/* * __cursor_var_next -- * Move to the next, variable-length column-store item. */ static inline int __cursor_var_next(WT_CURSOR_BTREE *cbt, bool newpage) { WT_CELL *cell; WT_CELL_UNPACK unpack; WT_COL *cip; WT_INSERT *ins; WT_PAGE *page; WT_SESSION_IMPL *session; WT_UPDATE *upd; uint64_t rle, rle_start; session = (WT_SESSION_IMPL *)cbt->iface.session; page = cbt->ref->page; rle_start = 0; /* -Werror=maybe-uninitialized */ /* Initialize for each new page. */ if (newpage) { cbt->last_standard_recno = __col_var_last_recno(cbt->ref); if (cbt->last_standard_recno == 0) return (WT_NOTFOUND); __cursor_set_recno(cbt, cbt->ref->ref_recno); cbt->cip_saved = NULL; goto new_page; } /* Move to the next entry and return the item. */ for (;;) { if (cbt->recno >= cbt->last_standard_recno) return (WT_NOTFOUND); __cursor_set_recno(cbt, cbt->recno + 1); new_page: /* Find the matching WT_COL slot. */ if ((cip = __col_var_search(cbt->ref, cbt->recno, &rle_start)) == NULL) return (WT_NOTFOUND); cbt->slot = WT_COL_SLOT(page, cip); /* Check any insert list for a matching record. */ cbt->ins_head = WT_COL_UPDATE_SLOT(page, cbt->slot); cbt->ins = __col_insert_search_match(cbt->ins_head, cbt->recno); upd = cbt->ins == NULL ? NULL : __wt_txn_read(session, cbt->ins->upd); if (upd != NULL) { if (upd->type == WT_UPDATE_DELETED) { if (__wt_txn_upd_visible_all(session, upd)) ++cbt->page_deleted_count; continue; } return (__wt_value_return(session, cbt, upd)); } /* * If we're at the same slot as the last reference and there's * no matching insert list item, re-use the return information * (so encoded items with large repeat counts aren't repeatedly * decoded). Otherwise, unpack the cell and build the return * information. */ if (cbt->cip_saved != cip) { if ((cell = WT_COL_PTR(page, cip)) == NULL) continue; __wt_cell_unpack(cell, &unpack); if (unpack.type == WT_CELL_DEL) { if ((rle = __wt_cell_rle(&unpack)) == 1) continue; /* * There can be huge gaps in the variable-length * column-store name space appearing as deleted * records. If more than one deleted record, do * the work of finding the next record to return * instead of looping through the records. * * First, find the smallest record in the update * list that's larger than the current record. */ ins = __col_insert_search_gt( cbt->ins_head, cbt->recno); /* * Second, for records with RLEs greater than 1, * the above call to __col_var_search located * this record in the page's list of repeating * records, and returned the starting record. * The starting record plus the RLE is the * record to which we could skip, if there was * no smaller record in the update list. */ cbt->recno = rle_start + rle; if (ins != NULL && WT_INSERT_RECNO(ins) < cbt->recno) cbt->recno = WT_INSERT_RECNO(ins); /* Adjust for the outer loop increment. */ --cbt->recno; continue; } WT_RET(__wt_page_cell_data_ref( session, page, &unpack, cbt->tmp)); cbt->cip_saved = cip; } cbt->iface.value.data = cbt->tmp->data; cbt->iface.value.size = cbt->tmp->size; return (0); } /* NOTREACHED */ }
/*移向下条variable-length column-store 记录*/ static inline int __cursor_var_next(WT_CURSOR_BTREE* cbt, int newpage) { WT_CELL *cell; WT_CELL_UNPACK unpack; WT_COL *cip; WT_ITEM *val; WT_INSERT *ins; WT_PAGE *page; WT_SESSION_IMPL *session; WT_UPDATE *upd; uint64_t rle, rle_start; session = (WT_SESSION_IMPL *)cbt->iface.session; page = cbt->ref->page; val = &cbt->iface.value; rle_start = 0; /* -Werror=maybe-uninitialized */ if (newpage){ cbt->last_standard_recno = __col_var_last_recno(page); if (cbt->last_standard_recno == 0) return (WT_NOTFOUND); __cursor_set_recno(cbt, page->pg_var_recno); goto new_page; } for (;;){ if (cbt->recno >= cbt->last_standard_recno) return (WT_NOTFOUND); __cursor_set_recno(cbt, cbt->recno + 1); new_page: /*定位到recno对应的WT_COL slot*/ if ((cip = __col_var_search(page, cbt->recno, &rle_start)) == NULL) return (WT_NOTFOUND); cbt->slot = WT_COL_SLOT(page, cip); /*读取内容值*/ cbt->ins_head = WT_COL_UPDATE_SLOT(page, cbt->slot); cbt->ins = __col_insert_search_match(cbt->ins_head, cbt->recno); upd = cbt->ins == NULL ? NULL : __wt_txn_read(session, cbt->ins->upd); if (upd != NULL) { if (WT_UPDATE_DELETED_ISSET(upd)) { ++cbt->page_deleted_count; continue; } val->data = WT_UPDATE_DATA(upd); val->size = upd->size; return (0); } /* * If we're at the same slot as the last reference and there's * no matching insert list item, re-use the return information * (so encoded items with large repeat counts aren't repeatedly * decoded). Otherwise, unpack the cell and build the return * information. * upd == NULL, 记录可能被删除放入到了insert列表中,slot可能被重用了,那么需要进行cell unpack取值 */ if (cbt->cip_saved != cip) { if ((cell = WT_COL_PTR(page, cip)) == NULL) continue; __wt_cell_unpack(cell, &unpack); if (unpack.type == WT_CELL_DEL) { if ((rle = __wt_cell_rle(&unpack)) == 1) continue; /*定位到修改列表中的记录*/ ins = __col_insert_search_gt(cbt->ins_head, cbt->recno); cbt->recno = rle_start + rle; if (ins != NULL && WT_INSERT_RECNO(ins) < cbt->recno) cbt->recno = WT_INSERT_RECNO(ins); /* Adjust for the outer loop increment. */ --cbt->recno; continue; } /*取出cell中的值到tmp中*/ WT_RET(__wt_page_cell_data_ref(session, page, &unpack, &cbt->tmp)); cbt->cip_saved = cip; } val->data = cbt->tmp.data; val->size = cbt->tmp.size; return 0; } }