/*
 *  wi.h
 *
 *  $Id$
 *
 *  Data structures
 *
 *  This file is part of the OpenLink Software Virtuoso Open-Source (VOS)
 *  project.
 *
 *  Copyright (C) 1998-2012 OpenLink Software
 *
 *  This project is free software; you can redistribute it and/or modify it
 *  under the terms of the GNU General Public License as published by the
 *  Free Software Foundation; only version 2 of the License, dated June 1991.
 *
 *  This program is distributed in the hope that it will be useful, but
 *  WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
 *  General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License along
 *  with this program; if not, write to the Free Software Foundation, Inc.,
 *  51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 *
 */


/* Main header file for database engine */

#ifndef _WI_H
#define _WI_H

#define VAJRA
/*#undef CL6*/
#define KEYCOMP GPF_T1 ("not done with key comp");
#define O12 GPF_T1("Database engine does not support this deprecated function. Please contact OpenLink Support.")
/*#define PAGE_TRACE 1 */
/*#define BUF_BOUNDS*/
/* #define DBG_BLOB_PAGES_ACCOUNT */
#undef OLD_HASH
#if !defined (NEW_HASH)
#define NEW_HASH
#endif

#define AUTO_COMPACT

#ifndef bitf_t
#define bitf_t unsigned
#endif

#include "Dk.h"

#undef log

/*
 *  Global features
 */
typedef struct free_set_cache_s	free_set_cache_t;
typedef struct index_tree_s	index_tree_t;
typedef struct search_spec_s	search_spec_t;
typedef struct placeholder_s	placeholder_t;
typedef struct it_cursor_s	it_cursor_t;
typedef struct page_map_s	page_map_t;
typedef struct extent_map_s extent_map_t;
typedef struct buffer_desc_s	buffer_desc_t;
typedef struct it_map_s it_map_t;
#if 0
typedef int errcode;
#else
#ifndef _ERRCODE_DEFINED
typedef int errno_t;
#endif
#endif
typedef struct remap_s remap_t;
typedef struct buffer_pool_s buffer_pool_t;
typedef struct hash_index_s hash_index_t;
typedef struct row_delta_s row_delta_t;
typedef struct row_fill_s  row_fill_t;
typedef struct page_fill_s  page_fill_t;
typedef int (*key_cmp_t) (buffer_desc_t * buf, int pos, it_cursor_t * itc);
typedef struct pf_hash_s pf_hash_t;

#define WI_OK		0
#define WI_ERROR	-1

#include "widisk.h"
#include "widv.h"
#include "numeric.h"
#include "widd.h"
#include "ltrx.h"
#include "blobio.h"
#include "wifn.h"
#include "bitmap.h"
#include "extent.h"



#define IT_DP_MAP(it, dp) \
  (&(it)->it_maps[(dp) & IT_N_MAPS_MASK])

extern int it_n_maps;
#define IT_N_MAPS it_n_maps
#define IT_N_MAPS_MASK (it_n_maps - 1)

struct it_map_s
{
  dk_mutex_t 	itm_mtx;
  dk_hash_t	itm_dp_to_buf;
  dk_hash_t	itm_remap;
  dk_hash_t	itm_locks;
};

#define IT_DP_MAP(it, dp) \
  (&(it)->it_maps[(dp) & IT_N_MAPS_MASK])





#define IT_DP_TO_BUF(it, dp) \
  (buffer_desc_t *) gethash (DP_ADDR2VOID (dp), &IT_DP_MAP ((it), dp)->itm_dp_to_buf)

#define IT_DP_REMAP(it, dp, remap_dp) \
{ \
  it_map_t * itm = IT_DP_MAP (it, dp); \
  remap_dp = (dp_addr_t)(ptrlong) gethash (DP_ADDR2VOID (dp), &itm->itm_remap); \
  if (!remap_dp) \
    remap_dp = (dp_addr_t)(ptrlong) gethash (DP_ADDR2VOID (dp), it->it_storage->dbs_cpt_remap); \
  if (!remap_dp) \
    remap_dp = dp; \
}

#define IT_DP_TO_BUF(it, dp) \
  (buffer_desc_t *) gethash (DP_ADDR2VOID (dp), &IT_DP_MAP ((it), dp)->itm_dp_to_buf)

#define IT_DP_REMAP(it, dp, remap_dp) \
{ \
  it_map_t * itm = IT_DP_MAP (it, dp); \
  remap_dp = (dp_addr_t)(ptrlong) gethash (DP_ADDR2VOID (dp), &itm->itm_remap); \
  if (!remap_dp) \
    remap_dp = (dp_addr_t)(ptrlong) gethash (DP_ADDR2VOID (dp), it->it_storage->dbs_cpt_remap); \
  if (!remap_dp) \
    remap_dp = dp; \
}



typedef unsigned int32 bp_ts_t; /* timestamp of buffer, use in  cache replacement to distinguish old buffers.  Faster than double linked list for LRU.  Wraparound does not matter since only differences of values are considered.  */

#define BP_N_BUCKETS 5

struct buffer_pool_s
{
  /* Buffer cache pool.  Many pools exist to avoid a single critical section for cache replacement */
  buffer_desc_t *	bp_bufs;
  int			bp_n_bufs;
  int			bp_next_replace; /*index into bp_bufs, points where the previous cache replacement took place */
  buffer_desc_t *	bp_first_free; /* when bufs become available through delete, they are pushed here and linked via bd_next */
  dk_mutex_t *	bp_mtx; /* serialize free buffer lookup in this pool */
  bp_ts_t	bp_ts; /* ts to assign to next touched buffer */
  bp_ts_t	bp_last_buf_ts;
  bp_ts_t	bp_stat_ts; /* bp_ts as of when the pool age stats were last computed */
  char		bp_stat_pending; /* flag for autocompact/stats gathering in progress */

  unsigned char * bp_storage;	/* pointer to storage area */

  /* Each pool is divided into BP_N_BUCKETS, each holding a approx
   * * equal no f buffers.  They are divided by bp_ts, with the 1st bucket
   * holding the oldest 1/BP_N_BUCKETS and so on.  Used for scheduling
   * relatively old dirty buffers for flush to disk.  Each */

  int32		bp_bucket_limit[BP_N_BUCKETS];
  int		bp_n_clean[BP_N_BUCKETS]; /* bp_ts at the boundary between buckets */
  int 		bp_n_dirty[BP_N_BUCKETS];
  buffer_desc_t **	bp_sort_tmp;
};


#define IN_BP(in) \
  mutex_enter (bp->bp_mtx)

#define LEAVE_BP(in) \
  mutex_leave (bp->bp_mtx)


#define wi_schema wi_master_wd->wd_schema
#define wi_master wi_master_wd->wd_primary_dbs


typedef struct wi_inst_s
{
  /* Global data representing a Virtuoso server instance */
  wi_db_t *		wi_master_wd;  /* initial and only logical database */
  dk_set_t 		wi_dbs; /* list of database file groups  */
  uint32			wi_n_dirty; /* dirty buffer count, approx The real count is in the buffer pools.  */
  dk_set_t		wi_free_schemas; /* schema structs awaiting idle moment for safe deallocation */
  int32			wi_max_dirty;
  char			wi_is_checkpoint_pending; /* true if no new activity should be started due to checkpoint */
  char			wi_atomic_ignore_2pc; /* do not wait for prepared uncommitted to be finished  before atomic.  Need that when resetting cluster cfg after node failures */
  char			wi_checkpoint_atomic;
  char			wi_checkpoint_rollback; /* use special cpt delta space for rb results? */
  lock_trx_t *		wi_cpt_lt; /* used to keep stuff rolled back for cpt duration */
  dk_set_t		wi_waiting_checkpoint; /* threads suspended for checkpoint duration */

  dk_set_t	wi_storage;
  char *		wi_open_mode; /* various crash recovery options */
  dk_mutex_t *	wi_txn_mtx; /* serialize lock wait graph and transaction thread counts */
  buffer_pool_t **	wi_bps;  /* set of buffer pools */
  short	wi_n_bps;
  unsigned short	wi_bp_ctr; /* round robin buffer pool counter, not serialized, used for picking  a different pool on consecutive buffer replacements */

  dbe_storage_t *	wi_temp;  /* file group for temp db, sort temps, hash indices etc. */
  short			wi_temp_allocation_pct;
  id_hash_t * 		wi_files;
} wi_inst_t;


