[Experimental], JSON polymorphism, and the JSON source generator don't play well together

[stephentoub]

You have a base type that you serialize/deserialize with JsonSerializer.
You also use the JSON source generator.
Now you add a new derived type from that base class and you mark the derived type as [Experimental].

Anyone using your base class, even if they don't use the derived experimental type, now gets compilation errors stemming from the source generator emitting use of the derived [Experimental] type.

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-text-json, @gregsdennis
See info in area-owners.md if you want to be subscribed.

[eiriktsarpalis]

Seems like it belongs in the same class of issues as #101738 or #76829? I wouldn't say this is an issue specific to polymorphism, it's a matter of the source generator walking a type graph and hitting a node that emits warnings. One possible solution we were debating in the past is letting the source generator emit user-defined suppressions, which would at least address any warnings that the generator induces.

[stephentoub]

Have we considered just having the source generator emit #pragma warning disable at the top of every file?

[eiriktsarpalis]

We could. We've been disabling them on an individual diagnostic id basis on account of potential source generator bugs going unnoticed, but I suppose one could argue that such warnings only matter if they translate to observable bugs when executing the code.

[stephentoub]

For the regex source generator, we now use individual suppressions when the source generator is built as debug and use the suppress-everything approach when it's built for release. I don't think there's anything meaningful about such warnings that a developer can do any more than they could for other bugs in the source generator; in release, they would seem to only serve as blocking noise. I think of them like asserts.

[eiriktsarpalis]

Sounds good. We should probably do the same with CB.