AWS invoke API ops giving java.lang.Character cannot be cast to java.lang.Number Error

[gzmask]

This PR should fix it: #70

Error example in DynamoDB calls (other APIs have the same issue):

(def ddb
  (aws/client
    {:api :dynamodb
     :credentials-provider (credentials/basic-credentials-provider ...)}))

(aws/invoke ddb {:op :ListTables})

Stacktrace:
[[clojure.lang.RT uncheckedByteCast "RT.java" 1379]
[cognitect.aws.signers$uri_encode invokeStatic "signers.clj" 30]
[cognitect.aws.signers$uri_encode doInvoke "signers.clj" 15]
[clojure.lang.RestFn invoke "RestFn.java" 423]
[cognitect.aws.signers$canonical_uri invokeStatic "signers.clj" 59]
[cognitect.aws.signers$canonical_uri invoke "signers.clj" 51]
[cognitect.aws.signers$canonical_request invokeStatic "signers.clj" 106]
[cognitect.aws.signers$canonical_request invoke "signers.clj" 103]
[cognitect.aws.signers$string_to_sign invokeStatic "signers.clj" 115]
[cognitect.aws.signers$string_to_sign invoke "signers.clj" 113]
[cognitect.aws.signers$signature invokeStatic "signers.clj" 135]
[cognitect.aws.signers$signature invoke "signers.clj" 131]
[cognitect.aws.signers$v4_sign_http_request invokeStatic "signers.clj" 157]
[cognitect.aws.signers$v4_sign_http_request doInvoke "signers.clj" 137]
[clojure.lang.RestFn invoke "RestFn.java" 470]
[cognitect.aws.signers$eval36549$fn__36550 invoke "signers.clj" 161]
[clojure.lang.MultiFn invoke "MultiFn.java" 244]
[cognitect.aws.client$send_request invokeStatic "client.clj" 69]
[cognitect.aws.client$send_request invoke "client.clj" 63]
[cognitect.aws.client.api.async$invoke$send__36444 invoke "async.clj" 67]
[cognitect.aws.retry$with_retry$fn__36062$state_machine__23274__auto____36071$fn__36073 invoke "retry.clj" 15]
[cognitect.aws.retry$with_retry$fn__36062$state_machine__23274__auto____36071 invoke "retry.clj" 15]
[clojure.core.async.impl.ioc_macros$run_state_machine invokeStatic "ioc_macros.clj" 973]
[clojure.core.async.impl.ioc_macros$run_state_machine invoke "ioc_macros.clj" 972]
[clojure.core.async.impl.ioc_macros$run_state_machine_wrapped invokeStatic "ioc_macros.clj" 977]
[clojure.core.async.impl.ioc_macros$run_state_machine_wrapped invoke "ioc_macros.clj" 975]
[cognitect.aws.retry$with_retry$fn__36062 invoke "retry.clj" 15]
[clojure.lang.AFn run "AFn.java" 22]
[java.util.concurrent.ThreadPoolExecutor runWorker "ThreadPoolExecutor.java" 1149]
[java.util.concurrent.ThreadPoolExecutor$Worker run "ThreadPoolExecutor.java" 624]
[java.lang.Thread run "Thread.java" 748]]

[dchelimsky]

I don't see this issue. What versions of cognitect.aws/api, cognitect.aws/endpoints, and cognitect.aws/dynamodb are you using?

[gzmask]

All with newest version. Did your test exercise the uri-encode function? https://github.com/cognitect-labs/aws-api/blob/master/src/cognitect/aws/signers.clj#L15
It's a simple type mismatched error on the (byte \A) bits.

[dchelimsky]

There are already numerous tests that exercise this function indirectly in https://github.com/cognitect-labs/aws-api/blob/master/test/src/cognitect/aws/signer_test.clj. Here's a direct unit test of the fn:

(require '[cognitect.aws.signers :as signers])
(require '[clojure.spec.test.alpha :as stest])

(s/fdef signers/uri-encode
  :args (s/cat :string (s/and string? not-empty) :extra-chars string?)
  :ret string?)

(stest/check `signers/uri-encode))

This produces this outcome:

({:spec
  #object[clojure.spec.alpha$fspec_impl$reify__2524 0x32c975de "clojure.spec.alpha$fspec_impl$reify__2524@32c975de"],
  :clojure.spec.test.check/ret
  {:result true, :num-tests 1000, :seed 1555691156677},
  :sym cognitect.aws.signers/uri-encode})

Please show me a failing test case I can execute.

[gzmask]

Hmmm, I tested this with (set! *unchecked-math* true) and (set! *unchecked-math* false), the latter passes while the first one failed. So this is an issue with the byte function in core really 🤔

[dchelimsky]

I ran the generative test above with both (set! *unchecked-math* true) and (set! *unchecked-math* false) and it keeps passing for me.

[dchelimsky]

I'm closing this as I'm not able to reproduce this. If you can provide a failing test case (i.e. code that I can paste into a REPL and execute and see fail) I'll be glad to take another look and reopen this if appropriate.

[gzmask]

uncheck-math is hard to test due to run time characteristics. You need to re-eval the defn to see the change in effect.

Try this:

(set! *unchecked-math* true)
(byte \A) ;; will fail
(set! *unchecked-math* false)
(byte \A) ;; will pass

[dchelimsky]

Ah, tricky. I now see the failure. Fix coming soon.

[dchelimsky]

Ended up a/ w faster implementation, too. Thanks!