/* wi_is_checkpoint_pending */
#define CPT_NONE 0
#define CPT_CHECKPOINT 1
#define CPT_ATOMIC_PENDING 2 /* in the process of killing transactions before entering into atomic mode */
#define CPT_ATOMIC 3 /* in atomic mode, only one transaction allowed */


extern wi_inst_t	wi_inst;
EXE_EXPORT (struct wi_inst_s *, wi_instance_get, (void));


struct wi_db_s
{
  /* Logical database.  Can in principle have multiple file groups , although now only one is supported */  caddr_t		wd_qualifier;
  dbe_storage_t *	wd_primary_dbs;
  dk_set_t 		wd_storage;
  dbe_schema_t *	wd_schema;
};


struct dbe_storage_s
{
  /* database file group */
  char		dbs_type;
  int		dbs_stripe_unit;
  caddr_t	dbs_name;
  caddr_t	dbs_cfg_file;
  dk_set_t		dbs_disks; /* list of disk_segment_t for multifile dbs */
  disk_segment_t *	dbs_last_segment;
  dk_set_t 	dbs_trees;
  dk_set_t 	dbs_deleted_trees; /*  dropped indices between now and last checkpoint.  Checkpoint finalizes the drop */

  char *		dbs_file; /* file name if single file dbs. */
  int			dbs_fd;
  OFF_T		dbs_file_length; /* file len if single file. */
  dk_mutex_t *	dbs_file_mtx; /* serializes  dbs_fd, if single file dbs */

  dp_addr_t		dbs_n_pages; /* Total pages in disk array or single file */
  dk_mutex_t *	dbs_page_mtx;  /* serializes page alloc/free */
  du_thread_t *	dbs_owner_thr;  /* thread owning this dbs, also owner of  dbs_page_mtx */
  buffer_desc_t *	dbs_free_set;  /* page allocation bitmap pages */
  buffer_desc_t *	dbs_incbackup_set; /* set of backuped pages for incremental backup */
  dp_addr_t		dbs_n_pages_in_sets; /* space for so many bits in free set and backup set */
  uint32		dbs_n_free_pages;
  uint32		dbs_n_pages_on_hold;  /* no of pages provisionally reserved for possible delta caused by tree splitting */
  dk_set_t		dbs_cp_remap_pages; /* list of page no's for writing checkpoint remap */
  uint32		dbs_max_cp_remaps;  /* max checkpoint remap */

  dk_session_t *	dbs_log_session;
  OFF_T		dbs_log_length;
  char *		dbs_log_name;
  log_segment_t *	dbs_log_segments; /* if log over several volumes */
  log_segment_t *	dbs_current_log_segment;  /* null if log segments off */
  dk_session_t *	dbs_2pc_log_session;
  caddr_t		dbs_2pc_file_name;
  id_hash_t *		dbs_registry_hash; /* em start dp's and treeroots here if many dbs's */
  dp_addr_t		dbs_registry; /* first page of registry */
  dp_addr_t	dbs_pages_changed;	/* bit map of changes since last backup.  Linked list like free set */
  dk_hash_t *	dbs_cpt_remap; /* checkpoint remaps in this storage unit.  Accessed with no mtx since changes only at checkpoint. */
  index_tree_t *	dbs_cpt_tree;  /* dummy tree for use during checkpoint */
  wi_db_t *		dbs_db;
  dp_addr_t		dbs_dp_sort_offset; /* when sorting buffers for flush, offset by this so as not to mix file groups */
  int			dbs_extend; /* size extend increment in pages */
  dk_hash_t *		dbs_unfreeable_dps;
  char * 		dbs_cpt_file_name;
  dk_session_t *	dbs_cpt_recov_ses; /* during cpt recov write or recov, the file ses with recov data */
  extent_map_t *	dbs_extent_map; /* system shared general purpose disk extents, housekeeping and small tables */
  dk_hash_t *	dbs_dp_to_extent_map;
  buffer_desc_t *	dbs_extent_set;
  dp_addr_t		dbs_n_pages_in_extent_set;
  int32			dbs_initial_gen; /* generic no of exe tat inited the db */
  char 			dbs_id[16];
} ;


/* dbs_type */
#define DBS_PRIMARY 0
#define DBS_SECONDARY 1
#define DBS_TEMP 2
#define DBS_RECOVER 3


#define IN_DBS(dbs) \
  if (dbs->dbs_owner_thr != THREAD_CURRENT_THREAD) \
    { \
      mutex_enter (dbs->dbs_page_mtx); \
      dbs->dbs_owner_thr = THREAD_CURRENT_THREAD; \
    }

#define LEAVE_DBS(dbs) \
{ \
  if (THREAD_CURRENT_THREAD != dbs->dbs_owner_thr)  \
    GPF_T1 ("Leaving dbs without owning it"); \
  dbs->dbs_owner_thr = NULL; \
  mutex_leave (dbs->dbs_page_mtx); \
}



typedef struct hi_signature_s
{
  /* hash index signature.  Indicates what cols of what table are the key and dependent parts */
  caddr_t 	hsi_super_key; /* key id of as int box for the key used to fill the hash inx. If key changes, hash inx is invalidated */
  caddr_t	hsi_n_keys; /* n first of hsi_col_ids relevant for lookup */
  oid_t *	hsi_col_ids; /* columns of the source key in the hash inx */
#ifdef NEW_HASH
  caddr_t	hsi_isolation;  /* record isolation because if made with read committed cannot be reused with repeatable read */
#endif
} hi_signature_t;


#ifdef OLD_HASH
typedef struct hash_inx_elt_s
{
  uint32		he_no;
  dp_addr_t		he_page;
  short			he_pos;
  struct hash_inx_elt_s * 	he_next;
} hash_inx_elt_t;
#endif

#ifdef NEW_HASH
typedef struct hash_inx_b_ptr_s
{
  uint32                hibp_no;
  dp_addr_t		hibp_page;
  short			hibp_pos;
} hash_inx_b_ptr_t;
#endif

struct hash_index_s
{
  mem_pool_t *		hi_pool;
  id_hash_t *		hi_memcache;
  int			hi_size;
  int64			hi_count;
#ifdef OLD_HASH
  hash_inx_elt_t **	hi_elements;
#endif
  dk_set_t 	hi_pages;
  int		hi_page_fill;
  index_tree_t *	hi_it;
  dp_addr_t	hi_last_source_dp;
  dp_addr_t 	*hi_buckets;
  dk_hash_t     *hi_source_pages;
  index_tree_t  *hi_source_tree;
  char		hi_lock_mode;
  char		hi_isolation;
};


typedef struct hash_index_cache_s
{
  /* Global cache of all hash join indices. The idex_reee_t's used for storage are a double linked list for LRU */
  id_hash_t *		hic_hashes;  /* hash from hi_signature_t  to index_tree_t */
  dk_hash_t *		hic_col_to_it; /* from member col id to dk_set_t  of hash inx's invalidated if col changes */
  dk_hash_t *		hic_pk_to_it; /* pk key table to dk_set_t of index_tree_t's of dependent hash indices. Invalidate if ins/del */
  dk_mutex_t *		hic_mtx;
  index_tree_t *	hic_first;
  index_tree_t *	hic_last;
} hash_index_cache_t;


extern hash_index_cache_t hash_index_cache;

#define ASSERT_IN_DBS(dbs) \
  ASSERT_IN_MTX (dbs->dbs_page_mtx)


struct index_tree_s
  {
    /* struct for any index tree, hash inx, sort temp or other set of pages that form a unit */
    dbe_storage_t *	it_storage; /* file group used for storage */
    dbe_key_t *		it_key; /* if index, this is the key */
    volatile dp_addr_t	it_root;
    buffer_desc_t * volatile	it_root_image;
    int volatile		it_root_image_version;
    unsigned short		it_root_version_ctr;
    char		it_is_single_page;
    buffer_desc_t *	it_root_buf;
    dk_mutex_t *	it_lock_release_mtx;
    int		it_fragment_no; /* always 0. I If horiz. fragmentation were supported, would be frg no. */
    hash_index_t * 	it_hi; /* Ifhash index */
    dp_addr_t	it_hash_first;
    char		it_shared;
    char		it_hi_isolation; /* if hash index, isolation used for filling this */
    int		it_ref_count; /* if hash inx, count of qi's using this */
    hi_signature_t *	it_hi_signature;
    dk_set_t 		it_waiting_hi_fill; /* if thi is a hash inx being filled, list of threads waiting for the fill to finish */
    index_tree_t *	it_hic_next; /* links for LRU queue of hash indices */
    index_tree_t *	it_hic_prev;
    long		it_last_used;
    int			it_hi_reuses; /* if hash inx, count of reuses */
    bitf_t		it_all_in_own_em:1;
    bitf_t		it_blobs_with_index:1;
    dp_addr_t		it_n_index_est; /* estimate of index pages */
    dp_addr_t		it_n_blob_est; /* estimate of blob pages */
    extent_map_t *	it_extent_map;
    it_map_t *		it_maps;
};


