Problems w/ Android 4.x

[brianegan]

Hello! Trying to get Injekt going with a test project of mine, and running into trouble. It works great on Lollipop+, but on Android 4.x I'm running into trouble.

The error I'm seeing:

java.lang.RuntimeException: Unable to instantiate application com.brianegan.bansa.counter.Application: uy.kohesive.injekt.api.InjektionException: No registered instance or factory for type com.brianegan.bansa.Store<com.brianegan.bansa.counter.ApplicationState, com.brianegan.bansa.Action>
    at android.app.LoadedApk.makeApplication(LoadedApk.java:526)
    at android.app.ActivityThread.handleBindApplication(ActivityThread.java:4390)
    at android.app.ActivityThread.access$1500(ActivityThread.java:139)
    at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1260)
    at android.os.Handler.dispatchMessage(Handler.java:102)
    at android.os.Looper.loop(Looper.java:136)
    at android.app.ActivityThread.main(ActivityThread.java:5105)
    at java.lang.reflect.Method.invokeNative(Native Method)
    at java.lang.reflect.Method.invoke(Method.java:515)
    at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:792)
    at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:608)
    at dalvik.system.NativeStart.main(Native Method)
Caused by: uy.kohesive.injekt.api.InjektionException: No registered instance or factory for type com.brianegan.bansa.Store<com.brianegan.bansa.counter.ApplicationState, com.brianegan.bansa.Action>
    at uy.kohesive.injekt.registry.default.DefaultRegistrar.getInstance(DefaultRegistrar.kt:99)
    at uy.kohesive.injekt.api.InjektScope.getInstance(Scope.kt)
    at com.brianegan.bansa.counter.Application.<init>(Application.kt:30)
    at java.lang.Class.newInstanceImpl(Native Method)
    at java.lang.Class.newInstance(Class.java:1208)
    at android.app.Instrumentation.newApplication(Instrumentation.java:990)
    at android.app.Instrumentation.newApplication(Instrumentation.java:975)
    at android.app.LoadedApk.makeApplication(LoadedApk.java:521)
    ... 11 more

My code can be found here: https://github.com/brianegan/bansa/blob/master/examples/counter/src/main/kotlin/com/brianegan/bansa/counter/Application.kt

I also tried Injekting a module in onCreate instead of using a companion object, with no luck. Feeling like a bit of a n00b here though, trying to figure out what I'm doing wrong! Any advice would be much appreciated :)

[apatrida]

Hi, first a note on your code:

 val store = Injekt.get(fullType<Store<ApplicationState, Any>>())

should simply be a delegate when used as a property:

 val store: Store<ApplicationState, Any> by injectLazy() // or injectValue()

Or can be assigned:

 val store: Store<ApplicationState, Any> = Injekt.get()

and in the injekt registration:

 addSingleton(
                    fullType<Store<ApplicationState, Any>>(),
                    createStore(ApplicationState(), counterReducer));

can be:

  addSingleton<Store<ApplicationState, Any>>(createStore(ApplicationState(), counterReducer))

Or if the return type of createStore is already Store<ApplicationState, Any> then just:

 addSingleton(createStore(ApplicationState(), counterReducer))

No need to use fullType anywhere. Reified generics will take care of this. If you cannot get the IDE to show the correct extension function, then import:

import uy.kohesive.injekt.api.*
import uy.kohesive.injekt.*

temporarily until you have everything in use and happy.

This shouldn't be your problem though unless the newer version of Android is causing the instance to be created without loading the reference to the companion object. With some magic in their VM they could do this, although evil it is possible.

If you only have one application instance anyway, just change it to a member as well:

val injektMain = object : InjektMain() {
        override fun InjektRegistrar.registerInjectables() {
            // ...
        }
    }

It is a small change and will make sure all registrations are done since members are instantiated and initialized in the order they appear top to bottom. Let me know if this helps.

[brianegan]

Great, thanks for the tips! I tried a couple of similar patterns, but not this exact one. I'll give it a shot tonight :)

[apatrida]

How did this work out?

