/dev/xvdb" }
data_block_device[:ebs][:volume_id]

[mullermp]

That seems wrong. It should be a struct. We will investigate.

[alextwoods]

Thanks for reporting that - I agree with @mullermp - it should be a struct there. I have confirmed this does NOT impact V3 of the SDK (note: V2 of the SDK is no longer supported).