/* it_shared */
#define HI_PRIVATE 0
#define HI_OK 1
#define HI_RETRY 2
#define HI_FILL 3
#define HI_OBSOLETE 4


#define IN_TXN \
  mutex_enter (wi_inst.wi_txn_mtx);

#define LEAVE_TXN \
  mutex_leave (wi_inst.wi_txn_mtx)

#define ASSERT_IN_TXN \
  ASSERT_IN_MTX (wi_inst.wi_txn_mtx)

#define ASSERT_OUTSIDE_TXN \
  ASSERT_OUTSIDE_MTX (wi_inst.wi_txn_mtx)

#define IT_PAGE_IN_RANGE(it, x) \
  (((dp_addr_t) x) < it->it_storage->dbs_n_pages)

#define DBS_PAGE_IN_RANGE(dbs, x) \
  (((dp_addr_t) x) < dbs->dbs_n_pages)

#define IN_CPT_1 \
  mutex_enter (checkpoint_mtx)

#define LEAVE_CPT_1 \
  mutex_leave (checkpoint_mtx)

#define IN_CPT(lt) \
  do { \
    if (lt) { \
      IN_TXN; \
      lt_rollback (lt, TRX_CONT); \
      lt_threads_set_inner (lt, 0); \
      LEAVE_TXN; \
    } \
    IN_CPT_1; \
  } while (0)

#define LEAVE_CPT(lt) \
  do { \
       if (lt) { \
         IN_TXN; \
	 lt_threads_set_inner (lt, 1); \
         LEAVE_TXN; \
       } \
       LEAVE_CPT_1; \
     } while (0)

/* inlined comparison funcs */


typedef struct cmp_desc_s
{
    char	cmd_min_op;
  char		cmd_max_op;
  dtp_t		cmd_dtp;
  char		cmd_non_null;
} cmp_desc_t;


typedef struct cmp_func_desc_s
{
  key_cmp_t 	cfd_func;
  cmp_desc_t *	cfd_compares;
} cmp_func_desc_t;

struct search_spec_s
  {
    char		sp_is_boxed; /* always 1, not used */
    char		sp_min_op; /* compare operator for lower bound */
    char		sp_max_op; /* if this is a range match, compare op for upper bound */
    char		sp_is_reverse; /* true if inserting a DESC sorted item */
    short			sp_min;  /* index into itc_search_params */
    short			sp_max; /* ibid */
    search_spec_t *	sp_next;
    dbe_col_loc_t	sp_cl;  /* column on key, if key on page matches key in compilation */
    dbe_column_t *	sp_col; /* col descriptor, use for finding the col if key on page is obsolete */
    struct state_slot_s *sp_min_ssl;  /* state slot for initing   the cursor's  itc_search_params[sp_min] */
    struct state_slot_s *sp_max_ssl;
    collation_t	 *sp_collation;
    char		sp_like_escape;
  };


typedef struct out_map_s
{
  dbe_col_loc_t	om_cl;
  char 		om_is_null;
} out_map_t;


#define OM_NULL 1
#define OM_ROW 2
#define OM_BM_COL 3
    /* flags for page_wait_access, itc_dive_mode  */
#define PA_READ 0 /* code relies on 0 being PA_READ, as per result of memset 0 */
#define PA_WRITE 1


#define ITC_STORAGE(itc) (itc)->itc_space->isp_tree->it_storage

#define ITC_AT_END -1 /* itc_map_pos when at end of buffer in derection of read */
#define ITC_DELETED -2 /* during page rewrite, just delete, goes to next non-delete or to end if none */

    /* placeholder_t i the common superclass of a placeholder, a bookmark whose position survives index updates, and of the index tree cursor */

#define PLACEHOLDER_MEMBERS \
  bitf_t		itc_type:3;  \
  bitf_t		itc_is_on_row:1; \
  bitf_t		itc_is_registered:1;	\
  bitf_t		itc_desc_order:1; /* true if reading index from end to start */ \
  char			itc_lock_mode; \
  short			itc_map_pos; \
  volatile dp_addr_t		itc_page; \
  dp_addr_t		itc_owns_page;  /* cache last owned lock */ \
  buffer_desc_t *	itc_buf_registered; \
  it_cursor_t *		itc_next_on_page; \
  index_tree_t *	itc_tree; \
  bitmap_pos_t		itc_bp

#define itc_row_no itc_bp.bp_value

#define ITC_PLACEHOLDER_BYTES \
	((int)(ptrlong)(&((it_cursor_t *)0x0)->itc_to_reset))

/* itc_type */
#define ITC_PLACEHOLDER	0
#define ITC_CURSOR	1

/* itc_search_mode */
#define SM_INSERT	1
#define SM_READ		0
#define SM_READ_EXACT	2
#define SM_TEXT 4
#define SM_INSERT_BEFORE 5
#define SM_INSERT_AFTER 6

struct placeholder_s
  {
    PLACEHOLDER_MEMBERS;
  };


typedef void (*itc_clup_func_t) (it_cursor_t *);

#define MAX_SEARCH_PARAMS TB_MAX_COLS + 10

#define RA_MAX_ROOTS 80

#define SQLO_RATE_NAME "rnd-stat-rate"

typedef enum { RANDOM_SEARCH_OFF = 0, RANDOM_SEARCH_ON = 1, RANDOM_SEARCH_AUTO = 2 ,
RANDOM_SEARCH_COND = 3} random_search_mode;


struct it_cursor_s
  {
    PLACEHOLDER_MEMBERS;

    char		itc_to_reset; /* what level of change took place while itc waited for page buffer */
    char		itc_max_transit_change; /* quit waiting and do not enter the buffer if itc_transit_change >= this */
    char		itc_acquire_lock; /*when wait over want to own it? */
    char 		itc_search_mode; /* unique match or not */
    char		itc_isolation;
    unsigned char	itc_key_spec_nth;
    char		itc_has_blob_logged:3; /*if blob to log, can't drop blob when inlining it until commit */
    char		itc_random_search:3;
    bitf_t		itc_is_allocated:1;
    bitf_t		itc_dive_mode:1;
    bitf_t		itc_at_data_level:1;
    bitf_t		itc_landed:1; /* true if found position on or between leaves, false if in initial descent through the tree */
    bitf_t		itc_no_bitmap:1;  /* ignore bitmap logic if on bitmap inx */
    bitf_t		itc_desc_serial_landed:1; /* if set, failure to get first lock (right above the selected range) resets search */
    bitf_t		itc_desc_serial_reset:1;
    bitf_t		itc_is_vacuum:1;
    bitf_t		itc_ac_parent_deld:1; /* set by autocompact to indicate that the parent page was popped off because of having only one leaf left */
    bitf_t		itc_cl_results:1; /* in cluster server, send stuff in out map to the client node */
    bitf_t		itc_cl_local:1; /* if cluster but running local */
    bitf_t		itc_cl_batch_done:1; /* set if reset due ti batch done */
    bitf_t		itc_cl_set_done:1;
    bitf_t		itc_cl_from_temp:1; /* last search param is the id of the qf with the setp and the temp data */
    bitf_t		itc_cl_qf_any_passed:1; /* in cluster query frag output itc, used to know if nulls hould be sent in oj */
    bitf_t		itc_must_kill_trx:1;
    unsigned char	itc_search_par_fill;
    unsigned char	itc_owned_search_par_fill;
    unsigned char	itc_pars_from_end; /* no of places in search params used for temp cast search pars */
    short		itc_hash_buf_fill;
    short		itc_hash_buf_prev;
    short			itc_write_waits; /* wait history. Use for debug */
    short			itc_read_waits;
    short			itc_n_lock_escalations; /* no of times row locks escalated to page locks on this read.  Used for claiming page lock as first choice after history of escalating */