[brianegan]

Hey there! Sorry about the delay -- busy times from my side. Thanks so much again for the help. I made some time tonight to play around with it. Here my findings:

Good call on the code cleanup -- addSingleton and by injectLazy() are really cool improvements.

With regards to Android 4.x, I tried a few things.

Attempt 1 - InjektMain() becomes a field

I tried your suggestion of assigning the InjektMain object to a field in the Application class. Unfortunately, I ran into the same error as described in my original post. You can see this code here:

https://github.com/brianegan/bansa/blob/injekt-android-4.x-fix-1/examples/counter/src/main/kotlin/com/brianegan/bansa/counter/Application.kt

Attempt 2 - Move val store to onCreate

A slight variation on Attempt #1 -- I've simply moved the val store from a class field into the onCreate function to delay the access even further. This also didn't work. Same error.

https://github.com/brianegan/bansa/blob/injekt-android-4.x-fix-2/examples/counter/src/main/kotlin/com/brianegan/bansa/counter/Application.kt

Attempt 3 - importModule

A combination of #1, #2 and now I've introduced a StoreModule : InjektModule class and tried injecting that. This didn't work either :(

https://github.com/brianegan/bansa/blob/injekt-android-4.x-fix-3/examples/counter/src/main/kotlin/com/brianegan/bansa/counter/Application.kt

Attempt 4 - Don't inject in the application class

This isn't my favorite solution as someone might need to both register and then use a module inside the application init phase, but I wanted to give it a shot. I simply move the injection out of the Application class and into the RootActivity. Unfortunately, this also didn't work :(

https://github.com/brianegan/bansa/blob/injekt-android-4.x-fix-4/examples/counter/src/main/kotlin/com/brianegan/bansa/counter/Application.kt

https://github.com/brianegan/bansa/blob/injekt-android-4.x-fix-4/examples/counter/src/main/kotlin/com/brianegan/bansa/counter/RootActivity.kt

Attempt 5 - No Injekt (Control)

As a sanity check, I removed Injekt and the app worked on Android 4.x.

https://github.com/brianegan/bansa/blob/injekt-android-4.x-fix-5/examples/counter/src/main/kotlin/com/brianegan/bansa/counter/RootActivity.kt

Thanks again for helping me track this down! This is such a cool project, I hope we can get it working on Android 4.x.

Hopefully these branches will provide some helpful code samples. If I get a bit more time, I'll try to dig into the root of the problem.

[brianegan]

Another attempt: Tried injecting a basic Kotlin type String into the root activity and that worked... Not sure what it is about my types that work on Android 5+ vs 4.x :(

[apatrida]

can you add logging to see the order of operations, log inside the injekt registration, log inside your onCreate method. Is OnCreate called from the ancestor constructor? or is it called later?

[brianegan]

Sure, I also stepped through the app, and it runs through in this order (which seemed correct to me):

1. injektMain field is created, runs registerInjectables
2. registerInjectables runs importModule(StoreModule())
3. StoreModule#registerInjectables runs, executing the addSingleton<Store<ApplicationState, Any>>
4. Application#onCreate is called
5. RootActivity#onCreate is called, runs Injekt.get<Store<ApplicationState, Any>>()
6. App Crash with error message above

[apatrida]

1. the error must be different because you are asking for Any instead of Action as the second generic parameter. Can I see the current error with latest attempt?

[apatrida]

also can you try it as a addSingletonFactory and pass a lambda where you cast it to Any if you are going to do that cast.

 override fun InjektRegistrar.registerInjectables() {
            addSingletonFactory { 
                  createStore(ApplicationState(), counterReducer) as Store<ApplicationState, Any>
            }
        }

The compiler should give you warnings about this cast, or say it isn't needed, or something, I'd like to see what it tells you.

Also if you debug, you can look at the injekt registry and see what type is stored in the lookup hashmap which will also give a clue.

[apatrida]

I can mock these out and see if there is a problem with generics somewhere, but I've never seen one.

