ZST reallocs are a minefield.

[MaulingMonkey]

• malloc(0) may allocate where realloc(ptr, 0) frees or is undefined behavior!
• Some allocators can create zero-sized allocs, but then can't reallocate them!
• Kinda wish my realloc traits took NonZeroUsizes

[MaulingMonkey]

So after brainstorming options for how to respond to zero sized alloc requests, I apparently chose all the options.

[workingjubilee]

malloc(0) may allocate where realloc(ptr, 0) frees or is undefined behavior!

Some notes:

Before C23 the realloc(ptr, 0) behavior is a de facto "implementation dependent" choice from the 2~3 actual implementations in the wild (across many OS, so there was strong convergence). It wasn't really meant to be implementation dependent but no one was able to agree on a fully converged implementation for C23, so it got undefined to allow any implementation. People on WG14 have asserted that it won't be suddenly exploited by compilers tomorrow, but the record shows how that's gone before, as tomorrow has its own tomorrows. So you can choose to trust WG14 assertions and use a cfged table of behaviors for different OS, or you can choose to treat realloc(ptr, 0) as a time bomb (which incidentally is also the more portable option).

malloc(0) must be either a null pointer or a unique pointer (i.e. it can be NULL or it can effectively allocate a byte).