Here is mine, from 1999. The null-terminated data starts after the header.

  typedef struct {
    size_t len;
    long refcount;
  } dstr_t;

  dstr_t *dstr_alloc(const char *);
  dstr_t *dstr_duplicate(const dstr_t *);
  dstr_t *dstr_addref(dstr_t *);
  void dstr_release(dstr_t *);
  void dstr_dt(void *);
  dstr_t *dstr_concat(const dstr_t *, const dstr_t *);
  int dstr_cmp(const dstr_t *, const dstr_t *);
  int dstr_cmp_void(const void *, const void *);
  size_t dstr_len(const dstr_t *);
  const char *dstr_peek(const dstr_t *);
  char *dstr_take(const dstr_t *);
  dstr_t *dstr_extract(const dstr_t *, size_t, size_t);
  size_t dstr_chr(const dstr_t *, char);
  size_t dstr_rchr(const dstr_t *, char);

  #define dstr_len(D) ((D)->len)
  #define dstr_auto_clean(D) except_cleanup_push(dstr_dt, (D))
  #define dstr_clean() except_checked_cleanup_pop(dstr_dt,   except_call)

In the implementation, this is used for getting at the string data:

  #define dstr_str(D) ((char *)((dstr_t *) (D) + 1))

For some reason I didn't bother with the flexible array member trick.