[brianegan]

Ah, sorry -- that error message above is slightly out of date as this library I'm working on has moved from Action to Any. The error is essentially the same (this is after moving all injection out of the app and into the activity):

java.lang.RuntimeException: Unable to start activity ComponentInfo{com.brianegan.bansa.counter/com.brianegan.bansa.counter.RootActivity}: uy.kohesive.injekt.api.InjektionException: No registered instance or factory for type com.brianegan.bansa.Store<com.brianegan.bansa.counter.ApplicationState, java.lang.Object>

I'll give it a quick shot as a singletonFactory instead. Thanks again for your help debugging this!

I also think the types are correct overall, as it is working in Android 5+ -- I'm not sure if there's something unique to the 4.x environment?

[brianegan]

"Also if you debug, you can look at the injekt registry and see what type is stored in the lookup hashmap which will also give a clue."

Ah, cool -- I'll try this as well to see if it's a weird type mismatch! Thanks again.

[brianegan]

"The compiler should give you warnings about this cast, or say it isn't needed, or something, I'd like to see what it tells you."

Compiler: "No cast needed"

[apatrida]

I'm writing a test case now. Back in a bit with results, maybe something happens when two generics are present.

[brianegan]

Cool, Thanks!

[apatrida]

This test passes:

package uy.kohesive.injekt

import org.junit.Test
import uy.kohesive.injekt.Injekt
import uy.kohesive.injekt.InjektMain
import uy.kohesive.injekt.api.*
import java.math.BigDecimal
import java.util.concurrent.atomic.AtomicInteger
import kotlin.test.assertEquals
import kotlin.test.fail

class TestGithub33 {
    data class Store<T1, T2>(val state: T1, val obj: T2)
    data class ApplicationState(val id: String)
    data class ObjType1(val id: String)
    data class ObjType2(val id: String)
    data class OtherState(val id: String)

    companion object : InjektMain() {
        override fun InjektRegistrar.registerInjectables() {
             addSingleton(Store(ApplicationState("as1"), ObjType1("ot1")))
             addSingleton(Store(ApplicationState("as2"), ObjType2("ot2")))
             addSingleton(Store(OtherState("os1"), ObjType1("ot1")))
             addSingletonFactory<Store<ApplicationState, Any>> { Store(ApplicationState("as3"), "howdy") }
        }
    }

    @Test
    fun testWithGenerics() {
        val check1: Store<ApplicationState, ObjType1> = Injekt.get()
        assertEquals(Store(ApplicationState("as1"), ObjType1("ot1")), check1)

        val check2: Store<ApplicationState, ObjType2> = Injekt.get()
        assertEquals(Store(ApplicationState("as2"), ObjType2("ot2")), check2)

        val check3: Store<OtherState, ObjType1> = Injekt.get()
        assertEquals(Store(OtherState("os1"), ObjType1("ot1")), check3)

        val check4: Store<ApplicationState, Any> = Injekt.get()
        assertEquals(Store<ApplicationState, Any>(ApplicationState("as3"), "howdy"), check4)
    }

    @Test
    fun testDoesNotExistWithGenericType() {
        try {
            val check: Store<OtherState, Any> = Injekt.get()
            fail("Should not exist, this type")
        }
        catch (ex: Throwable) {
            // pass
        }
    }
}

using inject 1.14.x (actually is 1.15.0 with kotlin 1.0.1 but no other changes and reverting Kotlin doesn't change it)

[brianegan]

Ah, interesting. Here's a screenshot of the registrar. Looked correct to me, does this look fishy at all to you?

[apatrida]

yes, this is really odd. can we see the type value passed into the 2nd layer of the Injekt.get() call, after it creates and internal type key.

[apatrida]

if the key looks the same, and it isn't being found in the hashMap, it is something to do with hashcode() and equals() for the Type objects.

[apatrida]

it is a map of Map<Type, lambdaType> that would be failing to compare in the map.

   private val factories = ConcurrentHashMap<Type, () -> Any>()
    private val keyedFactories = ConcurrentHashMap<Type, (Any) -> Any>()

[apatrida]

this method in the DefaultRegistry.kt is being called:

    @Suppress("UNCHECKED_CAST")
    override fun <R: Any> getInstance(forType: Type): R {
        val factory = factories.getByKey(forType) ?: throw InjektionException("No registered instance or factory for type ${forType}")
        return factory.invoke() as R
    }

so if its type matches the type you see in the Map and it is not found, it'll be ugly issue.

