For C++17's parallel algorithms, where we separately declare and define many of them, we developed a convention where the declaration is = 0 and the definition is /* = 0 */. This is a reminder that while the default template argument can't be repeated, we haven't simply forgotten it:

Thanks to Michael Grier for noticing that we've accumulated inconsistency elsewhere, with relics of an older convention where we'd name the template parameter in the definition _Enabled. That older convention was less clear and doesn't really provide new information (as we're already using enable_if_t).

• Change _Enabled = 0 to = 0 for countr_one() and _Popcount().
  • There are no other declarations.
• Change _Enabled to /* = 0 */ for duration_cast(), rotr(), countl_zero(), and swap() for arrays.
• Cleanup swap(_Ty&, _Ty&).
  • For C++17, follow the convention where the declaration is = 0 and the definition is /* = 0 */.
  • For C++14, we want to emphasize the purpose of the second template parameter, so the declaration is int _Enabled = 0 and the definition is int _Enabled /* = 0 */.
• Add /* = 0 */ to tuple's ctors and uses_allocator_construction_args().

After this cleanup, the only remaining uses of _Enabled in product code are in other contexts. I've audited all occurrences where enable_if_t appears in separate declarations and definitions, including through helper typedefs, so this cleanup should be fairly exhaustive.