does anybody know what the `SerialDescriptor` gene...

# serialization

[thana]

does anybody know what the SerialDescriptor generated for a @Serializable class depends on?

[sandwwraith]

what do you mean?

[thana]

well i have 2 identical* classes in different projects dependening on two identical* classes in those very projects. the serializers are generated with the exact same versions. yet the json strings differ.
Using the debugger i can see the SerialDescriptor is different - so my guess was that might be a good starting point for further debugging

* identical except on packages and file structure

[sandwwraith]

SerialDescriptor includes fully-qualified name of the class (i.e. with the package). However, it becomes visible only when using polymorphism. Besides this, it contains the order in which properties in class are declared and some additional metadata from annotations. What do you mean exactly by 'json strings differ' ?

[thana]

in one case it's completly broken, in the other it's not. i was trying to create some sourcecode for a bugreport. but i cannot reproduce it outside the real project 😞

[thana]

i was already guessing it would depend on the order of declared members but... the order is 100% identical - yet in the real project there are even members missing!
I have no clue how to hunt this bug down 😞

[thana]

i think i will duplicate the project and then remove code until the bug disappears 🤔

[r4zzz4k]

* identical except on packages

SerialDescriptor includes fully-qualified name of the class (i.e. with the package)

@thana, did you try to change package name in your reproducer project to match the original one? I understand you've most probably did, just didn't mention this here, but still.

[thana]

Yep in the end i aligned the package names, top.
Lucky that didnt trigger the error 😁

[thana]

I will Stop for today. Im Tor tired already anyway

[hallvard]

... and, of course, you did a clean rebuild, or even maybe a File --> Invalidate Caches/Restart ?

[thana]

i always did a clean build

[thana]

and it looks like its a source set related problem

[thana]

@sandwwraith i could boil it down to https://github.com/mschirmacher/SerializationTest

[thana]

the branch master contains a broken version working contains the working version. it only differs inthe sourceset that contains the class DomainEvent

[thana]

yet ./gradlew jvmTest will pass in the working branch but fail in master

[thana]

would you agree here that this is a bug?

[sandwwraith]

Can this be related to your case? https://github.com/Kotlin/kotlinx.serialization/issues/378

[thana]

Yes, this is somehow related.
i say "somehow" because versions 1.3.21/0.10.0 do not seem to reproduce the problem. i this fact known, too?