Definition of ssize_t is wrong

[snej]

sockpp has its own declaration of ssize_t in platform.h, but on non-Windows systems it defines it as int. This is wrong on 64-bit OSs that use 32-bit ints, which includes all Apple platforms. (Dunno about Linux or BSD.) This leads to compile errors if a source file ends up including both platform.h and the OS's declaration. That aside, there are potentially actual integer overflow bugs if dealing with sizes > 4GB.

sockpp/include/sockpp/platform.h

Lines 78 to 86 in 180f18d

	#ifndef _SSIZE_T_DEFINED
	#define _SSIZE_T_DEFINED
	#undef ssize_t
	#ifdef _WIN64
	using ssize_t = int64_t;
	#else
	using ssize_t = int;
	#endif // _WIN64
	#endif // _SSIZE_T_DEFINED

You shouldn't need to declare this at all. Instead:

• On a POSIX-compliant OS, #include <sys/types.h>.
• On Windows, #include <BaseTsd.h> and then using ssize_t = SSIZE_T;.

(Also, it's kind of icky that the types declared in platform.h are all in the global namespace. Could they be put in the sockpp namespace like the rest of the API?)

[snej]

Aw crap, I didn't notice that all that stuff is wrapped in #if defined(_WIN32).

@borrrden tells me that the definition on Windows should still be fixed to use the one from the system, because the system one ends up being typedef'd to long not int, which is incompatible.

[fpagliughi]

Ha. I saw this and thought, "That must be something new". But I find references to it going back to 2002, so I must have missed something.

Sure, I'd much rather use definitions from each system than make them up myself. Just wish Microsoft was more aligned to the rest of the world.

[fpagliughi]

OK. I'll cherry-pick a1271fb, and a few other non-SSL commits in #17, and include then in an upcoming point release.

[fpagliughi]

Done with a1271fb.