Basile Starynkevitch <bas...@starynkevitch.net> writes: > What are exactly GCC code types, defined in gcc/coretypes.h?
They are types for which there was no obvious place to put them. > Of course, they are very important types used in GCC. And there are > probably wired in gengtype (but maybe not all of them; I didn't check > for all; certainly gengtype handle tree & gimple specially in some > way.). gengtype only handles a few types specially. There is no particular connection between types recognized by gengtype and coretypes.h. E.g., gengtype handles VEC specially, but it is not in coretypes.h. > However, I don't understand why gimple_seq_node is a core type. It seems > rarely used (outside of gimple*.[ch] files), and I would imagine it > could be kept in gimple.h. I agree. I expect it's there just because gimple_seq is there. > I also don't understand why > union section; > typedef union section section; > struct cl_target_option; > struct cl_optimization; > should be kept in gcc/coretypes.h I don't know either. > I could imagine that the transition to C++ might have to C++-ify these > core types (all of gcc/coretypes.h) very quickly, or on the contrary to > keep them as they are, without making a class of them, for a long time. I don't think the transition to C++ will have to change any of these types. We should not make it a goal to convert all types to classes. > Do we have a clear criterium to add, keep, or remove a type from > gcc/coretypes.h? No. Ian
