mdbx: Merge branch 'devel'.

lmdb.h

@@ -1039,8 +1039,16 @@ size_t mdb_txn_id(MDB_txn *txn);
 	 *
 	 * The transaction handle is freed. It and its cursors must not be used
 	 * again after this call, except with #mdb_cursor_renew().
-	 * @note Earlier documentation incorrectly said all cursors would be freed.
+	 *
+	 * @note MDBX-mode:
+	 * A cursor must be closed explicitly always, before
+	 * or after its transaction ends. It can be reused with
+	 * #mdb_cursor_renew() before finally closing it.
+	 *
+	 * @note LMDB-compatible mode:
+	 * Earlier documentation incorrectly said all cursors would be freed.
 	 * Only write-transactions free cursors.
+	 *
 	 * @param[in] txn A transaction handle returned by #mdb_txn_begin()
 	 * @return A non-zero error value on failure and 0 on success. Some possible
 	 * errors are:
@@ -1057,8 +1065,16 @@ int  mdb_txn_commit(MDB_txn *txn);
 	 *
 	 * The transaction handle is freed. It and its cursors must not be used
 	 * again after this call, except with #mdb_cursor_renew().
-	 * @note Earlier documentation incorrectly said all cursors would be freed.
+	 *
+	 * @note MDBX-mode:
+	 * A cursor must be closed explicitly always, before
+	 * or after its transaction ends. It can be reused with
+	 * #mdb_cursor_renew() before finally closing it.
+	 *
+	 * @note LMDB-compatible mode:
+	 * Earlier documentation incorrectly said all cursors would be freed.
 	 * Only write-transactions free cursors.
+	 *
 	 * @param[in] txn A transaction handle returned by #mdb_txn_begin()
 	 */
 int mdb_txn_abort(MDB_txn *txn);

mdb.c

@@ -2700,8 +2700,9 @@ mdb_cursors_eot(MDB_txn *txn, unsigned merge)
 
 	for (i = txn->mt_numdbs; --i >= 0; ) {
 		for (mc = cursors[i]; mc; mc = next) {
-			mdb_ensure(NULL, mc->mc_signature == MDBX_MC_SIGNATURE
+			unsigned stage = mc->mc_signature;
-				|| mc->mc_signature == MDBX_MC_WAIT4EOT);
+			mdb_ensure(NULL, stage == MDBX_MC_SIGNATURE
+				|| stage == MDBX_MC_WAIT4EOT);
 			next = mc->mc_next;
 			if ((bk = mc->mc_backup) != NULL) {
 				if (merge) {
@@ -2714,10 +2715,8 @@ mdb_cursors_eot(MDB_txn *txn, unsigned merge)
 					if ((mx = mc->mc_xcursor) != NULL)
 						mx->mx_cursor.mc_txn = bk->mc_txn;
 				} else {
-					/* Abort nested txn, but save current cursor's stage */
+					/* Abort nested txn */
-					unsigned stage = mc->mc_signature;
 					*mc = *bk;
-					mc->mc_signature = stage;
 					if ((mx = mc->mc_xcursor) != NULL)
 						*mx = *(MDB_xcursor *)(bk+1);
 				}
@@ -2725,11 +2724,12 @@ mdb_cursors_eot(MDB_txn *txn, unsigned merge)
 				bk->mc_signature = 0;
 				free(bk);
 			}
-			if (mc->mc_signature == MDBX_MC_WAIT4EOT) {
+			if (stage == MDBX_MC_WAIT4EOT) {
 				mc->mc_signature = 0;
 				free(mc);
 			} else {
 				mc->mc_signature = MDBX_MC_READY4CLOSE;
+				mc->mc_flags = 0 /* reset C_UNTRACK */;
 			}
 #else
 				mc = bk;
@@ -3244,7 +3244,12 @@ mdb_txn_reset(MDB_txn *txn)
 	if (unlikely(!(txn->mt_flags & MDB_TXN_RDONLY)))
 		return EINVAL;
 
