SEO

Not to come off as "pushy" but wanted to make sure these patches
didn't fall through the cracks, assuming they were rejected for
some reason.

Cheers,


On Sun, Apr 06, 2008 at 06:32:39PM -0700, patrick keshishian wrote:
> On Sun, Apr 06, 2008 at 09:39:20AM -0400, Todd C. Miller wrote:
> > In message <20080406082118.GA19070@roppongi.boxsoft.com>
> > so spake patrick keshishian (sidster):
> >
> > > Also, hashp->cndx is decremented (adjusted) during deletion of
> > > the key/data pair if hashp->cndx is pointing to the entry being
> > > deleted. The side-effect of not doing this is entries are
> > > skipped due to the way __delpair "shuffles" keys.
> >
> > Looks reasonable, though I wonder if it would not be better to do
> > the cndx fixup inside __delpair() for the shuffle case.
>
> Does this look OK?
>
>
> (Again hash.c diff is based on first patch in my original email)
>
> --- /usr/src/lib/libc/db/hash/hash.c Tue Aug 1 03:11:31 2006
> +++ hash/hash.c Sun Apr 6 18:18:38 2008
> @@ -705,7 +708,7 @@ hash_seq(const DB *dbp, DBT *key, DBT *data, u_int32_t
> hashp->cndx = 1;
> hashp->cpage = NULL;
> }
> -
> + next_bucket:
> for (bp = NULL; !bp || !bp[0]; ) {
> if (!(bufp = hashp->cpage)) {
> for (bucket = hashp->cbucket;
> @@ -724,8 +727,18 @@ hash_seq(const DB *dbp, DBT *key, DBT *data, u_int32_t
> hashp->cbucket = -1;
> return (ABNORMAL);
> }
> - } else
> + } else {
> bp = (u_int16_t *)hashp->cpage->page;
> + if (flag == R_NEXT) {
> + hashp->cndx += 2;
> + if (hashp->cndx > bp[0]) {
> + hashp->cpage = NULL;
> + hashp->cbucket++;
> + hashp->cndx = 1;
> + goto next_bucket;
> + }
> + }
> + }
>
> #ifdef DEBUG
> assert(bp);
> @@ -755,13 +768,6 @@ hash_seq(const DB *dbp, DBT *key, DBT *data, u_int32_t
> key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
> data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
> data->size = bp[ndx] - bp[ndx + 1];
> - ndx += 2;
> - if (ndx > bp[0]) {
> - hashp->cpage = NULL;
> - hashp->cbucket++;
> - hashp->cndx = 1;
> - } else
> - hashp->cndx = ndx;
> }
> return (SUCCESS);
> }
>
> --- /usr/src/lib/libc/db/hash/hash_page.c Fri Aug 5 06:03:00 2005
> +++ hash/hash_page.c Sun Apr 6 18:18:05 2008
> @@ -149,6 +149,14 @@ __delpair(HTAB *hashp, BUFHEAD *bufp, int ndx)
> bp[i - 1] = bp[i + 1] + pairlen;
> }
> }
> + if (ndx == hashp->cndx) {
> + /*
> + * We just removed pair we were "pointing" to.
> + * By moving back the cndx we ensure subsequent
> + * hash_seq() calls won't skip over any entries.
> + */
> + hashp->cndx -= 2;
> + }
> }
> /* Finally adjust the page data */
> bp[n] = OFFSET(bp) + pairlen;