@WorkerThread
    suspend fun fetchPokemonList(
        page: Int,
        onSuccess: () -> Unit,
        onError: (String) -> Unit
    ) = flow {
        var pokemons = pokemonDao.getPokemonList(page)
        if (pokemons.isEmpty()) {
            when(val result = apiClient.fetchPokemonList(page = page)) {
                is ApiResult.Success -> {
                    pokemons = result.response.results
                    pokemons.forEach { pokemon -> pokemon.page = page }
                    pokemonDao.insertPokemonList(pokemons)
                    emit(pokemons)
                    onSuccess()
                }
                is ApiResult.Failure -> onError("Error!")

            }
        } else {
            emit(pokemons)
            onSuccess()
        }
    }.flowOn(Dispatchers.IO)`

ApiClient.kt

suspend fun fetchPokemonList(
        page: Int
    ) = apiService.fetchPokemonList(
        limit = PAGING_SIZE,
        offset = page * PAGING_SIZE
    )

ApiService.kt

@GET("pokemon")
    suspend fun fetchPokemonList(
        @Query("limit") limit: Int = 20,
        @Query("offset") offset: Int = 0
    ): ApiResult<PokemonResponse, Nothing>

@Provides
    @Singleton
    fun provideRetrofit(okHttpClient: OkHttpClient): Retrofit {
        return Retrofit.Builder()
            .client(okHttpClient)
            .baseUrl(BuildConfig.BASE_URL)
            .addConverterFactory(MoshiConverterFactory.create())
            .addConverterFactory(ApiResultConverterFactory)
            .addCallAdapterFactory(ApiResultCallAdapterFactory)
            .build()
    }

    @Provides
    @Singleton
    fun provideApiService(retrofit: Retrofit): ApiService {
        return retrofit.create(ApiService::class.java)
    }

    @Provides
    @Singleton
    fun provideApiClient(apiService: ApiService): ApiClient {
        return ApiClient(apiService)
    }`

Am I missing something?

My sense was that this should Just Work™️ with other adapters, but upon further inspection we seem to be running up against a Java reflection limitation with regard to nested generic return types.

public Single<ApiResult<String, String>> foo();

Retrofit only passes on an ApiResult class and not the full ApiResult<String, String> type. Upon further digging, this seems to be coming from Java itself - the Method.getGenericReturnType() returns a Single<ApiResult>.

Not sure we can really work around this, will dig around a bit. I suspect a proper fix, if any, would require fixing in Retrofit itself.

Describe the bug

A clear and concise description of what the bug is.

When I make an API call that returns a 204 No Content and the response in null I get

kotlin.KotlinNullPointerException: Response from com.mediciland.datacollector.android.data.network.block.BlockApis.getBlock was null but response body type was declared as non-null
        at retrofit2.KotlinExtensions$await$2$2.onResponse(KotlinExtensions.kt:43)
        at com.slack.eithernet.ApiResultCallAdapterFactory$ApiResultCallAdapter$adapt$1$enqueue$1.onResponse(ApiResult.kt:245)
        at retrofit2.OkHttpCall$1.onResponse(OkHttpCall.java:161)
        at com.google.firebase.perf.network.InstrumentOkHttpEnqueueCallback.onResponse(InstrumentOkHttpEnqueueCallback.java:69)
        at okhttp3.internal.connection.RealCall$AsyncCall.run(RealCall.kt:519)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1167)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:641)
        at java.lang.Thread.run(Thread.java:764)

Requirements (place an x in each of the [ ])

• [x] I've read and understood the Contributing guidelines and have done my best effort to follow them.
• [x] I've read and agree to the Code of Conduct.
• [x] I've searched for any related issues and avoided creating a duplicate issue.

To Reproduce

Steps to reproduce the behavior: The retrofit call is defined as

  @DecodeErrorBody
  @GET("block/{blockId}")
  suspend fun get(@Path("blockId") blockId: Long): ApiResult<BlockDto?, ErrorDto>

Expected behavior

As long as the the ApiResult response type is defined as nullable I would expect to not get a NPE

Summary

Resolves issues/50.

As soon as we reach the end of nextAnnotations, we avoid an illegal array access. Since this is the last iteration, it's implied that resultType will be null.

Another potential approach would be to check if the target annotation is in the original array first, and if so, we could create a new filtered copy, but I assume this is not desired so we can keep running a single array traversal.