[brianegan]

[brianegan]

I may be wrong, but they look like the same type, no? Sorry if that's a tough one to tackle :(

[apatrida]

They do appear the same. But they are no longer comparing as equal or hashcode is no longer the same for the two Type objects.

[apatrida]

maybe a bug in the new Android, I'm asking someone else to look at this from Reflection team of Kotlin itself. He is in Russia so not likely tonight.

[brianegan]

It actually works fine in the newer android (Android Lollipop+, aka 5+). It is broken on the 4.x range though.

[apatrida]

ah, they maybe had a bug they fixed.

[brianegan]

I'm thinking so... Anyhow, thanks so much for the help! Been a fun github-issue debugging session :)

[apatrida]

There could be a workaround of creating a different object type to hold the values, but that means handling 4-5 possible forms of the generics we could have and the whatever stack of values they might contain.

[apatrida]

Ah, one thing that could break equals and hashCode is if these classes came from different classloaders. One step used one class loader, the other step a different. maybe.

[apatrida]

if for the Type, you drill into it and see somewhere in there is a reference to the classloader.

[apatrida]

another work around for you is to put the injekt main code into OnCreate as well and use it within there, any factories it sets up will be available after to anything. That "instance" doesn't do anything since the registry itself is a singleton Kotlin object ... only if you create your own Injekt instances do you have to keep a reference around. You can just Injekt.addSingleton as well within the OnCreate.

you would surely be on the same classloader then.

[brianegan]

Classloader inspection:

[brianegan]

Also, I tried putting it in onCreate with no luck :(

[apatrida]

no clue at this point. We aren't doing anything weird. Unless Type docs say we shouldn't trust its hashcode/equals (maybe it does, I never checked)

[brianegan]

I wonder if something like a TypeToken might be an alternative? http://docs.guava-libraries.googlecode.com/git/javadoc/com/google/common/reflect/TypeToken.html

[apatrida]

The work around would be custom equals/hashcode for types and come up with something. maybe the string version of the Type would be a valid key. But then would not be correct for multiple classloaders. But not sure that is a very useful case.

TypeToken like Guava is what we do to get the Type, just like they do. It isn't any different.

[apatrida]

The problem is using Type as a hash map key. And it not working for older Android

[brianegan]

Ah, ok... for some reason I thought the Guava TypeToken had a slightly different equals and hashcode impl.

[brianegan]

"The problem is using Type as a hash map key. And it not working for older Android"

Yah for sure, was thinking something like TypeToken (which I mistakenly had a slightly different equals and hashcode than Type) would act in place of the Type being used now for the hash map key. Or, as you said, implementing something custom yourself

[brianegan]

Well, interesting getting to the root of it at least! Need to get to sleep from my end, but once again, I really appreciate the help getting to the bottom of it :) Fun journey!

[apatrida]

TypeToken delegates hashcode and equals to Type. one quick change is to use Type.toString and see if it helps. But, there is no docs for toString for Type to say that it should generate anything in particular. so just as it might work, it might not in other environments. The interface Type and ParameterizedType does not say anything about hashcode and equals.

[apatrida]

I could see if I get a Kotlin KType proxy that is reliable or if that needs reflection jar.

[apatrida]

ok, will see what Alex says tomorrow and check in on this again.

[apatrida]

one other approach is to store everything with two maps, one by Type as key and other as Type.toString() as Key and if either hits, it works. Covers at least ONE of them working correctly.

[brianegan]

Hey Hey :) Just a curious ping: Hear anything from your contact Alex? No biggies either way, just curious about the problem!

[apatrida]

He agreed with what I said before,. Equals and hash code broken for Type class or due to classloader differences. It can be fixed by wrapping Type with something reliable for comparison.

[apatrida]

Is this still an issue?