Fix issue with anon union in generated code

[watzon]

Ok I still don't completely understand the root cause of this issue, but I documented my discovery in olbat/libgen#59. Basically in trying to generate bindings for ruby.h I was running into an issue where I was getting the error Bug: missing struct name. I eventually tracked the error to this line and this line in ruby/intern.h where there is an anonymous union inside some generated code. In trying to generate a Crystal type for the anonymous union it threw an error because there was no name to assign the type to.

So I did something simple that could probably be improved upon, and added a counter, removed the raise call that was causing things to break, and if all the other conditions in TypeMapper#check_anonymous_name fail, it returns a generated name anon_#{@anon_count += 1}. So now instead of breaking it gives that union a crystal type name Anon1.

Let me know if there is anything I can do to improve this, but I personally see it as an acceptable solution. This looks like an edge case that probably won't pop up much. I couldn't even figure out how to reproduce the bug for the specs.

[bcardiff]

@watzon I think we would need a small spec at least for this. Otherwise it will get broken easy in the future.

[watzon]

If you can help me come up with something that reproduces that behavior I'd be happy to add one. I'm having a hard time understanding what's going on in that C header that was causing things to break. I tried creating a minimal example for a spec, but was unable to.

[olbat]

👋, here is a short code snippet containing a function declaration to reproduce the issue:

void func(union{ char x; int i; });

[watzon]

Ok I added a parser spec and fixed a couple warnings coming from the 0.35.0 update. I think to really reproduce the original issue I might have to add another spec to the lib_body_transformer_spec though.