+#if MDBX_MODE_ENABLED
+	/* LY: don't close DBI-handles in MDBX mode */
+	return mdb_txn_end(txn, MDB_END_RESET|MDB_END_UPDATE);
+#else
 	return mdb_txn_end(txn, MDB_END_RESET);
+#endif /* MDBX_MODE_ENABLED */
 }
 
 int
@@ -3256,6 +3261,12 @@ mdb_txn_abort(MDB_txn *txn)
 	if(unlikely(txn->mt_signature != MDBX_MT_SIGNATURE))
 		return MDB_VERSION_MISMATCH;
 
+#if MDBX_MODE_ENABLED
+	if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY))
+		/* LY: don't close DBI-handles in MDBX mode */
+		return mdb_txn_end(txn, MDB_END_ABORT|MDB_END_UPDATE|MDB_END_SLOT|MDB_END_FREE);
+#endif /* MDBX_MODE_ENABLED */
+
 	if (txn->mt_child)
 		mdb_txn_abort(txn->mt_child);
 
@@ -5763,7 +5774,7 @@ mdb_page_search(MDB_cursor *mc, MDB_val *key, int flags)
 		return MDB_BAD_TXN;
 	} else {
 		/* Make sure we're using an up-to-date root */
-		if (*mc->mc_dbflag & DB_STALE) {
+		if (unlikely(*mc->mc_dbflag & DB_STALE)) {
 				MDB_cursor mc2;
 				if (unlikely(TXN_DBI_CHANGED(mc->mc_txn, mc->mc_dbi)))
 					return MDB_BAD_DBI;
@@ -7826,15 +7837,14 @@ mdb_cursor_init(MDB_cursor *mc, MDB_txn *txn, MDB_dbi dbi, MDB_xcursor *mx)
 	mc->mc_pg[0] = 0;
 	mc->mc_flags = 0;
 	mc->mc_ki[0] = 0;
+	mc->mc_xcursor = NULL;
 	if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
 		mdb_tassert(txn, mx != NULL);
 		mx->mx_cursor.mc_signature = MDBX_MC_SIGNATURE;
 		mc->mc_xcursor = mx;
 		mdb_xcursor_init0(mc);
-	} else {
-		mc->mc_xcursor = NULL;
 	}
-	if (*mc->mc_dbflag & DB_STALE) {
+	if (unlikely(*mc->mc_dbflag & DB_STALE)) {
 		mdb_page_search(mc, NULL, MDB_PS_ROOTONLY);
 	}
 }
@@ -10226,7 +10236,7 @@ mdbx_stat(MDB_txn *txn, MDB_dbi dbi, MDBX_stat *arg, size_t bytes)
 	if (unlikely(txn->mt_flags & MDB_TXN_BLOCKED))
 		return MDB_BAD_TXN;
 
-	if (txn->mt_dbflags[dbi] & DB_STALE) {
+	if (unlikely(txn->mt_dbflags[dbi] & DB_STALE)) {
 		MDB_cursor mc;
 		MDB_xcursor mx;
 		/* Stale, must read the DB's root. cursor_init does it for us. */

mdbx.c

@@ -172,7 +172,7 @@ typedef struct mdb_walk_ctx {
 
 /** Depth-first tree traversal. */
 static int __cold
-mdb_env_walk(mdb_walk_ctx_t *ctx, const char* dbi, pgno_t pg, int flags, int deep)
+mdb_env_walk(mdb_walk_ctx_t *ctx, const char* dbi, pgno_t pg, int deep)
 {
 	MDB_page *mp;
 	int rc, i, nkeys;
@@ -238,7 +238,7 @@ mdb_env_walk(mdb_walk_ctx_t *ctx, const char* dbi, pgno_t pg, int flags, int dee
 		payload_size += NODESIZE + node->mn_ksize;
 
