Hi,

this regression is about literal_type_p ICEing for types which cannot be 
completed. Indeed, for the testcase, complete_type cannot complete the 
type but doesn't error out either, just returns the type as-is, and the 
gcc_assert triggers. We could imagine handling such types in the caller 
- check_field_decls - but in my opinion makes more sense to just allow 
such types and return false. I also considered changing literal_type_p 
to use complete_type_or_else but then it's easy to produce duplicate 
diagnostics, for example. What do you think?

Tested x86_64-linux.

Thanks,
Paolo.

PS: eventually I guess we want to fix this in mainline and 4.7.1.

///////////////////////////