    /* dp_addr_t		itc_parent_page; */
    int			itc_n_pages_on_hold; /* if inserting, amount provisionally reserved for deltas made by tree split */

    it_map_t *		itc_itm1;  /* points to the iot_map_t if this itc holds the it_map_t's itm_mtx */
    it_map_t *		itc_itm2;
    jmp_buf_splice *	itc_fail_context; /* throw when deadlock or other exception inside index operation */

    du_thread_t *	itc_thread;
    it_cursor_t *	itc_next_waiting; /* next itc waiting on same buffer_desc_t's read/write gate */

    lock_trx_t *	itc_ltrx;
    it_cursor_t *	itc_next_on_lock; /* next itc waiting on same lock */
    struct page_lock_s * itc_pl; /*page_lock_t  of the current page */

    dbe_key_t *		itc_insert_key; /* Key n which operation takes place */
    key_spec_t		itc_key_spec; /* search specs used for indexed lookup */
    search_spec_t *	itc_row_specs; /* earch specs for checking  rows where itc_specs match */
    out_map_t *		itc_out_map;  /* one for each out ssl of the itc_ks->ks_out_slots */
    search_spec_t *	itc_bm_col_spec; /* if set, this is the indexable condition on the bitmapped col */

    db_buf_t		itc_row_data; /* pointer in mid page buffer , where the itc's rows data starts */
    dbe_key_t *		itc_row_key;
    placeholder_t *	itc_bm_split_left_side;
    /* hash index */
    buffer_desc_t *	itc_buf; /* cache the buffer when keeping buffer wired down between rows.  Can be done because always read only.  */
    buffer_desc_t * 	itc_hash_buf; /* when filling a hash, the last buffer, constantly wired down, can be because always oen writer */
    struct word_stream_s *	itc_wst; /* for SM_TEXT search mode */
    caddr_t *		itc_out_state;  /* place out cols here. If null copy from itc_in_state */
    struct key_source_s *	itc_ks;

    /* data areas. not cleared at alloc */
    caddr_t		itc_search_params[MAX_SEARCH_PARAMS];
    caddr_t		itc_owned_search_params[MAX_SEARCH_PARAMS];
    extent_map_t *	itc_hold_em; /* if pages on hold, record where so they can be returned if the em changes */
    short			itc_ra_root_fill;
    int			itc_n_reads;
    int			itc_nth_seq_page; /* in sequential read, nth consecutive page entered.  Use for starting read ahead.  */

    placeholder_t *	itc_bm_split_right_side;
    int			itc_root_image_version;
    char		itc_bm_spec_replaced; /* true if bm inx dive set the key_spec */
    char		itc_cl_org_desc;
    dp_addr_t		itc_ra_root[RA_MAX_ROOTS];
    buffer_desc_t *	itc_buf_entered; /* this is set to the entered buf when another thread enters this itc into a buf as a result of page_leave_inner on that other thread */
    key_spec_t 	itc_cl_org_spec;

    struct {
      int	sample_size;  /* stop random search after this many rows */
      int	n_sample_rows; /* count of rows retrieved in random traversal */
      dk_hash_t *	cols;	/* hash from de_col_t to col_stat_t *for random sample col stats. */
    } itc_st;
  };





#define ITC_NULL_CK(itc, cl) \
  (itc->itc_row_data[cl.cl_null_flag[IE_ROW_VERSION (itc->itc_row_data)]] & cl.cl_null_mask[IE_ROW_VERSION (itc->itc_row_data)])

#define ROW_INT_COL(buf, row, rv, cl, ref, n) \
{\
  short __off = (cl).cl_pos[rv];\
  if ((cl).cl_row_version_mask & rv)\
    {\
      unsigned short __irow2 = SHORT_REF (row + __off);\
      db_buf_t __row2 = buf->bd_buffer + buf->bd_content_map->pm_entries[__irow2 & ROW_NO_MASK];\
      __off = (cl).cl_pos[IE_ROW_VERSION(__row2)];\
      n = ref (__row2 + __off) + (__irow2 >> COL_OFFSET_SHIFT);\
    }\
  else \
    n =ref(row + __off); \
}


/* when searching, do you set a lock before returning hit?  Note that serializable sets the lock before checking the hit */
#define itc_lock_after_match(itc) \
  ((ISO_REPEATABLE == itc->itc_isolation || (PL_EXCLUSIVE == itc->itc_lock_mode && itc->itc_isolation == ISO_COMMITTED)) \
   && itc->itc_page != itc->itc_owns_page)
#define ROW_FIXED_COL(buf, row, rv, cl, ptr) \
{\
  short __off = (cl).cl_pos[rv];\
  if ((cl).cl_row_version_mask & rv)\
    {\
      int __irow2 = SHORT_REF (row + __off);\
      db_buf_t __row2 = buf->bd_buffer + buf->bd_content_map->pm_entries[__irow2 & ROW_NO_MASK];\
      __off = (cl).cl_pos[IE_ROW_VERSION(__row2)];\
      ptr = __row2 + __off;\
    }\
  else \
    ptr = row + (cl).cl_pos[rv];\
}


#define ROW_STR_COL(key, buf, row, cl, p1, l1, p2, l2, offset) \
  kc_var_col (key, buf, row, cl, &p1, &l1, &p2, &l2, &offset)

#define KEY_PRESENT_VAR_COL(key, row, cl, off, len)\
{\
  row_ver_t rv = IE_ROW_VERSION (row);\
  len = (cl).cl_pos[rv];\
  if (CL_FIRST_VAR == len)\
    {\
      len = SHORT_REF (row + key->key_length_area[rv]);\
      off = IE_KEY_VERSION (row) ? key->key_row_var_start[rv] : key->key_key_var_start[rv];\
      len -= off;\
    }\
  else \
    {\
      off = SHORT_REF (row - len) & COL_VAR_LEN_MASK;\
      len = SHORT_REF (row + 2 - len) - off;\
    }\
}


#define ITC_PRESENT_VAR_COL(itc, cl, off, le) \
  KEY_PRESENT_VAR_COL (itc->itc_insert_key, itc->itc_row_data, cl, off, len)

#define ROW_LENGTH(row, key, len) \
len = row_length (row, key)


#if 0
/* itc_extension_state - not in use */
#define REXT_NO_EXTENSION	0
#define REXT_UNREAD_EXTENSION	1
#define REXT_EXTENSION		2
#endif
/* itc_acquire_lock */
#define ITC_NO_LOCK 0
#define ITC_GET_LOCK 1
#define ITC_LOCK_IF_ON_ROW 2

/* itc_to_reset - unsigned char, order is important */
/* when calling page_wait_access, the itc_max_transit_change is one of these.
 * If the change during wait is greater than indicated here, the itc does not enter the buffer.
 * For example if the page of the buffer  splits, the itc will not know whether it still wants to enter the buffer and must restart the search.
 * When the wait is over, itc_to_reset is set to reflect what happened during the wait, again one of the below */

#define RWG_WAIT_NO_ENTRY_IF_WAIT 0 /* Just check if buffer available, wait until is but do not go in */
#define RWG_NO_WAIT	1 /* Only go in if immediately available */
#define RWG_WAIT_NO_CHANGE 2
#define RWG_WAIT_DISK	3
#define RWG_WAIT_DATA	4 /* Data change but no split. */
#define RWG_WAIT_KEY	5 /* insert/delete  but no split */
#define RWG_WAIT_SPLIT	6 /* page split or delete */
#define RWG_WAIT_DECOY 7 /* waited for a decoy buffer which was not replaced by a real buffer.  Retry buffer lookup */
#define RWG_WAIT_ANY 8 /* Means get the buffer no matter what changes during read */


#define ITC	it_cursor_t *


#define ASSERT_BUFF_WIRED(it,buf) do { \
  if ((it)->itc_page != (buf)->bd_page) \
    GPF_T1 ("it->itc_page != buf->bd_page"); \
  if ((it)->itc_landed) \
    { \
      if (!(buf)->bd_is_write) \
	GPF_T1 ("Landed search w/o bd_is_write"); \
    } \
  else \
    { \
      if (!(buf)->bd_readers) \
	GPF_T1 ("Non-landed search w/o db_readers"); \
    } \
} while (0)



#define ITC_INIT(itc, isp, trx) \
  memset ((ITC) itc, 0, ((ptrlong) &(itc)->itc_search_params) - ((ptrlong) itc)); \
  itc->itc_type = ITC_CURSOR; \
  itc->itc_ltrx = trx; \
  itc->itc_lock_mode = PL_SHARED; \

#define ITC_START_SEARCH_PARS(it) \
  it->itc_search_par_fill =0, \
    it->itc_owned_search_par_fill =0,		\
    it->itc_pars_from_end = 0

#define ITC_SEARCH_PARAM(it, par) \
{ \
  it->itc_search_params[it->itc_search_par_fill++] =  ((caddr_t) (par)); \
}



