folly::max_align_v

constexpr std::size_t max_align_v = detail::max_align_v_::value();

Description

max_align_t is a type which is aligned at least as strictly as the most-aligned basic type (see the specification of std::max_align_t). This implementation exists because 32-bit iOS platforms have a broken std::max_align_t (see below).

You should refer to this as ::folly::max_align_t in portable code, even if you have using namespace folly; because C11 defines a global namespace max_align_t type.

To be certain, we consider every non-void fundamental type specified by the standard. On most platforms long double would be enough, but iOS 32-bit has an 8-byte aligned double and long long int and a 4-byte aligned long double.

So far we've covered locals and other non-allocated storage, but we also need confidence that allocated storage from malloc, new, etc will also be suitable for objects with this alignment requirement.

Apple document that their implementation of malloc will issue 16-byte granularity chunks for small allocations (large allocations are page-size granularity and page-aligned). We think that allocated storage will be suitable for these objects based on the following assumptions:

1. 16-byte granularity also means 16-byte aligned. 2. new and other allocators follow the malloc rules.

We also have some anecdotal evidence: we don't see lots of misaligned-storage crashes on 32-bit iOS apps that use double.

Apple's allocation reference: http://bit.ly/malloc-small