When we find a record in the slot's update skiplist, but then want to

jump past the rest of the deleted records, we have to adjust based on
the starting record of the slot, use the page's repeat array to find
that starting record.

Another run at the __col_insert_search_gt (the greater-than skiplist
search), hopefully it's finally correct.

1 files changed, 22 insertions, 16 deletions

diff --git a/src/include/column.i b/src/include/column.i
index 5c1f4ee36b8..62422d3ddde 100644
--- a/src/include/column.i
+++ b/src/include/column.i
@@ -28,26 +28,29 @@ __col_insert_search_gt(WT_INSERT_HEAD *inshead, uint64_t recno)
 	 * The insert list is a skip list: start at the highest skip level, then
 	 * go as far as possible at each level before stepping down to the next.
 	 */
+	ins = NULL;
 	for (i = WT_SKIP_MAXDEPTH - 1, insp = &inshead->head[i]; i >= 0;)
-		if (*insp != NULL && recno <= WT_INSERT_RECNO(*insp)) {
-			ins = *insp;
+		if (*insp != NULL && recno > WT_INSERT_RECNO(*insp)) {
+			ins = *insp;	/* GTE: keep going at this level */
 			insp = &(*insp)->next[i];
 		} else {
-			--i;
+			--i;		/* LT: drop down a level */
 			--insp;
 		}
 
 	/*
-	 * If the skiplist consisted of a single record larger than the target,
-	 * we never updated ins and it references that record. Otherwise, ins
-	 * references a record less-than-or-equal to the target, move to a
-	 * record after it, that is, the subsequent record that's greater than
-	 * the target. Because inserts happen concurrently, additional records
-	 * may be inserted after the searched-for record that are still smaller
-	 * than the target, continue to move forward until reaching a record
-	 * larger than the target. There isn't any safety testing because we
-	 * confirmed such a record exists before searching.
+	 * If we didn't find any records smaller than the target, we never set
+	 * the return value, set it to the first record in the list. Otherwise,
+	 * it references a record less-than-or-equal to the target, move to a
+	 * later record, that is, a subsequent record greater than the target.
+	 * Because inserts happen concurrently, additional records might be
+	 * inserted after the searched-for record that are still smaller than
+	 * the target, continue to move forward until reaching a record larger
+	 * than the target. There isn't any safety testing because we confirmed
+	 * such a record exists before searching.
 	 */
+	if (ins == NULL)
+		ins = WT_SKIP_FIRST(inshead);
 	while (recno >= WT_INSERT_RECNO(ins))
 		ins = WT_SKIP_NEXT(ins);
 	return (ins);
@@ -77,10 +80,10 @@ __col_insert_search_lt(WT_INSERT_HEAD *inshead, uint64_t recno)
 	 */
 	for (i = WT_SKIP_MAXDEPTH - 1, insp = &inshead->head[i]; i >= 0;)
 		if (*insp != NULL && recno < WT_INSERT_RECNO(*insp)) {
-			ins = *insp;
+			ins = *insp;	/* LT: keep going at this level */
 			insp = &(*insp)->next[i];
 		} else  {
-			--i;
+			--i;		/* GTE: drop down a level */
 			--insp;
 		}
 
@@ -235,7 +238,7 @@ __col_fix_last_recno(WT_PAGE *page)
  *	Search a variable-length column-store page for a record.
  */
 static inline WT_COL *
-__col_var_search(WT_PAGE *page, uint64_t recno)
+__col_var_search(WT_PAGE *page, uint64_t recno, uint64_t *start_recnop)
 {
 	WT_COL_RLE *repeat;
 	uint64_t start_recno;
@@ -255,8 +258,11 @@ __col_var_search(WT_PAGE *page, uint64_t recno)
 
 		repeat = page->pg_var_repeats + indx;
 		if (recno >= repeat->recno &&
-		    recno < repeat->recno + repeat->rle)
+		    recno < repeat->recno + repeat->rle) {
+			if (start_recnop != NULL)
+				*start_recnop = repeat->recno;
 			return (page->pg_var_d + repeat->indx);
+		}
 		if (recno < repeat->recno)
 			continue;
 		base = indx + 1;