filterx: perf improvement and fix race condition in readonly FilterXJson::to_json_literal()

[MrAnno]

json_object_to_json_string_ext() writes/cached into the given JSON object, so it's not thread safe.

[alltilla]

Nice catch, sorry for introducing this problem.

What would you think about making make_readonly() virtual and implementing the json specific parts there, so we don't need separate json_make_readonly() functions?

[MrAnno]

Much better, thanks.

[MrAnno]

The problem with this is that make_readonly() is transitive through get_subscript() and getattr() calls, and those are called from different threads during runtime.

So either we say we only let objects be marked readonly() at startup and we let go of the transitive behavior, or we keep everything but then I will have to either use locks or do a deep copy of the object (which defeats the whole idea of the read-only thing).

[alltilla]

I think we could replace the make_readonly calls with an assert, and make sure to take care of it in the json and otel filterx object implementations. Maybe the assert could be in the list and dict interface getters.

The json object caches all the values, so we can do a deep traversal and make everything readonly when the root object becomes readonly, otel does not cache, it creates a new object, so we can must it readonly instantly.

Let's talk about this IRL.

Thanks for the investigation!

[alltilla]

One minor suggestion, but it is completely okay for me to merge this as is. Let me know if you don't want to change the code based on my suggestion, and I will merge the PR.

Thanks!

[alltilla]

Awesome, this turned out really nice :)