Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add API resource instance methods to StripeClient #954

Closed
wants to merge 3 commits into from

Conversation

joeltaylor
Copy link
Contributor

This change introduces convenience methods to access API resources
through StripeClient for per-client configuration. The instance client
can be configured with the same global Stripe configurations. As a
result, an instance of StripeClient is able to override or fallback
to the global configuration that was present at the time of
initialization.

Here's an example:

Stripe::Customer.list() == StripeClient.new.customers.list()

The primary workhorse for this feature is a new module called
Stripe::ClientAPIOperations that defines instance methods on
StripeClient when it is included. A ClientProxy is used to send any
method calls to an API resource with the instantiated client injected.
There are a few noteworthy aspects of this approach:

  • Many resources are namespaced, which introduces a unique challenge
    when it comes to method chaining calls (e.g.
    client.issuing.authorizations). In order to handle those cases, we
    create a ClientProxy object for the root namespace (e.g., "issuing")
    and define all resource methods (e.g. "authorizations") at once to
    avoid re-defining the proxy object when there are multiple resources
    per namespace.

  • Sigma deviates from other namespaced API resources and does not have
    an OBJECT_NAME separated by a period. We account for that nuance
    directly.

  • method_missing is substantially slower than direct calls. Therefore,
    methods are defined where possible but method_missing is still used
    at the last step when delegating resource methods to the actual
    resource.

  • Each API resource is pluralized to align with the conventions of other
    Stripe libraries (e.g. Node and PHP). The pluralization itself is
    quite naive but can easily be switched out for something more advanced
    once the need arises.

  • Each API resource spec was converted to use instance based
    methods and was done to ensure adequate test coverage. Since this
    entire feature is built on proxying methods, testing via the client
    implicitly tests the original implementation for "free".

This is a clean-up of #921 as I felt that it didn't have reliable enough test coverage.

Comment on lines +47 to +56
def update_args_with_client!(method, args)
opts_pos = @resource.method(method).parameters.index(%i[opt opts])

return unless opts_pos

opts = opts_pos >= args.length ? {} : args[opts_pos]

normalized_opts = Stripe::Util.normalize_opts(opts)
args[opts_pos] = { client: @client }.merge(normalized_opts)
end
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The primary regressions that I found during the first PR all had to do with nuances when it came time to merge the options. I was able to identify the scenarios and lock them in under test: https://github.com/stripe/stripe-ruby/pull/954/files#diff-50b82fc5b424d8d4c21c2ddd6c67485e40fbc76fdf5cc1e4320949a4ad20fb62R56

@@ -45,12 +45,8 @@ def resource_url
end

# @override To make id optional
def self.retrieve(id = ARGUMENT_NOT_PROVIDED, opts = {})
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ARGUMENT_NOT_PROVIDED doesn't play very well with how we merge options. The args object in method_missing is an array and merging options when an ID is omitted results in [nil, {}] being sent to the resource, which isn't desirable.

However, I wasn't able to re-create any of the scenarios that warranted the addition of it in c269e9b and believe it's safe for removal.

lib/stripe/stripe_client.rb Outdated Show resolved Hide resolved
lib/stripe/stripe_client.rb Outdated Show resolved Hide resolved
Comment on lines 265 to 297
def store_last_response(object_id, resp)
return unless last_response_has_key?(object_id)

self.class.current_thread_context.last_responses[object_id] = resp
end

def last_response_has_key?(object_id)
self.class.current_thread_context.last_responses&.key?(object_id)
end
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Rubocop didn't like the complexity of execute_request so I extracted these out.

lib/stripe/stripe_configuration.rb Outdated Show resolved Hide resolved
Comment on lines 103 to 110
def max_network_retry_delay=(val)
@max_network_retry_delay = val.to_i
end

def initial_network_retry_delay=(val)
@initial_network_retry_delay = val.to_i
end
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Figured these are worth exposing since everything else is.

@joeltaylor
Copy link
Contributor Author

joeltaylor commented Nov 16, 2020

@brandur-stripe this is the alternative approach I mentioned. The most noteworthy change is the testing strategy. Considering that testing the client implicitly tests the original implementation, I thought it was a viable path but I'm certainly open to feedback. Either way, I was able to find some edge cases and refactor some of the original logic with a high level of confidence.

There are a few outstanding questions:

  • I didn't remove capability from the dynamic method generation so someone can technically do StripeClient.new.capabilitys.list(). It'll throw the same error as Stripe::Capability.list(), but figured it's worth noting even though they're accessed via Stripe::Account. I'm happy to remove it and any others that fall into the same category.
  • There are two funny methods that may be worth modifying:
    • StripeClient.new.three_d_secures
    • StripeClient.new.invoiceitems
  • CI is failing due to Ruby 2.3, which has reached EOL. I'm happy to open a PR to drop 2.3 (and 2.4) support unless you all prefer to support EOL versions.
  • This feature is likely worth a README update but I'm punting until the end on that in case any significant changes come along.