 		if (IS_BRANCH(mp)) {
-			rc = mdb_env_walk(ctx, dbi, NODEPGNO(node), flags, deep);
+			rc = mdb_env_walk(ctx, dbi, NODEPGNO(node), deep);
 			if (rc)
 				return rc;
 			continue;
@@ -286,7 +286,7 @@ mdb_env_walk(mdb_walk_ctx_t *ctx, const char* dbi, pgno_t pg, int flags, int dee
 				name[namelen] = 0;
 			}
 			rc = mdb_env_walk(ctx, (name && name[0]) ? name : dbi,
-					db->md_root, node->mn_flags & F_DUPDATA, deep + 1);
+					db->md_root, deep + 1);
 			if (rc)
 				return rc;
 		}
@@ -314,9 +314,9 @@ mdbx_env_pgwalk(MDB_txn *txn, MDBX_pgvisitor_func* visitor, void* user)
 	rc = visitor(0, 2, user, "lmdb", "meta", 2, sizeof(MDB_meta)*2, PAGEHDRSZ*2,
 				 (txn->mt_env->me_psize - sizeof(MDB_meta) - PAGEHDRSZ) *2);
 	if (! rc)
-		rc = mdb_env_walk(&ctx, "free", txn->mt_dbs[FREE_DBI].md_root, 0, 0);
+		rc = mdb_env_walk(&ctx, "free", txn->mt_dbs[FREE_DBI].md_root, 0);
 	if (! rc)
-		rc = mdb_env_walk(&ctx, "main", txn->mt_dbs[MAIN_DBI].md_root, 0, 0);
+		rc = mdb_env_walk(&ctx, "main", txn->mt_dbs[MAIN_DBI].md_root, 0);
 	if (! rc)
 		rc = visitor(P_INVALID, 0, user, NULL, NULL, 0, 0, 0, 0);
 	return rc;
@@ -354,6 +354,47 @@ size_t mdbx_canary_get(MDB_txn *txn, mdbx_canary* canary)
 	return txn->mt_txnid;
 }
 
+int mdbx_cursor_on_first(MDB_cursor *mc)
+{
+	if (unlikely(mc == NULL))
+		return EINVAL;
+
+	if (unlikely(mc->mc_signature != MDBX_MC_SIGNATURE))
+		return MDB_VERSION_MISMATCH;
+
+	if (!(mc->mc_flags & C_INITIALIZED))
+		return MDBX_RESULT_FALSE;
+
+	unsigned i;
+	for(i = 0; i < mc->mc_snum; ++i) {
+		if (mc->mc_ki[i])
+			return MDBX_RESULT_FALSE;
+	}
+
+	return MDBX_RESULT_TRUE;
+}
+
+int mdbx_cursor_on_last(MDB_cursor *mc)
+{
+	if (unlikely(mc == NULL))
+		return EINVAL;
+
+	if (unlikely(mc->mc_signature != MDBX_MC_SIGNATURE))
+		return MDB_VERSION_MISMATCH;
+
+	if (!(mc->mc_flags & C_INITIALIZED))
+		return MDBX_RESULT_FALSE;
+
+	unsigned i;
+	for(i = 0; i < mc->mc_snum; ++i) {
+		unsigned nkeys = NUMKEYS(mc->mc_pg[i]);
+		if (mc->mc_ki[i] != nkeys - 1)
+			return MDBX_RESULT_FALSE;
+	}
+
+	return MDBX_RESULT_TRUE;
+}
+
 int mdbx_cursor_eof(MDB_cursor *mc)
 {
 	if (unlikely(mc == NULL))

mdbx.h

@@ -219,10 +219,19 @@ typedef struct mdbx_canary {
 int mdbx_canary_put(MDB_txn *txn, const mdbx_canary* canary);
 size_t mdbx_canary_get(MDB_txn *txn, mdbx_canary* canary);
 
-/* Returns 1 when no more data available or cursor not positioned,
+/* Returns:
- * 0 otherwise or less that zero in error case. */
+ *	- MDBX_RESULT_TRUE	when no more data available
+ *				or cursor not positioned;
+ *	- MDBX_RESULT_FALSE	when data available;
+ *	- Otherwise the error code. */
 int mdbx_cursor_eof(MDB_cursor *mc);
 
+/* Returns: MDBX_RESULT_TRUE, MDBX_RESULT_FALSE or Error code. */
+int mdbx_cursor_on_first(MDB_cursor *mc);
+
+/* Returns: MDBX_RESULT_TRUE, MDBX_RESULT_FALSE or Error code. */
+int mdbx_cursor_on_last(MDB_cursor *mc);
+
 #define MDBX_EMULTIVAL (MDB_LAST_ERRCODE - 42)
 #define MDBX_RESULT_FALSE MDB_SUCCESS
 #define MDBX_RESULT_TRUE (-1)