/* when cast of search param to column type makes a new box, it is registered with this so that it is freed with the itc */
#define ITC_OWNS_PARAM(it, par) \
  it->itc_owned_search_params[it->itc_owned_search_par_fill++] = par

/* Relation of itc and page mtxs */


#ifdef MTX_DEBUG
#define mtx_assert(a) assert ((a))
#else
#define mtx_assert(a)
#endif


/* this means that if the page to be entered looks like a leaf, the access mode is preferentially exclusive */
#define ADAPTIVE_LAND
/* the convention for transits is to enter the mtx at the lower address first.  This becomes itc->itm1, the other is itc->ittc_itm2 */

#define ITC_IN_TRANSIT(itc, from_dp, to_dp)\
{\
  it_map_t * itm1 = IT_DP_MAP (itc->itc_tree, from_dp);\
  it_map_t * itm2 = IT_DP_MAP (itc->itc_tree, to_dp);\
  if (itm1 == itm2)\
    {\
      if (itc->itc_itm2 || (itc->itc_itm1 && itc->itc_itm1 != itm1)) GPF_T1 ("single map transit tried while other transit in effect"); \
      itc->itc_itm2 = NULL;\
      if (itc->itc_itm1 != itm1) \
        mutex_enter (&itm1->itm_mtx);\
      itc->itc_itm1 = itm1;\
    }\
  else if (itm1 < itm2)\
    {\
      if (itc->itc_itm1 && itc->itc_itm1 != itm1) GPF_T1 ("entering different transit from that in effect"); \
      if (itc->itc_itm2 && itc->itc_itm2 != itm2) GPF_T1 ("entering different transit from that in effect"); \
      if (!itc->itc_itm1) \
	{ \
	  itc->itc_itm1 = itm1;			\
	  itc->itc_itm2 = itm2;			\
	  mutex_enter (&itm1->itm_mtx);		\
	  mutex_enter (&itm2->itm_mtx);		\
	}\
    }					\
  else \
    {\
      if (itc->itc_itm1 && itc->itc_itm1 != itm2) GPF_T1 ("entering different transit from that in effect"); \
      if (itc->itc_itm2 && itc->itc_itm2 != itm1) GPF_T1 ("entering different transit from that in effect"); \
      if (!itc->itc_itm1) \
	{ \
	  itc->itc_itm1 = itm2;			\
	  itc->itc_itm2 = itm1;			\
	  mutex_enter (&itm2->itm_mtx);		\
	  mutex_enter (&itm1->itm_mtx);		\
	} \
    }\
\
}


#define ITC_ASSERT_TRANSIT(itc, dp1, dp2) \\
{\
  it_map_t itm1 = IT_DP_MAP (itc->itc_tree, dp1);\
  it_map_t itm2 = IT_DP_MAP (itc->itc_tree, dp2);\
  if (itm1 == itm2)\
    {\
      ASSERT_IN_MTX (&itm1->itm_mtx);\
      mtx_assert (itc->itc_itm1 == itm1\
	      && itc->itc_itm2 == NULL);\
    }\
  else\
    {\
      ASSERT_IN_MTX (&itm1->itm_mtx);\
      ASSERT_IN_MTX (&itm2->itm_mtx);\
      mtx_assert ((itc->itc_itm1 == itm1 && itc->itc_itm2 == itm2)\
	      || (itc->itc_itm1 == itm2 && itc->itc_itm2 == itm1));\
    }\
}


#define IT_ASSERT_TRANSIT(it, dp1, dp2) \
{\
  it_map_t * itm1 = IT_DP_MAP (it, dp1);\
  it_map_t * itm2 = IT_DP_MAP (it, dp2);\
  if (itm1 == itm2)\
    {\
      ASSERT_IN_MTX (&itm1->itm_mtx);\
    }\
  else\
    {\
      ASSERT_IN_MTX (&itm1->itm_mtx);\
      ASSERT_IN_MTX (&itm2->itm_mtx);\
    }\
}


#define ITC_LEAVE_MAPS(itc)\
{\
  if (itc->itc_itm1)\
    {\
      mutex_leave (&itc->itc_itm1->itm_mtx);\
      itc->itc_itm1 = NULL;\
    }\
  if (itc->itc_itm2)\
    {\
      mutex_leave (&itc->itc_itm2->itm_mtx);\
      itc->itc_itm2 = NULL;\
    }\
}


#define ITC_LEAVE_MAP_NC(itc) \
{ \
  mtx_assert (itc->itc_itm1 && !itc->itc_itm2); \
  mutex_leave (&itc->itc_itm1->itm_mtx); \
  itc->itc_itm1 = NULL; \
}


#define ASSERT_IN_MAP(it, dp) \
{\
  ASSERT_IN_MTX (&IT_DP_MAP (it, dp)->itm_mtx);	\
}


#define ASSERT_OUTSIDE_MAP(it, dp) \
  ASSERT_OUTSIDE_MTX (&IT_DP_MAP (it, dp)->itm_mtx)

#define ASSERT_OUTSIDE_MAPS(itc)\
  mtx_assert (!itc->itc_itm1 && !itc->itc_itm2)	\


#define ITC_IN_VOLATILE_MAP(itc, dp)\
{\
  ASSERT_OUTSIDE_MAPS (itc);\
  for (;;)\
    {\
      dp_addr_t __to = dp;\
      it_map_t * itm = IT_DP_MAP (itc->itc_tree, __to);\
      mutex_enter (&itm->itm_mtx);\
      if (__to == dp)\
	{\
	  itc->itc_itm1 = itm;\
	  break;\
	}\
      mutex_leave (&itm->itm_mtx);\
      TC (tc_dp_changed_while_waiting_mtx);\
    }\
}\


#define IN_VOLATILE_MAP(it, dp)\
{\
  for (;;)\
    {\
      dp_addr_t __to = dp;\
      it_map_t * itm = IT_DP_MAP (it, __to);\
      mutex_enter (&itm->itm_mtx);\
      if (__to == dp)\
	{\
	  break;\
	}\
      mutex_leave (&itm->itm_mtx);\
      TC (tc_dp_changed_while_waiting_mtx);\
    }\
}\





#define ITC_IN_KNOWN_MAP(itc, dp)\
{\
  it_map_t * itm = IT_DP_MAP (itc->itc_tree, dp);\
  mtx_assert (!itc->itc_itm2); \
  mtx_assert (!itc->itc_itm1 || itc->itc_itm1 == itm);	\
  if (!itc->itc_itm1) \
    mutex_enter (&itm->itm_mtx);\
  itc->itc_itm1 = itm;\
}


#define ITC_IN_OWN_MAP(itc) ITC_IN_KNOWN_MAP ((itc), (itc)->itc_page)


/* Page Content Map */

#define PM_MAX_ENTRIES	 (PAGE_DATA_SZ / 4)


struct page_map_s
  {
    /* For a buffer with index tree content, this struct  holds the starting positions of all entries in index order plus avail. space */
    short	pm_size; /* number of entries actually in pm_entries.  Different sizes of map are allocated for different buffers since having a page full of minimum length entries is very rare */
    short		pm_count;  /* count of rows. This many first entries in pm_entries are valid */
    short		pm_filled_to; /* First free byte of the page's trailing contiguous free space */
    short		pm_bytes_free; /* count of free bytes, including gaps.  If a row of this or less size is inserted, it will fit, maybe needing page compaction */
    short		pm_n_non_comp; /*n inserts that have not been checked for compressible cols */
    short		pm_entries[PM_MAX_ENTRIES]; /* the start offsets of the index entries  on the page */
  };


#define PM_ENTRIES_OFFSET ((int) (ptrlong) &((page_map_t *)0)->pm_entries)

#define DO_ROWS(buf, map_pos, row, key)		\
{ \
  int map_pos; \
  for (map_pos = 0; map_pos < buf->bd_content_map->pm_count; map_pos++) \
  {\
    db_buf_t row = buf->bd_buffer + buf->bd_content_map->pm_entries[map_pos];

#define END_DO_ROWS } }



/* the different standard sizes of page_map_t */
#define PM_SZ_1 50
#define PM_SZ_2 200
#define PM_SZ_3 700
#define PM_SZ_4 (PM_MAX_ENTRIES)


#define PM_SIZE(ct) \
  (ct < PM_SZ_1 ? PM_SZ_1 : (ct < PM_SZ_2 ? PM_SZ_2 : (ct < PM_SZ_3 ? PM_SZ_3 : PM_SZ_4)))
extern resource_t * pm_rc_1;
extern resource_t * pm_rc_2;
extern resource_t * pm_rc_3;
extern resource_t * pm_rc_4;