@joeltaylor
Copy link
Contributor Author

I wanted to see if I could consume the OpenAPI spec to automate the testing, but hit a few roadblocks. I'm not seeing a super clear route outside of duplicating the API resource specs for both usages. I may also be able to do something dynamic and/or create a test helper. Curious what your thoughts are?

Not super opinionated here, although it'd be nice to land with something that exercises both cases, and hopefully without too much metaprogramming. I think duplicated tests would probably be fine, although that's a lot of copying and pasting. The other option is that we don't go too crazy with the per-resource per-action testing, because if you look at what most of these are doing, they're not really testing anything all that thoroughly anyway.

Incidentally, we got a pull request recently in #966 wherein users were basically looking to have something close to per-client configuration. We could bring that in, but your more comprehensive approach here would be the much better solution IMO.

Even if we take a little further on getting all the per-client resources working correctly, would you mind if we brought in a slice of this PR that makes StripeClient configurable? If you're busy, I can make that PR.

@brandur-stripe Moving the conversation over from #921. I'm a fan of extracting out the StripeClient configuration portion from this PR. Considering how large this PR is, that seems like a good seam for more manageable chunks. I'll follow-up with a PR this week.

Regarding testing, I'm curious what your thoughts are one #954 (comment)?

@stripe stripe deleted a comment from travisbarronsanabria Mar 15, 2021
@brandur-stripe
Copy link
Contributor

brandur-stripe commented Mar 15, 2021

@brandur-stripe Moving the conversation over from #921. I'm a fan of extracting out the StripeClient configuration portion from this PR. Considering how large this PR is, that seems like a good seam for more manageable chunks. I'll follow-up with a PR this week.

Awesome! And thanks for the quick reply on this. I also took a shot at this on Friday and although I didn't get it 100% working before the day ended, it was relatively straightforward. It was all your code though, and it'd be better if it all came in under your name, so will wait for your PR.

Testing answers below.

@brandur-stripe this is the alternative approach I mentioned. The most noteworthy change is the testing strategy. Considering that testing the client implicitly tests the original implementation, I thought it was a viable path but I'm certainly open to feedback. Either way, I was able to find some edge cases and refactor some of the original logic with a high level of confidence.

@joeltaylor Nice. As long as the two usage types are relatively bound together, I think this approach is okay. The only part that strikes me as a bit a problem is that it's now hard to find usage examples of the original "style" in the test suite. IMO though, we should try to bias towards getting this merged and figure that out later.

There are a few outstanding questions:

  • I didn't remove capability from the dynamic method generation so someone can technically do StripeClient.new.capabilitys.list(). It'll throw the same error as Stripe::Capability.list(), but figured it's worth noting even though they're accessed via Stripe::Account. I'm happy to remove it and any others that fall into the same category.

Hmm, would the code to have it not be at the top level be a little gnarly? (Like, did you not add it because no special casing is needed right now?) I think it's fine as long as we're erroring, but hmm, I guess it would be slightly better if the invalid methods were not available at all.

Separately: what do you think about potentially removing the pluralization of resources? It's not as "Ruby-ish", but the upsides are that we don't have to either (1) deal with pluralization oddities like "capabilitys", or (2) introduce a pluralization engine that can map the special exceptions. It might be simpler overall.

  • There are two funny methods that may be worth modifying:

Ah, good to call these out.

  • StripeClient.new.three_d_secures

Let's leave this one as is.

  • StripeClient.new.invoiceitems

Hm, I know this is based off our weird endpoint naming, but IMO we should remap this to invoice_items as long as it's relatively easy to do.

  • CI is failing due to Ruby 2.3, which has reached EOL. I'm happy to open a PR to drop 2.3 (and 2.4) support unless you all prefer to support EOL versions.

We don't formally support EOLed Ruby, but we do sort of have a policy of trying to keep them on life support as long as possible. It kind of looks like from the failing build that it might be relatively easy to put in a patch that keeps 2.3 working (probably just an extra check on nil somewhere). What do you think?

  • This feature is likely worth a README update but I'm punting until the end on that in case any significant changes come along.

Yeah, good call. That's the type of thing that I'm more than willing to do as well.

Copy link
Contributor

@ob-stripe ob-stripe left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Left some comments. Thanks for working on this @joeltaylor!

