to GList or not to GList?

[Joshua N Pritikin <vishnu pobox com>]

i've implement linked-lists about a million times, but i'm still confused.

The GList structure is stored separately from the data.  What is the
rationale for this organization?  Is it mainly for GINT_TO_POINTER & friends?
To contrast, here is a linked-list implementation that must be embedded in
the data object (see MIME attachment).  For example,

  struct mydata {
    pe_ring peer;
    pe_ring queue;
    gint hits;
    gint priority;
  };

In some cases, this organization seems like a better fit for the problem.

Ignoring the question of API freezes, would such an embedded list be appropriate
for inclusion in glib?  Or is this style of design generally a bad idea?

-- 
Get self-realization at <http://sahajayoga.org> ... <http://why-compete.org> ?
  Victory to the Divine Mother!!
struct pe_ring { void *self; pe_ring *next, *prev; };

#define PE_RING_INIT(LNK, SELF) 		\
STMT_START {					\
  (LNK)->next = LNK;				\
  (LNK)->prev = LNK;				\
  (LNK)->self = SELF;				\
} STMT_END

#define PE_RING_EMPTY(LNK) ((LNK)->next == LNK)

#define PE_RING_UNSHIFT(LNK, ALL)		\
STMT_START {					\
  assert((LNK)->next==LNK);			\
  (LNK)->next = (ALL)->next;			\
  (LNK)->prev = ALL;				\
  (LNK)->next->prev = LNK;			\
  (LNK)->prev->next = LNK;			\
} STMT_END

#define PE_RING_ADD_BEFORE(L1,L2)		\
STMT_START {					\
  assert((L1)->next==L1);			\
  (L1)->next = L2;				\
  (L1)->prev = (L2)->prev;			\
  (L1)->next->prev = L1;			\
  (L1)->prev->next = L1;			\
} STMT_END

#define PE_RING_DETACH(LNK)			\
STMT_START {					\
  if ((LNK)->next != LNK) {			\
    (LNK)->next->prev = (LNK)->prev;		\
    (LNK)->prev->next = (LNK)->next;		\
    (LNK)->next = LNK;				\
  }						\
} STMT_END