Requirements (place an x in each [ ])

• [X] I've read and understood the Contributing Guidelines and have done my best effort to follow them.
• [X] I've read and agree to the Code of Conduct.

Describe the bug

I tried to add the test fixtures to my gradle project:

testImplementation testFixtures("com.slack.eithernet:eithernet:1.2.0")

But my test sources don't seem to pick up the test-fixture dependencies.

When I run dependencyInsight i get:

:dependencyInsight
com.slack.eithernet:eithernet:1.2.0 FAILED
   Failures:
      - Could not resolve com.slack.eithernet:eithernet:1.2.0.
          - Unable to find a variant of com.slack.eithernet:eithernet:1.2.0 providing the requested capability com.slack.eithernet:eithernet-test-fixtures:
               - Variant apiElements provides com.slack.eithernet:eithernet:1.2.0
               - Variant runtimeElements provides com.slack.eithernet:eithernet:1.2.0
               - Variant testFixturesApiElements provides com.slack.eithernet:EitherNet-test-fixtures:1.2.0
               - Variant testFixturesRuntimeElements provides com.slack.eithernet:EitherNet-test-fixtures:1.2.0

Which, according to gradle docs seem to be cause by missing gradle module metadata.

I'm no gradle expert, so perhaps I'm doing something wrong.

Requirements (place an x in each of the [ ])**

• [x] I've read and understood the Contributing guidelines and have done my best effort to follow them.
• [x] I've read and agree to the Code of Conduct.
• [x] I've searched for any related issues and avoided creating a duplicate issue.

Reproducible in:

{project_name} version: 1.2.0

Summary

Some gentle code-gardening to replace the hand-rolled array-copy-plus-add with the standard array + item syntax.

Requirements (place an x in each [ ])

• [x] I've read and understood the Contributing Guidelines and have done my best effort to follow them.
• [x] I've read and agree to the Code of Conduct.

Summary

You can create an ApiResult with the syntax ApiResult.httpFailure(error), but there is no utility function to create an ApiResult from a success value, this PR adds this utility function.

Requirements (place an x in each [ ])

• [x] I've read and understood the Contributing Guidelines and have done my best effort to follow them.
• [x] I've read and agree to the Code of Conduct.

Does this fit in ApiFailure or HttpFailure somewhere?

Some ideas:

A. Reuse ApiFailure

In this event, upon 4xx responses we would attempt to request a delegate converter to decode the error body and report in an ApiFailure with a code.

Awkward because it combines ApiFailure and HttpFailure

B. Add an error to HttpFailure

Simple, but potentially awkward because we then add a type to HttpFailure too. If we do this, I think we should reuse the same error type as ApiError

Other questions

• How do we enable this?
  • parameter to the call adapter factory?
  • @DecodeErrorBody opt-in annotation?
• What if there are multiple error types? Some APIs return different errors for different status codes
  • Could we pass the error code via annotation in the converter lookup, similar to how we use @ResultType

Summary

Add proguard rule to keep ApiResult. Fixes #26.

Requirements (place an x in each [ ])

• [x] I've read and understood the Contributing Guidelines and have done my best effort to follow them.
• [x] I've read and agree to the Code of Conduct.

Description

At @tonalfitness, we use microservices. Many endpoints across different services use HTTP error codes with meaningful bodies. Turns out the body for a given status code might be totally different for two endpoints. Because of that, the advice provided here doesn't really work for us because the status code itself is not enough to know what type to return.

Here's what would be great:

interface MyApi {
    @DecodeErrorBody
    @POST("/some/path")
    suspend fun someEndpoint(): ApiResult<SuccessModel, GenericErrorModel>
}

sealed class GenericErrorModel {
    @HttpError(400, 403)
    @JsonClass(generateAdapter = true)
    data class SpecificErrorModel1(
        @Json(name = "intProperty") val intProperty: Int,
    ) : GenericErrorModel()

    @HttpError(500)
    @JsonClass(generateAdapter = true)
    data class SpecificErrorModel2(
        @Json(name = "stringArrayProperty") val stringProperty: String,
        @Json(name = "booleanProperty") val booleanProperty: Boolean,
    ) : GenericErrorModel()
}

The approach above is possible by extracting the response type inside a custom converter factory and then checking for a potential sealed error type. The actual response status code could be matched against annotated subtypes to figure out what model to deserialize the response to.

