Use KSP2

[eygraber]

No description provided.

[eygraber]

@evant I updated all dependencies to latest versions, but looks like google/ksp#1854 is still an issue, as well as some issues with typealiases which might be related to https://kotlinlang.slack.com/archives/C013BA8EQSE/p1716431911867909. I tested with 2.1.0-RC and both issues are present there as well.

I probably won't have much time to dedicate to this for the next few months, in case there's anyone who'd like to pick it up.

[ting-yuan]

The exception KtInvalidLifetimeOwnerAccessException: Access to invalid KtAlwaysAccessibleLifetimeToken: PSI has changed since creation, was thrown in the second round in the SymbolProcessor.finish()

invoke:79, AbstractKSDeclarationImpl$containingFile$2 (com.google.devtools.ksp.impl.symbol.kotlin)
invoke:78, AbstractKSDeclarationImpl$containingFile$2 (com.google.devtools.ksp.impl.symbol.kotlin)
lambda 'synchronized' in 'value':74, SynchronizedLazyImpl (kotlin)
getValue:69, SynchronizedLazyImpl (kotlin)
getContainingFile:78, AbstractKSDeclarationImpl (com.google.devtools.ksp.impl.symbol.kotlin)
lambda 'apply' in 'addOriginatingElement':95, KSAstProvider (me.tatarka.kotlin.ast)
addOriginatingElement:94, KSAstProvider (me.tatarka.kotlin.ast)
lambda 'forEach' in 'generate':27, KmpComponentCreateGenerator (me.tatarka.inject.compiler)
forEach:1863, KmpComponentCreateGenerator (me.tatarka.inject.compiler)
lambda 'apply' in 'generate':23, KmpComponentCreateGenerator (me.tatarka.inject.compiler)
lambda 'with' in 'generate':22, KmpComponentCreateGenerator (me.tatarka.inject.compiler)
generate:18, KmpComponentCreateGenerator (me.tatarka.inject.compiler)
lambda 'forEach' in 'generateKmpComponentCreateFiles':40, ProcessKmpComponentCreateKt (me.tatarka.inject.compiler.ksp)
forEach:216, ProcessKmpComponentCreateKt (me.tatarka.inject.compiler.ksp)
generateKmpComponentCreateFiles:39, ProcessKmpComponentCreateKt (me.tatarka.inject.compiler.ksp)
finish:113, InjectProcessor (me.tatarka.inject.compiler.ksp)
lambda 'forEach' in 'execute':586, KotlinSymbolProcessing (com.google.devtools.ksp.impl)
execute$stub_for_inlining$32:1863, KotlinSymbolProcessing (com.google.devtools.ksp.impl)
forEach:1863, KotlinSymbolProcessing (com.google.devtools.ksp.impl)
execute:586, KotlinSymbolProcessing (com.google.devtools.ksp.impl)
...

whereas the corresponding element of createNative was from the first round:

lambda 'with' in 'processKmpComponentCreate':29, ProcessKmpComponentCreateKt (me.tatarka.inject.compiler.ksp)
processKmpComponentCreate:16, ProcessKmpComponentCreateKt (me.tatarka.inject.compiler.ksp)
process$lambda$3:69, InjectProcessor (me.tatarka.inject.compiler.ksp)
invoke:-1, InjectProcessor$$Lambda/0x00007f43f947f670 (me.tatarka.inject.compiler.ksp)
calcNext:171, FilteringSequence$iterator$1 (kotlin.sequences)
hasNext:194, FilteringSequence$iterator$1 (kotlin.sequences)
toList:813, SequencesKt___SequencesKt (kotlin.sequences)
process:70, InjectProcessor (me.tatarka.inject.compiler.ksp)
invoke:552, KotlinSymbolProcessing$execute$1$1 (com.google.devtools.ksp.impl)
invoke:550, KotlinSymbolProcessing$execute$1$1 (com.google.devtools.ksp.impl)
closeFilesOnException:409, IncrementalContextBase (ksp.com.google.devtools.ksp.common)
lambda 'forEach' in 'execute':550, KotlinSymbolProcessing (com.google.devtools.ksp.impl)
forEach:1863, KotlinSymbolProcessing (com.google.devtools.ksp.impl)
execute:549, KotlinSymbolProcessing (com.google.devtools.ksp.impl)
...

Please avoid caching and using the elements across rounds.

[evant]

@eygraber ^ probably due to

kotlin-inject/kotlin-inject-compiler/ksp/src/main/kotlin/me/tatarka/inject/compiler/ksp/InjectProcessor.kt

Line 35 in 7f7dcd9

private val kmpComponentCreateFunctionsByComponentType = mutableMapOf<AstClass, MutableList<AstFunction>>()

we should change it to only store the KSName like we do for other deferred symbols.

[eygraber]

Yup, I started working on that. Not sure if I'm doing something wrong because I'm getting other errors but at least the KtInvalidLifetimeOwnerAccessException seems to have stopped 😄

[eygraber]

@evant I believe the latest commit fixes the KtInvalidLifetimeOwnerAccessException issue.

There is still an issue with typealiases (probably related to google/ksp#1921).

Most of the RoundsTests are failing because of a change in dev.zacsweers.kctfork:ksp:0.6.0 where the messages are split between a messages: String property, and a diagnosticMessages: List<DiagnosticMessage>. diagnosticMessages is used in messagesWithSeverity (which is what we call to get the error output), however the errors that we are looking for only appear in messages.

[evant]

Is it OK if kotlin-inject doesn't work with KSP1 anymore?

I've seen enough issues that I think it would be good to have overlap supporting both for a bit. What breaks here with ksp1? I don't know if it'll easy to detect which one you are using but we could add an additional option to enable it for now

[eygraber]

Actually it looks like it does might work with KSP1. The integration tests are all passing, but about half of the unit tests fail. Maybe I'm not configuring kotlin-compile-testing correctly, or maybe the tests are actually failing, it's hard to say for sure.

[eygraber]

Got some inspiration from https://github.com/amzn/kotlin-inject-anvil/blob/main/compiler-utils/src/testFixtures/kotlin/software/amazon/lastmile/kotlin/inject/anvil/Compilation.kt and figured it out 😄

The only unit tests failing now are the ones that check for a specific error format that changed from KSP1 to KSP2, so I'm thinking that it is fine.

[eygraber]

Updated the unit and integration tests to run with both KSP1 and KSP2.

[evant]

CI is blocked by google/ksp#2230

looks like it's already fixed! I'm ok with throwing a clean on ci for now until that's released though

[eygraber]

Updated, hopefully that's all that's needed.

[sahilbajaj]

@eygraber @evant Eagerly awaiting action.

[eygraber]

@evant anything still needed here (aside from getting past New Years)?

[evant]

Will take a final pass when I get a chance but don't imagine anything major blocking merge, no

[eygraber]

It looks like function injection isn't working where the typealias points to another typealias. I don't have the time to investigate now, but that wasn't working before these changes either, so I'm guessing not a blocker?

[evant]

so I'm guessing not a blocker

yeah, if it's not a regression it's not a blocker

[eygraber]

@evant with KSP 2.0.0 enabling KSP2 by default, can we get a release that includes this?