Comment on lines 10 to 13
opts[:client] ||= params[:client] || StripeClient.active_client
opts[:api_base] ||= opts[:client].config.connect_base

params.delete(:client)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

OAuth methods are a bit weird, but I think the client should always be passed in opts, not in params.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Great call out. I'll see if I can get that behavior under test and will make the change.

client = params[:client] || StripeClient.active_client
base = opts[:connect_base] || client.config.connect_base
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same here.

Comment on lines +10 to +14
}.merge(api_object_names_to_classes)
end

# business objects
def self.api_object_names_to_classes
{
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

OOC, why is this change necessary? list is an API object name 🤔

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ob-stripe it was done so I could get a list of API resources to dynamically define the methods here: https://github.com/stripe/stripe-ruby/pull/954/files#diff-992f231b0add61aeb46ab11fb68529936809a99a74ed9f1c1755a4b01f57fb58R63

Certainly open to other suggestions!

lib/stripe/stripe_client.rb Outdated Show resolved Hide resolved
@system_profiler = SystemProfiler.new
@last_request_metrics = nil
@config_overides = config_overides
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In e.g. stripe-php we support two different syntaxes for initializing clients:

  1. passing a configuration hash (what you're doing already)
  2. passing just the API key as a string

Would you mind adding support for the latter? I think it should be as simple as:

@config_overides = config_overrides.is_a?(String) ? {api_key: config_overrides} : config_overrides

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's a great suggestion and certainly improves usability. I'll get it added!

@joeltaylor
Copy link
Contributor Author

@brandur-stripe I'll get this PR up to date with the main branch shortly. Wanted to give a 👍 to your answers and suggestions in #954 (comment)

To recap:

  • I'll remove the pluralization so we can get rid of the oddities
  • Will bias towards merging over fiddling with the testing strategy
  • Will be sure to support 2.3 (if I recall correctly it is just another nil check)

I think we're in the home stretch 😄

@joeltaylor joeltaylor force-pushed the add-client-instance-api branch 2 times, most recently from 55c4b2c to e3ed0a2 Compare April 6, 2021 02:14
@@ -11,7 +11,7 @@ def list(filters = {}, opts = {})

# set filters so that we can fetch the same limit, expansions, and
# predicates when accessing the next and previous pages
obj.filters = filters.dup
obj.filters = filters.dup unless filters.nil?
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's possible for filters to be nil due to how we delegate methods to the API resources.

joeltaylor and others added 3 commits April 5, 2021 20:05
This change introduces convenience methods to access API resources
through `StripeClient` for per-client configuration. The instance client
can be configured with the same global Stripe configurations. As a
result, an instance of `StripeClient` is able to override or fallback
to the global configuration that was present at the time of
initialization.

Here's an example:

```ruby
Stripe::Customer.list() == StripeClient.new.customer.list()
```

The primary workhorse for this feature is a new module called
`Stripe::ClientAPIOperations` that defines instance methods on
`StripeClient` when it is included. A `ClientProxy` is used to send any
method calls to an API resource with the instantiated client injected.
There are a few noteworthy aspects of this approach:

- Many resources are namespaced, which introduces a unique challenge
  when it comes to method chaining calls (e.g.
  client.issuing.authorizations).  In order to handle those cases, we
  create a `ClientProxy` object for the root namespace (e.g., "issuing")
  and define all resource methods (e.g. "authorizations") at once to
  avoid re-defining the proxy object when there are multiple resources
  per namespace.

- Sigma deviates from other namespaced API resources and does not have
  an `OBJECT_NAME` separated by a period. We account for that nuance
  directly.

- `method_missing` is substantially slower than direct calls. Therefore,
  methods are defined where possible but `method_missing` is still used
  at the last step when delegating resource methods to the actual
  resource.

- Each API resource spec was converted to use instance based
  methods and was done to ensure adequate test coverage. Since this
  entire feature is built on proxying methods, testing via the client
  implicitly tests the original implementation for "free".
It's possible for `filters` to be `nil`, which causes Ruby 2.3 to raise
an error: `TypeError: can't dup NilClass`
Comment on lines +67 to +69
# Update `invoiceitems` to match snake case convention
invoice_item_class = api_resources.delete("invoiceitem")
api_resources["invoice_item"] = invoice_item_class
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@brandur-stripe Here's the update to turn invoiceitem into invoice_item that you previously mentioned.

@ob-stripe
Copy link
Contributor

@richardm-stripe Reassigned this to you since Brandur is no longer around.

@joeltaylor
Copy link
Contributor Author

@richardm-stripe I'm going to close this PR out given it's been awhile and there are a few merge conflicts. Happy to revisit in the future if this is still a desired direction.

@joeltaylor joeltaylor closed this Jun 21, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

5 participants