Requirements for stabilisation?

[shadowcat-mst]

So, a number of components in Paws have something like this -

https://metacpan.org/source/JLMARTIN/Paws-0.36/lib/Paws/S3.pm#L2

Is there a list anywhere of what's required to get to the point where that warning can go away? If not, how would we go about getting to a point where there was such a list?

[pplu]

There are three services marked with warnings: S3, Route53 and Glacier. They basically pertain to the RestXML set of services. These services are a bit more complex than the others, each one having their special corner cases, and maybe the code even needs to be specialized for each service to get everything working correctly.

Having a comprehensive set of tests to verify if anything breaks would be a must (note: there are already some tests for stuff that was getting reported) I've revisited open issues that would need looking at before considering each one of these stable.

• S3

  • Handle S3 301 Moved responses #106 (Nice to have)
  • S3 PutObject URI encodes key? #111
  • URI's are being double encoded, first by _call_uri, then again by URI::Template #221
  • S3::ListObjectVersions does not support a prefix #171
  • S3 DeleteObjects Method missing 'Content-MD5' Header #149
  • CreateMultipartUpload generates invalid request #92

• Glacier

• Full Glacier support #30

• Route53

• Paws::Route53::ChangeResourceRecordSets problem... #204

• Paws::Route53::ListResourceRecordSets -- HostedZoneId not being passed through to request #128

[castaway]

Hi @pplu I've visited each one of the items listed above (and in some cases left comments), I also made a plan for a set of tests to use to verify fixes for each one, which we'll be working on implementing next. I see the tests generally use mocking for any checks against data sent to the APIs, we will use more of that.

Have you looked at / considered any end to end testing? I discovered https://github.com/localstack/localstack and wondered if we could use it, in a special subset of tests that only run if localstack is installed (indicated by an environment variable or similar) - any thoughts? Alternately we could run the existing set of tests against it with some tweaks based on env vars, and then maybe against AWS itself too, if the tester wants to.

[pplu]

I also made a plan for a set of tests to use to verify fixes for each one, which we'll be working on implementing next. I see the tests generally use mocking for any checks against data sent to the APIs, we will use more of that.

Great! If you want some help around the test suite or have doubts about how to test a specific case, don't doubt in contacting me.

Have you looked at / considered any end to end testing? I discovered https://github.com/localstack/localstack and wondered if we could use it, in a special subset of tests that only run if localstack is installed (indicated by an environment variable or similar) - any thoughts? Alternately we could run the existing set of tests against it with some tweaks based on env vars, and then maybe against AWS itself too, if the tester wants to.

I'd love end to end testing to happen. Most of the test suite is derived from real calls and real responses, so you could say that the tooling is in place. There is a pluggable caller called Paws::Net::MockCaller that accepts two modes: RECORD, where it actually calls the services, but records the input parameters and the response from the backend service. When in REPLAY mode it only accepts the same calling sequence, with the same parameters for each call, and instead of doing the HTTP call, it reads the recorded HTTP responses (so all the response handling logic is invoked). If you want to develop test suites against localstack I'd be happy to have them.

[castaway]

Having just gotten a bit confused with it - it seems if you "record" with credentials set to Test::CustomCredentials, it doesn't send creds that work, at least not for me, I get "The AWS Access Key Id you provided does not exist in our records.".. it records fine if I comment out that setting.

Is it supposed to work with them set? Do we need the CustomCredentials, since it seems to be the REPLAY/RECORD setting that tells it whether we are calling the live backend or not.

[pplu]

The MockCaller is supposed to be used with valid AWS credentials when in record mode (just using the default credential provider, developing with creds in ~/.aws/credentials or in ENV vars. In Replay mode the credential provider doesn't matter (the TestCredentials are useful in test so that Paws need anything special in the environment)

[pplu]

@castaway @piratefinn @shadowcat-mst : Tell me if you feel some of the branches can be merged for an upcoming release (as I want to merge them to release/0.39 as soon as possible). I'm kind of lost inside all the branches, so I ask you to please mention me what branches you think can be merged when you have them "ready" (I'm not actively trying to know if your branches are ready for merge or not).

[piratefinn]

@pplu After some discussion with @castaway, we decided #265 and #261 are good to pull. I'm closing 266 for now until we have tinkered with other elements more.