#define PM_RC(sz) (sz == PM_SZ_1 ? pm_rc_1 : (sz == PM_SZ_2 ? pm_rc_2 : (sz == PM_SZ_3 ? pm_rc_3 : (sz == PM_SZ_4 ? pm_rc_4 : (resource_t *)(GPF_T1("not a valid pm size"), NULL)))))





#define ITC_REAL_ROW_KEY(itc) \
{ \
  itc->itc_row_key = itc->itc_insert_key->key_versions[IE_KEY_VERSION (itc->itc_row_data)];\
}


/*#define BUF_DEBUG*/
#define buf_dbg_printf(a) /*printf a*/

#define bd_readers bdf.r.readers
#define bd_is_write bdf.r.is_write
#define bd_being_read bdf.r.being_read
#define bd_is_dirty bdf.r.is_dirty
#define bd_is_ro_cache bdf.r.is_ro_cache

#if defined (MTX_DEBUG) && !defined (PAGE_DEBUG)
#define PAGE_DEBUG
#endif

#ifdef NDEBUG
#undef PAGE_DEBUG
#endif

struct buffer_desc_s
{
  /* Descriptor of a page buffer.  Read/write gate and other fields */
  union {
    int64	flags; /* allow testing for all 0's with a single compare.  All zeros mens candidate for reuse. */
    struct {
      short readers; /* count of threads with read access */
      /* the below flagss are chars and not bit fields.  If bit fields, there is a read+write for setting and cache coherence will not protect against a change of a bit between the read and the write.  So for cache coherency the flags must be individually settable so that you do not end up setting neighbor flags to their former values.  The write will hit an obsolete cache line and will reload the line but the other bits will still come from the read that was done before the change.  */
      char	is_write;  /* if any thread the exclusive owner of this */
      char	being_read; /* is the buffer allocated for a page and awaiting the data coming from disk */
      char	is_dirty; /* Content changed since last written to disk */
      char	is_ro_cache;
      char	is_read_aside;
    } r;
  } bdf;
  bp_ts_t		bd_timestamp; /* Timestamp for estimating age for buffer reuse */
  it_cursor_t *	bd_read_waiting;  /* list of cursors waiting for read access */
  it_cursor_t *	bd_write_waiting; /* itc waiting for write access */

  db_buf_t		bd_buffer; /* the 8K bytes for the page */
  page_map_t *	bd_content_map; /* only if content is an index page */
  page_lock_t *	bd_pl; /* if lock associated, it's cached here in addition to the tree's hash */

  union {
    buffer_desc_t *	next; /* Link to next if this is in free set or inc backup set.  If regular buffer, this is a link to the next unused if this buffer is unused, else null */
    it_cursor_t *	registered;
  } bn;
  dp_addr_t		bd_page; /* The logical page number */
  dp_addr_t		bd_physical_page; /* The physical page number, can be different from bd_page if remapped */

  buffer_pool_t *	bd_pool;
  index_tree_t *	bd_tree; /* when caching a page, this is  the index tree  to which the page belongs */
  dbe_storage_t * 	bd_storage; /* the storage unit for reading/writing the page */
  io_queue_t *	 bd_iq; /* iq, if buffer in queue for read(write */
  buffer_desc_t *	bd_iq_prev; /* next and prev in double linked list of io queue */
  buffer_desc_t *	bd_iq_next;
#ifdef PAGE_DEBUG
  du_thread_t *	bd_writer; /* for debugging, the thread which has write access, if any */
  char * 		bd_enter_file;
  long 			bd_enter_line;
  char * 		bd_leave_file;
  long 			bd_leave_line;
  char 			bd_el_flag;	/* what operation was last: 1-enter, 2-leave */
  char *                bd_set_wr_file;
  long                  bd_set_wr_line;
  thread_t *		bd_thr_el;
#endif
#ifdef PAGE_TRACE
  long		bd_trx_no;
#endif
#ifdef BUF_DEBUG
  index_tree_t *	bd_prev_tree;
#endif
};

#define BUF_ROW(buf, pos) ((buf)->bd_buffer + (buf)->bd_content_map->pm_entries[pos])

#ifdef PAGE_DEBUG
#define BUF_DBG_ENTER_1(buf, __file, __line) \
    do { \
      if (buf) { \
	thread_t * __self = THREAD_CURRENT_THREAD; \
	(buf)->bd_enter_file = __file; \
	(buf)->bd_enter_line = __line; \
	(buf)->bd_el_flag = 1; \
	(buf)->bd_thr_el = __self; \
	if (!__self->thr_pg_dbg) \
	  __self->thr_pg_dbg = (void *) hash_table_allocate (31); \
	sethash ((void *)(buf), (dk_hash_t *) __self->thr_pg_dbg, (void*)(ptrlong)(buf)->bd_page); \
      } \
    } while (0)
#define BUF_DBG_LEAVE_1(buf, __file, __line) \
    do { \
      if (buf) { \
	thread_t * __self = THREAD_CURRENT_THREAD; \
	if (__self->thr_pg_dbg) { \
	  remhash ((void*) (buf), (dk_hash_t *) __self->thr_pg_dbg); \
	} else if ((buf)->bd_el_flag == 1) \
	  log_error ("Page debug info missing at %s:%ld, entered at %s:%ld", __file, __line, (buf)->bd_enter_file, (buf)->bd_enter_line); \
	(buf)->bd_leave_file = __file; \
	(buf)->bd_leave_line = __line; \
	(buf)->bd_el_flag = 2; \
	(buf)->bd_thr_el = __self; \
      } \
    } while (0)
#define BUF_DBG_LEAVE(buf) BUF_DBG_LEAVE_1((buf), file, line)
#define BUF_DBG_LEAVE_INL(buf) BUF_DBG_LEAVE_1((buf), __FILE__, __LINE__)
#define BUF_DBG_ENTER(buf) BUF_DBG_ENTER_1((buf), file, line)
#define BUF_DBG_ENTER_INL(buf) BUF_DBG_ENTER_1((buf), __FILE__, __LINE__)

#define THR_DBG_PAGE_CHECK \
  do \
  { \
    thread_t * self = THREAD_CURRENT_THREAD; \
    dk_hash_iterator_t hit; \
    buffer_desc_t * buf; \
    ptrlong page; \
    dk_hash_iterator (&hit, (dk_hash_t *) self->thr_pg_dbg); \
    while (NULL != self->thr_pg_dbg && dk_hit_next (&hit, (void**) &buf, (void**) &page)) \
      { \
	if (buf && buf->bd_tree && buf->bd_tree->it_key->key_id == KI_TEMP) continue; \
	GPF_T1 ("Buffer left occupied after thread is done"); \
      } \
    if (NULL != self->thr_pg_dbg) clrhash ((dk_hash_t *) self->thr_pg_dbg); \
  } \
  while (0)

#else
#define BUF_DBG_ENTER(buf)
#define BUF_DBG_ENTER_INL(buf)
#define BUF_DBG_LEAVE(buf)
#define BUF_DBG_LEAVE_INL(buf)
#define THR_DBG_PAGE_CHECK
#endif

#define bd_registered bn.registered
#define bd_next bn.next
#define BUF_AGE(buf) (buf->bd_pool->bp_ts - buf->bd_timestamp)

/* mark as recently used */
#define BUF_TOUCH(buf) \
{ \
  (buf)->bdf.r.is_read_aside = 0; \
  (buf)->bd_timestamp = (buf)->bd_pool->bp_ts;		\
  if ((bp_hit_ctr++ & 0x1f) == 0)			\
    (buf)->bd_pool->bp_ts++;				\
}


#define BUF_TICK(buf) buf->bd_pool->bp_ts++;

#define BUF_NONE_WAITING(buf) \
(!buf->bd_write_waiting && !buf->bd_read_waiting && !buf->bd_being_read)

#ifdef PAGE_DEBUG
#define BD_SET_IS_WRITE(bd, f) \
{ \
  (bd)->bd_is_write = f;			    \
  (bd)->bd_writer = f ? THREAD_CURRENT_THREAD : NULL;	\
  (bd)->bd_set_wr_file = __FILE__; \
  (bd)->bd_set_wr_line = __LINE__; \
}
#else
#define BD_SET_IS_WRITE(bd, f) \
  (bd)->bd_is_write = f

#endif


#ifdef BUF_BOUNDS
extern buffer_desc_t * bounds_check_buf;

