Re: RFC: New diff-delta.c implementation

2 messages, 2 authors, 2016-06-15 · open the first message on its own page

Re: RFC: New diff-delta.c implementation

From: Junio C Hamano <hidden>
Date: 2016-06-15 22:42:24

Nicolas Pitre [off-list ref] writes:
Well, actually I was measuring a 10% speed improvement with a quick and 
naive (not memory efficient) approach for pack-objects with the current 
algorithm.
...
The idea to avoid memory pressure is to reverse the window processing 
such that the object to delta against is constant for the entire window 
instead of the current logic where the target object is constant.  This 
way there would be only one index in memory at all time.
Your are right.  The first led to the latter unexplored idea.

I expect to be offline most of the day today, and have other
things I can work on for the next few days anyway, so if you or
somebody else have an inclination and energy to reverse the
delta window, I would appreciate that.

Maybe the calling convention of diff-delta.c would become
something like this?

struct delta_index; /* opaque to the caller; implementation
		     * defines what's in it.
                     */

/* returns a newly allocated struct delta_index.
 * input "buf" pointer can be stored in the struct, but "buf"
 * does not belong to diff-delta module (i.e. borrowed reference).
 */
struct delta_index *delta_index(
  void *buf,			/* input: from buffer */
  unsigned long size,		/* input: from size */
);

/* ... so free the structure and its internal data, but
 * do not free the borrowed reference!
 */
void free_delta_index(struct delta_index *);

/* Take "from", an already preprocessed delta_index for the
 * traditional from_buffer/from_size, and to_buf/to_size, and
 * produce delta in newly allocated buffer (caller should
 * free() when it is done), and return the result size in
 * *delta_size.  Stop early if the result would exceed max_size.
 */
void *diff_delta(
    struct delta_index *from,	/* input: prepared by delta_index() */
    void *to_buf,		/* input: destination buffer */
    unsigned long to_size,	/* input: destination size */
    unsigned long *delta_size,	/* output: result size */
    unsigned long max_size	/* input: do not waste cycles if
                                   you cannot generate result
                                   smaller than this */
);

and the calling convention would be:

	struct unpacked *s, *d;
	unsigned long max_size;

	/* precompute the index */
	struct delta_index *src = delta_index(s->data, s->entry->size);

	/* do the delta */
        void *delta_buf = diff_delta(src, d->data, d->entry->size,
                                     &sz, max_size);
        /* do useful thing here on delta_buf and sz */
        free(delta_buf);

	/* the caller can reuse *src with other *d,
         * but when it is done...
         */
        free_delta_index(src);

Re: RFC: New diff-delta.c implementation

From: Nicolas Pitre <hidden>
Date: 2016-06-15 22:42:24

On Sat, 22 Apr 2006, Junio C Hamano wrote:
Nicolas Pitre [off-list ref] writes:
quoted
Well, actually I was measuring a 10% speed improvement with a quick and 
naive (not memory efficient) approach for pack-objects with the current 
algorithm.
...
The idea to avoid memory pressure is to reverse the window processing 
such that the object to delta against is constant for the entire window 
instead of the current logic where the target object is constant.  This 
way there would be only one index in memory at all time.
Your are right.  The first led to the latter unexplored idea.

I expect to be offline most of the day today, and have other
things I can work on for the next few days anyway, so if you or
somebody else have an inclination and energy to reverse the
delta window, I would appreciate that.
I'll probably give it a try.

I'm still reviewing Geert's code right now and found minor things 
pertaining to the GIT delta encoding here and there which probably 
explain why it doesn't pack the Linux kernel archive yet.


Nicolas
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help