I explained how returning the address of the string buffer instead of the address of the struct would give you a C compatible string that you could pass to other C library functions. If those functions are "readonly" wrt the string, everything is copasetic.

if those string functions update/write the c-string (which is in the buffer) the strlen in the header will now be wrong. That has nothing to do with my suggestion, and it's already "broken" in that way you point out. My "string free bytes field" suggestion will also be broken by an operation like that, so my suggestion does not make this data structure worse than it already is wrt compatibility with C library functions.

However that strlen and free bytes problem can be managed (no worse than C standard strings themselves) and strlen and/or free bytes are useful features that make some other things easier so overall it's a win.