Tried again K2 and I'm getting the following warni...

# k2-adopters

[Edoardo Luppi]

Tried again K2 and I'm getting the following warnings.
What do they mean?

[dmitriy.novozhilov]

They indicate that some of your functions with contracts does no actually satisfy this contract

[dmitriy.novozhilov]

E.g. first and last diagnostics may be reported in such case

@OptIn(ExperimentalContracts::class)
fun test(block: () -> Unit) {
    contract { 
        callsInPlace(block, InvocationKind.EXACTLY_ONCE)
    }
    invokeLater(block)
}

fun invokeLater(block: () -> Unit) {
    // send `block` to thread pool
}

[dmitriy.novozhilov]

BTW please report an issue about incorrect warning message for the second case

[Edoardo Luppi]

Thanks!
Indeed the functions with the contracts were not inline, so good catch by K2

[dmitriy.novozhilov]

Note that it's not about inline but how you use passed lambda
This function is entirely correct

@OptIn(ExperimentalContracts::class)
fun notInlineRun(block: () -> Unit) {
    contract { 
        callsInPlace(block, InvocationKind.EXACTLY_ONCE)
    }
    block()
}

block is called inside function test, caleed exactly once and not saved in any state which potentially may invoke this lambda later

[dmitriy.novozhilov]

Also K2 assumes that every lambda passed to inline function (without noinline or crossinline modifier) automatically gets contract callsInPlace(block, InvocationKind.UNKNOWN)

[Edoardo Luppi]

What about

public suspend fun <R> ZSocket.use(block: suspend (ZSocket) -> R): R {
  contract {
    callsInPlace(block, InvocationKind.EXACTLY_ONCE)
  }

  var throwable: Throwable? = null
  return try {
    block(this) // Called here
  } catch (t: Throwable) {
    throwable = t
    throw t
  } finally {
    ...
  }
}

This is where it was marking the warning

[dmitriy.novozhilov]

Generally speaking contract here is correct
But performing control flow analysis over try/catch/finally expression is pain the ass quite hard, so there is a naive assumption which compiler makes, that exception can happen in any place of try block, including its very beginning, before call to block(this) (well, technically OutOfMemoryError actually can be thrown from anywhere). So compiler think that there is a way when function without invocation of block

[dmitriy.novozhilov]

Could you please report a ticket about this case to kotl.in/issue?
For me it looks like something which can be fixed

Contracts verification was added to K2 as a prototype and wasn't properly tested, so such reports will be quite valuable

[dmitriy.novozhilov]

For your code you can just add @Suppress("WRONG_INVOCATION_KIND") if warning annoys you

[Edoardo Luppi]

Makes sense.
I think in this specific case I could just get rid of the contracts (or add the suppression as you mentioned), but I'll report the issue tonight, no problem.

[kirillrakhman]

The warning message should already be fixed.

[Edoardo Luppi]

Thanks @kirillrakhman, 1.9.20-dev or 1.9.30-dev?

[kirillrakhman]

1.9.30-dev