Re: [PATCH 0/3] Using a bit more decoration

From: Junio C Hamano <gitster@pobox.com>
Date: Apr 9, 2013, 20:06 UTC
Message-ID: <7v8v4rqxj6.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20130409065256.GA20115@sigill.intra.peff.net>

Jeff King <peff@peff.net> writes:

Show 13 quoted lines

> +static inline int *slab_at(struct commit_slab *s, const struct commit *c)
> +{
> +	if (s->alloc <= c->index) {
> +		int new_alloc = alloc_nr(s->alloc);
> +		if (new_alloc <= c->index)
> +			new_alloc = c->index + 1;
> +
> +		s->buf = xrealloc(s->buf, new_alloc * sizeof(*s->buf));
> +		memset(s->buf + s->alloc, 0, new_alloc - s->alloc);
> +		s->alloc = new_alloc;
> +	}
> +	return s->buf + c->index;
> +}

This will hurt more as the number of objects we deal with grows (our ALLOC_GROW() shares the same).

I wonder if it might be a good idea to do

	struct commit_slab {
        	int piece_size;
		int piece_count;
		struct commit_slab_piece {
                	int *buf;
		} *piece;
	};

and then make the look-up logic like this:

	int nth_piece, nth_slot;

	nth_piece = c->index / s->piece_size;
        nth_slot = c->index % s->piece_size;
        if (s->piece_count <= nth_piece) {
		/* xrealloc s->piece to grow, update s->piece_count */
        }
        if (!s->piece[nth_piece]) {
		/* xcalloc s->piece[nth_piece] to allocate */
	}
        return s->piece[nth_piece]->buf + nth_slot;

Other than that, looks like a good technology demonstration.