#define BUF_BOUNDS_CHECK(buf)  \
{ \
  unsigned short flags = SHORT_REF (buf->bd_buffer + DP_FLAGS); \
  if (DPF_INDEX == flags && !buf->bd_content_map) { bounds_check_buf = buf; GPF_T1 ("inx buffer without map");}; \
  if (flags >= DPF_LAST_DPF) { bounds_check_buf = buf; GPF_T1 ("bad dp_flags"); };  \
  if (BUF_END_MARK != LONG_REF (buf->bd_buffer + PAGE_SZ)) { bounds_check_buf = buf; GPF_T1 ("bad buffer end mark"); } \
}
#define BUF_ALLOC_SZ (PAGE_SZ + sizeof (int32))
#define BUF_END_MARK 0xfeedbeef
#define BUF_SET_END_MARK(buf) LONG_SET (buf->bd_buffer + PAGE_SZ, BUF_END_MARK)
#else
#define BUF_BOUNDS_CHECK(buf)
#define BUF_ALLOC_SZ PAGE_SZ
#define BUF_SET_END_MARK(buf)
#endif





#define PFH_N_WAYS 29
#define PFH_MAX_COLS 5
#define PFH_N_SHORTS 4096
#define PFH_KV_ANY 255 /* in pfh_)kv when the kv is not yet set */


struct pf_hash_s
{
  short		pfh_start[PFH_MAX_COLS][PFH_N_WAYS];
  key_ver_t	pfh_kv; /* only this kv's keys are al;lowed in */
  short		pfh_hash[PFH_N_SHORTS];
  short		pfh_fill;
  short		pfh_n_cols; /* last cl_nth +1 that is inited */
  db_buf_t	pfh_page;
  page_fill_t *	pfh_pf;
};



struct row_fill_s
{
  db_buf_t	rf_row;
  db_buf_t	rf_large_row;
  row_size_t	rf_space;
  row_size_t	rf_fill;
  short		rf_map_pos;
  dbe_key_t *	rf_key;
  pf_hash_t *	rf_pf_hash; /* if doing compressing copy, mark the places that have full values here */
  char		rf_is_leaf;
  char		rf_no_compress; /* do not try col compression */
};


struct  page_fill_s
{
  it_cursor_t *	pf_itc;
  dk_set_t	pf_left; /* if more than one buffers result, the leftmost is first, then the rest, except for the current */
  buffer_desc_t *	pf_org; /* use this to decode compression if getting stuff as rows */
  buffer_desc_t *	pf_current;
  pf_hash_t *		pf_hash;
  row_lock_t **	pf_rls;
  placeholder_t **	pf_registered;
  int		pf_rl_fill;
  int		pf_cr_fill;
  char		pf_is_autocompact; /* when splitting, do not alloc real pages, just bufs with no disk page */
  char		pf_op;
  char		pf_rewrite_overflow;
  int		pf_dbg;
};


#define LOCAL_RF(rf, row, space, key)		\
  row_fill_t rf; \
  memset (&rf, 0, sizeof (rf)); \
  rf.rf_row = row;\
  rf.rf_space = space;\
  rf.rf_fill = key->key_row_var_start[0];\
  rf.rf_key = key;



#define RF_LARGE_CHECK(rf, off, len)\
{\
  int __off = off ? off : rf->rf_fill;\
  /*if (!!rf->rf_no_large && rf->rf_large_row) GPF_T1 ("rf_large_row not set"); */ \
  if (__off + len > MAX (rf->rf_space, MAX_ROW_BYTES)) GPF_T1 ("row fill overflow max bytes"); \
  if (__off + len > rf->rf_space)\
    {\
      memcpy (rf->rf_large_row, row, rf->rf_fill);\
      if (__off > rf->rf_fill) memset (rf->rf_large_row + rf->rf_fill, 0, __off - rf->rf_fill); \
      /* set at least __off worth but do not read more than rf_fill worth.  For valgrind. */ \
      row = rf->rf_row = rf->rf_large_row;\
      rf->rf_space = MAX_ROW_BYTES;\
    }\
}



struct row_delta_s
{
  char		rd_op;
  char		rd_make_ins_rbe; /* when plain on page, make ins rollback entry? */
  char		rd_copy_of_deleted; /* when writing a page with uncommitted deletes */
  char		rd_raw_comp_row; /* when copying and it is known that compression stays the same, rd_values is the row string */
  key_ver_t	rd_key_version; /* use this to see if left dummy or such */
  short		rd_map_pos;
  row_size_t	rd_non_comp_len;
  char		rd_is_double_lp; /* is 1st of a double leaf pointer in split? Special case with reg'd itcs on parent */
  char		rd_any_ser_flags;
  int		rd_non_comp_max;
  dp_addr_t		rd_leaf; /* if lp, if upd or ins concerns leaf ptr */
  dbe_col_loc_t **	rd_upd_change;
  dbe_key_t *		rd_key;
  caddr_t *		rd_values; /* if ins or upd, the values */
  short			rd_n_values;
  short			rd_temp_fill;
  short			rd_temp_max;
  char			rd_allocated;
  bitf_t		rd_cl_blobs_at_store:1; /* do not make blobs when reading the rd from cluster peer */
  db_buf_t		rd_temp; /* when copying, scratch space for non-allocd box images */
  row_lock_t *		rd_rl;
  it_cursor_t * 	rd_itc;
  caddr_t *		rd_qst;
  it_cursor_t *		rd_keep_together_itcs;
  dp_addr_t		rd_keep_together_dp;
  short			rd_keep_together_pos;
};

/* rd_allocated */
#define RD_AUTO 1
#define RD_ALLOCATED_VALUES 2
#define RD_ALLOCATED 3

/* rd_op */
#define RD_INSERT 1 /* 1 row, goes to rd_map_pos */
#define RD_DELETE 2 /* row at map pos rd_pos is deleted */
#define RD_UPDATE 3 /* row at rd_map_pos is replaced */
#define RD_LEAF_PTR 4 /* lp of lp at map pos rd_pos gets set to rd_leaf_ptr */
#define RD_LEFT_DUMMY 5
#define RD_UPDATE_LOCAL 6 /* replace of a row that does not affect any compressible */


#define MAX_ITCS_ON_PAGE 1000

typedef struct page_apply_frame_s
{
  placeholder_t *	paf_registered[MAX_ITCS_ON_PAGE];
  row_lock_t *	paf_rlocks[PM_MAX_ENTRIES];
  buffer_desc_t	paf_buf;
  page_map_t	paf_map;
  row_delta_t	paf_rd;
  dtp_t		paf_page[PAGE_SZ];
  caddr_t	paf_rd_values[TB_MAX_COLS];
  dtp_t paf_rd_temp[2 * MAX_ROW_BYTES];
} page_apply_frame_t;


struct io_queue_s
  {
    /* io queue.  One should exist for each independently addressable device.
     * Used for ascending order background flush and read ahead of buffers */
    caddr_t		iq_id;
    buffer_desc_t *	iq_first; /* firstof the double linked list of buffers in the queue. */
    buffer_desc_t *	iq_last;
    buffer_desc_t *	iq_current;
    semaphore_t *		iq_sem; /* the io server thread waits on this between batches of pages to be read(written */
    dk_mutex_t * 	iq_mtx; /* serializes access to the buffers list */
    dk_set_t	iq_waiting_shut; /* list of threads waiting for all activity on this iq to finish */
    int		iq_action_ctr; /* if a thread waits for sync, release ity anyway after so many increments of this &*/
  };


#define IN_IOQ(iq) \
  mutex_enter (iq->iq_mtx);

#define LEAVE_IOQ(iq) \
  mutex_leave (iq->iq_mtx);


struct remap_s
  {
    dp_addr_t rm_logical;
    dp_addr_t rm_physical;
  };


#define RA_MAX_BATCH 4000
#define RA_FREE_TEXT_BATCH 20

typedef struct ra_req_s
  {
    int		 ra_inx;
    int			ra_fill;
    int			ra_bfill;
    dp_addr_t		ra_dp[RA_MAX_BATCH];
    buffer_desc_t *	ra_bufs[RA_MAX_BATCH];
    int			ra_nsiblings;
  } ra_req_t;


extern int64 bdf_is_avail_mask; /* all bits on except read aside flag which does not affect reusability */

#ifdef MTX_DEBUG
#define BUF_NEEDS_DELTA(b) 1
#else
#define BUF_NEEDS_DELTA(b) (!(b)->bd_is_dirty)
#endif

