Alain Frisch:

> I support this suggestion. The standard equality/ordering/hashing
> functions are often adequate for most of the data structures, and it would
> be useful to use them and just place a hook on specific types to provide
> specialized implementation. It could almost be done with
> the "custom" block tag; the problem is that such blocks are not garbage
> collected.  What about "custom Caml blocks" ?
 
The memory manager Cmm (not to be confused with the IL in ocamlopt)
gives a generic garbage collection interface that allows custom garbage
collectors and custom heaps in such a way that one can use multiple
collectors with multiple heaps including collection of cross heap
pointers.  It is designed to do garbage collection of C++ so it is
complicated by C++ details.  But the basic design is good.  To use the
Cmm design with custom blocks would require 

-- defining a heap interface type

struct heap_operations {
	value (*alloc)(heap h, unsigned long size, mlsize_t mem, mlsize_t max);
	value (*alloc_block)(heap h, tag_t tag, mlsize_t size, mlsize_t mem, mlsize_t max);
	void (*register_root)(heap h, value r);
	void (*unregister_root)(heap h, value r);
	void (*scavenge)(heap h, value v);  /* ie mark, copy, etc */
	void (*collect)(heap h);
	void (*collect_slice)(heap h); /* may need more parameters */
	void (*compact)(heap h);
}; 

-- an addtional operation pointer: 

struct custom_operations {
	...
	void (*traverse)(value v); /* walks v calling scavenge on its pointers */
};

-- using the existing minor and major heaps to implement heap interfaces

value alloc_shr_prime(heap h, tag_t tag, mlsize_t size, mlsize_t mem, mlsize_t max){
	return alloc_shr(tag,size);
}

...

struct heap_operations major_heap = { 
	0, /* you can only alloc standard blocks in the major heap */
	&alloc_shr_prime, 
	&register_global_root_prime,
	... };
	
	
See

	ftp://ftp.di.unipi.it/pub/project/posso/cmm/ 
	
for Cmm papers and source.  Kaffe (an open source Java implementation)
also has custom allocators and collectors but I have forgotten the details.
-------------------
To unsubscribe, mail caml-list-request@inria.fr.  Archives: http://caml.inria.fr