__INTCAP_WIDTH__ missing

[brooksdavis]

• __(U)INTCAP_TYPE__ macros not defined
• __(U)INTCAP_WIDTH__ macros are not defined (intptr_t ones are wrong per __(U)INTPTR_WIDTH__ and __POINTER_WIDTH__ macros are too wide #795)
• __(U)INTCAP_FMT*__ macros are undefined (GCC doesn't provide these in general)
• __SIZEOF_(U)INTCAP__ are defined when that's not the case for (u)intptr_t

Purecap clang INTCAP macros:

$ clang -dM -E -x c /dev/null | grep INTCAP
#define __INTCAP_MAX__ 9223372036854775807L
#define __SIZEOF_INTCAP__ 16
#define __SIZEOF_UINTCAP__ 16
#define __UINTCAP_MAX__ 18446744073709551615UL

Purecap clang INTPTR macros:

clang -dM -E -x c /dev/null | grep INTPTR
#define __INTPTR_FMTd__ "Pd"
#define __INTPTR_FMTi__ "Pi"
#define __INTPTR_MAX__ 9223372036854775807L
#define __INTPTR_TYPE__ __intcap
#define __INTPTR_WIDTH__ 128
#define __UINTPTR_FMTX__ "PX"
#define __UINTPTR_FMTo__ "Po"
#define __UINTPTR_FMTu__ "Pu"
#define __UINTPTR_FMTx__ "Px"
#define __UINTPTR_MAX__ 18446744073709551615UL
#define __UINTPTR_TYPE__ unsigned __intcap
#define __UINTPTR_WIDTH__ 128

[jrtc27]

That's because __intcap (and __[u]intcap_t) is a primitive type, whilst uintptr_t is a typedef to a pre-existing primitive type. Contrast with grep INT128 or grep SHO?RT.

[jrtc27]

WIDTH should be defined though.

[brooksdavis]

Ah, that makes sense. I'm a touch skeptical about TYPE for non-standard types, but that does seem to be how it's done.

[jrtc27]

The oddity is that __[u]intcap_t are defined at all, rather than just a lone __intcap, but that's what happens for int128, you get __int128 and __[u]int128_t all pre-defined.