#define BUF_WIRED(buf) \
  (buf->bd_readers || buf->bd_is_write \
   || buf->bd_being_read \
   || buf->bd_read_waiting || buf->bd_write_waiting)

#if defined (WIN32)
#define BUF_AVAIL(buf) \
  (buf->bd_readers == 0 \
   && buf->bd_is_write == 0 \
   && buf->bd_being_read == 0 \
   && buf->bd_is_dirty == 0  \
   && !buf->bd_read_waiting && !buf->bd_write_waiting)
#else
#define BUF_AVAIL(buf) \
  ((buf->bdf.flags & bdf_is_avail_mask) == 0		\
   && !buf->bd_read_waiting && !buf->bd_write_waiting)
#endif


#define BUF_CANCEL_WRITE(buf) \
  buf_cancel_write (buf)


/* comparison */

/* for cmp <, <= ==, >, >= the test & 15 is true. For all else test & 15 == 0*/
#define DVC_MATCH 1
#define DVC_LESS 2
#define DVC_GREATER 4
#define DVC_DTP_LESS (DVC_LESS | DVC_NOORDER)
#define DVC_DTP_GREATER	(DVC_GREATER | DVC_NOORDER)
#define DVC_NOORDER 8
#define DVC_INDEX_END 16
#define DVC_CMP_MASK 15 /* or of bits for eq, lt, gt */
#define DVC_UNKNOWN	64  /* comparison of SQL NULL */
#define DVC_QUEUED 128  /* in cluster, not known yet, added to batch */
#define DVC_INVERT_CMP(res) do { \
  switch (res & (DVC_LESS | DVC_GREATER)) \
    { \
    case DVC_LESS: res = (res & ~DVC_LESS) | DVC_GREATER; break; \
    case DVC_GREATER: res = (res & ~DVC_GREATER) | DVC_LESS; break; \
    } } while (0)

/* Comparison operator codes for search_spec_t, sp_min_op, sp_max_op */
#define CMP_NONE 0
#define CMP_EQ 1
#define CMP_LT 2
#define CMP_LTE 3
#define CMP_GT 4
#define CMP_GTE 5
#define CMP_NEQ 14
#define CMP_LIKE 16
#define CMP_NULL 32
#define CMP_NON_NULL 48


#define NUM_COMPARE(n1,n2) \
  (n1 < n2 ? DVC_LESS : (n1 == n2 ? DVC_MATCH : DVC_GREATER))

#define IS_NUM_DTP(dtp) \
  (DV_LONG_INT == dtp || \
   DV_SHORT_INT == dtp || \
   DV_SINGLE_FLOAT == dtp || \
   DV_DOUBLE_FLOAT == dtp || \
   DV_NUMERIC == dtp \
  || DV_INT64 == dtp)

#ifndef dbg_printf
# ifdef DEBUG
#  define dbg_printf(a) { printf a; fflush (stdout); }
# else
#  define dbg_printf(a)
# endif
#endif


/* Catchers */

#define QR_RESET_CTX_T(thr) \
{ \
  du_thread_t * __self = thr; \
  int reset_code;  \
  jmp_buf_splice * __old_ctx = __self->thr_reset_ctx;\
  jmp_buf_splice __ctx;  \
  __self->thr_reset_ctx = &__ctx; \
  if (0 == (reset_code = setjmp_splice (&__ctx)))

#define QR_RESET_CTX  QR_RESET_CTX_T (THREAD_CURRENT_THREAD)

#define QR_RESET_CODE \
  else


#define END_QR_RESET \
    POP_QR_RESET; \
}

#define POP_QR_RESET \
  __self->thr_reset_ctx = __old_ctx




/* Reset  codes */

#define RST_ERROR	1
#define RST_ENOUGH	2 /* for a cursor, reached the end of the current batch of next rows */
#define RST_KILLED	3
#define RST_DEADLOCK	4
#define RST_TIMEOUT	5
#define RST_AT_END 6 /*  reached top or max rows in a select */


#ifdef DEBUG
# define FAILCK(it) if (! it -> itc_fail_context) GPF_T1("No fail context.");
#else
# define FAILCK(it)
#endif

# define ALLOC_CK(xx) dk_alloc_assert ((xx));
#ifdef MALLOC_DEBUG
# define ITC_ALLOC_CK(xx) if ((xx) -> itc_is_allocated) dk_alloc_assert ((((caddr_t)xx) - 8));
#else
# define ITC_ALLOC_CK(xx) ;
#endif


#define CHECK_SESSION_DEAD(lt, itc, buf)		\
{ \
  if (lt) \
    { \
      client_connection_t * cli = lt->lt_client; \
      char __term = cli ? cli->cli_terminate_requested : 0;			\
      if (__term && !wi_inst.wi_checkpoint_atomic) cli_terminate_in_itc_fail (cli, itc, buf); \
      if (cli && cli->cli_session && cli->cli_session->dks_to_close)	\
   { \
       LT_ERROR_DETAIL_SET (lt, \
	   box_dv_short_string ("Client session disconnected")); \
       (lt)->lt_error = LTE_SQL_ERROR; \
       (lt)->lt_status = LT_BLOWN_OFF; \
     } \
    } \
}


/* When inside an itc reset context, periodically call this  to check that no external async condition forces the search to abort or pause */
#define CHECK_TRX_DEAD(it, buf, may_ret) \
{ \
  lock_trx_t *__lt = it->itc_ltrx; \
      CHECK_DK_MEM_RESERVE (__lt); \
      CHECK_SESSION_DEAD (__lt, it, buf);		   \
      if ((__lt && __lt->lt_status != LT_PENDING)  \
|| (wi_inst.wi_is_checkpoint_pending && cpt_is_global_lock ())) \
	{ \
	  if (__lt && !wi_inst.wi_checkpoint_atomic) \
	itc_bust_this_trx (it, buf, may_ret); \
}\
 }								\



/* reset catch context around itc operations */
#define ITC_FAIL(it) \
{   \
  jmp_buf_splice failctx; \
  it->itc_thread = NULL; \
  it->itc_fail_context = &failctx; \
  if (0 == setjmp_splice (&failctx)) \
    {

#define ITC_FAILED \
  } else {

#define END_FAIL_THR(itc, thr) \
    longjmp_splice (thr->thr_reset_ctx, RST_DEADLOCK); \
  } \
  ITC_ALLOC_CK (itc); \
  (itc)->itc_fail_context = NULL; \
}

#define END_FAIL(itc) \
  END_FAIL_THR (itc, THREAD_CURRENT_THREAD)

#define ITC_ABORT_FAIL_CTX(itc) \
  itc->itc_fail_context = NULL

#define ITC_SAVE_FAIL(itc) \
{						    \
  jmp_buf_splice * _s = itc->itc_fail_context;   \

#define ITC_RESTORE_FAIL(itc) \
  itc->itc_fail_context = _s;	 \
  }


#define ITC_CHECK_FAIL(it) \
if (it->itc_ltrx && \
    it->itc_ltrx->lt_blown_off) \
  { \
    if (it->itc_fail_context) \
      longjmp_splice (it->itc_fail_context, RST_DEADLOCK); \
    else \
      GPF_T; \
  } else {};

/* When going to wait for a lock */
#define ITC_SEM_WAIT(it) \
{ \
  mtx_assert (it->itc_thread == THREAD_CURRENT_THREAD);	\
  ITC_LEAVE_MAPS (it); \
  itc_flush_client (it); \
  semaphore_enter (it->itc_thread->thr_sem); \
}


extern jmp_buf_splice structure_fault_ctx;
extern int assertion_on_read_fail;

#define STRUCTURE_FAULT \
  STRUCTURE_FAULT1("Structure fault. Crash recovery recommended.")


#define STRUCTURE_FAULT1(msg)			\
{ \
  if (assertion_on_read_fail) \
    GPF_T1 (msg); \
  else \
    longjmp_splice (&structure_fault_ctx, 1); \
}

extern char *run_as_os_uname;
extern long dbe_auto_sql_stats; /* from search.c */

extern int in_crash_dump;

#if defined (WITH_PTHREADS) && !defined (MTX_DEBUG) && !defined (MTX_METER)  && !defined (IN_ODBC_CLIENT)
#undef mutex_enter
#undef mutex_leave
#define mutex_enter(m)  pthread_mutex_lock (&((m)->mtx_mtx))
#define mutex_leave(m)  pthread_mutex_unlock (&((m)->mtx_mtx))
#endif

#endif /* _WI_H */