gd_intern.h is the best place we have for these things. the rest of the gd codebase should be able to assume a recent POSIX environment and completely basic types like ssize_t. so either we drop MSVC, or we workaround their crap setup somewhere. afaik, MSVC refuses to provide a standard ssize_t and such which is why we have these defines.

gd_intern.h is the best place we have for these things.

I'd prefer to have such fallback definitions in a generated config.h or such, but I agree that having them in gd_intern.h is better than to spread them all over the codebase.

the MSVC logic isn't really generated in this case ... it's hand-written stuff. it also requires including other headers in order to make the types work correctly (which we def don't want in config.h). even though the exact line is a define (#define ssize_t SSIZE_T), and is thus fine to setup in config.h, we have to check a couple diff headers in order to find SSIZE_T for MSVC, and that wouldn't work in config.h, and i don't want to include that boiler plate in every header file.

GNU projects in general rely on the gnulib project to hide this stuff, and it can go as far as generating local headers to overlay/replace system headers (e.g. sys/types.h). the downside is that gnulib is (L)GPL leaning while gd is BSD-like, and having us depend on gnulib will make the combined work subject to the (L)GPL (which personally i don't care about, but it seems an unreasonable change for the gd project). plus, gnulib really really wants to be linked by itself into a shared library or executable and not expose its symbols, which means putting it into a static libgd.a will easily cause problems for users.

i'm not aware of any other portability shims out there which means we're left open-coding our own ad-hoc ones like gd_intern.h. which is a fun name like: let the interns deal with this internal crap :p. on the flip side, it's a bit of pressure to try and keep our C library footprint somewhat smaller, which can help some embedded folks. shrug