We attempt to do a special encoding trick when serializing
`additional_owners` under an account: when updating a value, we actually
send the update parameters up as an integer-indexed hash rather than an
array. So instead of this:
field[]=item1&field[]=item2&field[]=item3
We send this:
field[0]=item1&field[1]=item2&field[2]=item3
The trouble is that this had previously been built into the core library
as the default handling for all arrays. Because of this, it was
impossible to resize a non-`additional_owners` array as described in
more detail in #340.
This patch special cases `additional_owners` and brings sane behavior
back to normal arrays along with a test suite so that we try to build
some better guarantees around both the general and non-general cases.
Unfortunately usage of `#update_attributes` had rolled over from a time
where `#update_attributes_with_options` was still in use and `opts` were
being passed in as an optional argument which had the result of further
nesting the hash internally (i.e. `:opts => { :opts => ... } }`).
This patch fixes that problem, adds a regression test to prevent it from
reappearing, and banishes the unused `#update_attributes_with_options`.
Fixes#334.
I added this for a regression suite so that 1.8.7 could pass its tests,
but unfortunately this caused a regression in the way that parameters
are encoded for arrays of hashes. This patch reverts the change and adds
tests to detect a future regression.
(And 1.8.7 is expected to fail on this initial commit.)
When additional filters were provided for pagination like an expansion
or a predicate, they would not propagate to any call after the first.
This patch addresses that issue by storing all filters and moving them
to any new page objects being created.
Fixes#331.
Adds a special helper to `StripeObject` that helps a developer to
determine whether or not an object is deleted.
As described originally in #257, this is a bit of a special case because
a non-deleted object does not respond with `deleted` as part of its
representation, so while a deleted object would have this accessor
available automatically, non-deleted objects would not. This made use of
the SDK awkward because the presence of the method was not guaranteed.
Fixes#257 (again, heh).
This dials down the safety of `StripeObject`'s `#update_attributes`
method so that it allows properties to be assigned that it doesn't yet
know about. We're doing this for a few reasons:
1. To reflect the current behavior of accessors (i.e. `obj.name = ...`)
through `method_missing`.
2. To allow `#update_attributes` to assign properties on new projects
that don't yet know their schema from an API call.
Fixes#324.
As discussed in #325, this deprecates the public visibility of
`#refresh_from` (by renaming it). It also adds some deprecation
infrastructure to produce warnings when it's used.
Replaces my original attempt in #319 in a way that doesn't depend on
`URI.encode_www_form` which doesn't exist in 1.8.7. This should
hopefully get us the best of all worlds.
Caveats around use of `+` instead of `%20` as detailed in #319 still
apply.
Fixes#286.
Pulls the test suite out of #319 so that we can get some coverage around
parameter encoding. This should prevent any recurrence of #318.
Also includes a little bit of refactoring.
Usage on a top-level collection:
```
Stripe::Customer.list.auto_paging_each do |customer|
puts customer
end
```
Usage on a subcollection:
``` ruby
customer.invoices.auto_paging_each do |invoice|
puts invoice
end
```
We've also renamed `#all` to `#list` to prevent confusion ("all" implies
that all resources are being returned, and in Stripe's paginated API
this was not the case). An alias has been provided for backward API
compatibility.
Fixes#167.
Replaces #211 and #248.
As detailed in issue #119, we've somewhat unfortunately been allowing
object attributes to be passed in during a #save because we mix any
arguments directly into the serialized hash (it seems that this was
originally intended to be used more for meta parameters that go to the
request).
As also noted in #119, this use causes problems when certain types of
parameters (like subobjects) are used. We're now left in the somewhat
awkward position of either:
1. Removing this functionality on #save and breaking what may be
behavior that people depend on.
2. Fully support this mass assignment.
This patch takes the second path by extracting a new #update_attributes
method and using it from #save. It's still far from a perfect approach
because keys that have the same name as certain options (e.g. `req_url`)
are not going to work, but it should capture the behavior that most
people want.
Fixes#119.
This patch adds question marks helpers (e.g. #paid?) for any values in a
StripeObject that are a boolean. This is fairly idiomatic Ruby in that
it behaves similarly to other libraries like ActiveRecord.
Note that a caveat here is that nullable booleans will not get a helper
added for them if their current value is null. For this reason, we
should eventually prefer to derive these methods from some sort of
programmatic API manifest.
Replaces #257 and #274.
This makes ListObject behave a little more like an Array in that it gets
an #empty? helper. This should fit pretty well with the Enumerable
methods that it already has.
Replaces #193.
This pulls the `Enumerable` mixin into `ListObject`. There is some
question in pulls like #167 as to the future of `ListObject` and how it
might change when pagination is introduced, but because we're unlikely
to make any backward incompatible changes to the API, it's likely that
`ListObject` will continue to represent a page of data that's been
extracted from the API. Given that assumption, pulling `Enumerable` in
should be relatively safe.
Fixes#227.
When constructing an object using .construct_from treat keys that are
strings the same as keys which are symbols by calling Util's
symbolize_names on an input hash. This makes guarantees around
consistency a little better.
Fixes#151.
Modifies the behavior of an update so that it can create an object if it
was new. This allows an "upsert"-like functionality where the SDK will
tend to do the right/expected thing.
This is a tad janky in that it checks for the existence of a class-level
`.create` method which is expected to come from a different module, but
most modules in the project are already fairly deeply intertwined and
co-dependent.
Fixes#165.