Correctness—like forgetting to annotate a subtype, or having more than one subtype bound to the same status code—could be enforced at compile time if either code generation or reflection is used (with a lint rule in this case).

Not sure if this is a too specific use case of if it's something that also makes sense to a broader audience that uses EitherNet.

Requirements (place an x in each of the [ ])

• [x] I've read and understood the Contributing guidelines and have done my best effort to follow them.
• [x] I've read and agree to the Code of Conduct.
• [x] I've searched for any related issues and avoided creating a duplicate issue.

Describe the bug

When the target annotation is missing, we get an exception trying to access an array position that is out of bounds at line 84.

https://github.com/slackhq/EitherNet/blob/5d1c8ec064ab4e9d645b394e3bd07364bbe8001a/src/main/java/com/slack/eithernet/Annotations.kt#L73-L95

The reason is because nextAnnotations is smaller than the original array by 1 and the loop code is optimistic about finding the target annotation in the original array. When this doesn't happen, the executed code path merely tries to copy the original array into nextAnnotations, overflowing it at the end.

Requirements (place an x in each of the [ ])**

• [x] I've read and understood the Contributing guidelines and have done my best effort to follow them.
• [x] I've read and agree to the Code of Conduct.
• [x] I've searched for any related issues and avoided creating a duplicate issue.

To Reproduce

Steps to reproduce the behavior:

1. Write your own Converter.Factory and plug it to Retrofit after ApiResultConverterFactory;
2. Call annotations.statusCode() inside responseBodyConverter();
3. Get the exception thrown.

Expected behavior

Instead of getting an exception, we should either get:

• null to (theseAnnotations as Array<Annotation>) which means we couldn't single out the target annotation (preferable IMO);
• Simply null.

Reproducible in:

Project version: 1.2.1

OS version(s): Any

Additional context

If this is recognized as a bug, I'm happy to push a PR for review.

Describe the bug

The GMM for v1.3.0 does not contains invalid capabilities entries for the testFixtures variants so Gradle fails to resolve the dependency even if you are not relying on the test fixtures.

Capabilities in v1.2.1

// curl -sL https://repo.maven.apache.org/maven2/com/slack/eithernet/eithernet/1.2.1/eithernet-1.2.1.module | jq '.variants[].capabilities? | select( . != null )'
[
  {
    "group": "com.slack.eithernet",
    "name": "eithernet-test-fixtures",
    "version": "1.2.1"
  }
]
[
  {
    "group": "com.slack.eithernet",
    "name": "eithernet-test-fixtures",
    "version": "1.2.1"
  }
]

Capabilities in v1.3.0

// curl -sL https://repo.maven.apache.org/maven2/com/slack/eithernet/eithernet/1.3.0/eithernet-1.3.0.module | jq '.variants[].capabilities? | select( . != null )'
[
  {
    "group": "",
    "name": "eithernet-test-fixtures",
    "version": "unspecified"
  }
]
[
  {
    "group": "",
    "name": "eithernet-test-fixtures",
    "version": "unspecified"
  }
]
[
  {
    "group": "",
    "name": "eithernet-test-fixtures",
    "version": "unspecified"
  }
]

Build failure log

* What went wrong:
Execution failed for task ':api:compileDebugKotlin'.
> Error while evaluating property 'filteredArgumentsMap' of task ':api:compileDebugKotlin'.
   > Could not resolve all files for configuration ':api:debugCompileClasspath'.
      > Could not resolve com.slack.eithernet:eithernet:1.3.0.
        Required by:
            project :api
         > Could not resolve com.slack.eithernet:eithernet:1.3.0.
            > Could not parse module metadata https://repo.maven.apache.org/maven2/com/slack/eithernet/eithernet/1.3.0/eithernet-1.3.0.module
               > missing 'group' at /variants[2]/capabilities[0]

Requirements

• [x] I've read and understood the Contributing guidelines and have done my best effort to follow them.
• [x] I've read and agree to the Code of Conduct.
• [x] I've searched for any related issues and avoided creating a duplicate issue.

To Reproduce

Add the EitherNet 1.3.0 dependency to a project and attempt to build it

Expected behavior

Gradle resolves the dependency and builds the project

Reproducible in:

Project version: 1.3.0

Additional context

I noticed this in a regular Renovate PR to my app so you can repro it there but it has a rather massive dependency tree so it's probably easier to use a fresh skeleton project .