[Tarantool-patches] [PATCH v5 51/52] sql: introduce mem_get_bytes_len()

imeevma at tarantool.org imeevma at tarantool.org
Sat Apr 10 00:08:28 MSK 2021


Thank you for the review! My answer and new patch below. Also, I see that I
did a lot of misspelling and other errors in my answers and I am very sorry
about this. I will try to add more info to my answers tomorrow.


On 30.03.2021 02:09, Vladislav Shpilevoy wrote:
> Thanks for the patch!
>
>> diff --git a/src/box/sql/func.c b/src/box/sql/func.c
>> index 199f3abef..6687fca3c 100644
>> --- a/src/box/sql/func.c
>> +++ b/src/box/sql/func.c
>> @@ -967,7 +968,7 @@ case_type##ICUFunc(sql_context *context, int argc, sql_value **argv)   \
>>  		return;                                                        \
>>  	}                                                                      \
>>  	z2 = mem_get_str(argv[0]);                                            \
>> -	n = sql_value_bytes(argv[0]);                                      \
>> +	n = mem_get_length(argv[0]);                                      \
>
> All these \ were aligned before your patchset. Please, keep it straight.
>
Fixed.

>>  	/*                                                                     \
>>  	 * Verify that the call to _bytes()                                    \
>>  	 * does not invalidate the _text() pointer.                            \


New patch:

commit 326fac6c97ae260b32be9987e48e3381c600ba9c
Author: Mergen Imeev <imeevma at gmail.com>
Date:   Mon Mar 22 12:27:33 2021 +0300

    sql: introduce mem_get_bytes_len()
    
    This patch introduces mem_get_bytes_len(). This function is used to
    receive length of string or binary value of MEM. If MEM is not
    of STRING or VARBINARy type this function returns -1.
    
    Part of #5818

diff --git a/src/box/sql/func.c b/src/box/sql/func.c
index 2896a5c31..746bda0f4 100644
--- a/src/box/sql/func.c
+++ b/src/box/sql/func.c
@@ -65,6 +65,14 @@ mem_as_bin(struct Mem *mem)
 	return s;
 }
 
+static int
+mem_get_length(struct Mem *mem)
+{
+	uint32_t len;
+	mem_get_bytes_len(mem, &len);
+	return len;
+}
+
 /*
  * Return the collating function associated with a function.
  */
@@ -187,14 +195,15 @@ lengthFunc(sql_context * context, int argc, sql_value ** argv)
 	case MP_UINT:
 	case MP_BOOL:
 	case MP_DOUBLE:{
-			sql_result_uint(context, sql_value_bytes(argv[0]));
+			mem_as_bin(argv[0]);
+			sql_result_uint(context, mem_get_length(argv[0]));
 			break;
 		}
 	case MP_STR:{
 			const unsigned char *z = mem_as_ustr(argv[0]);
 			if (z == 0)
 				return;
-			len = sql_utf8_char_count(z, sql_value_bytes(argv[0]));
+			len = sql_utf8_char_count(z, mem_get_length(argv[0]));
 			sql_result_uint(context, len);
 			break;
 		}
@@ -310,8 +319,8 @@ position_func(struct sql_context *context, int argc, struct Mem **argv)
 		return;
 	}
 
-	int n_needle_bytes = sql_value_bytes(needle);
-	int n_haystack_bytes = sql_value_bytes(haystack);
+	int n_needle_bytes = mem_get_length(needle);
+	int n_haystack_bytes = mem_get_length(haystack);
 	int position = 1;
 	if (n_needle_bytes > 0) {
 		const unsigned char *haystack_str;
@@ -457,18 +466,18 @@ substrFunc(sql_context * context, int argc, sql_value ** argv)
 	bool unused;
 	mem_get_int(argv[1], &p1, &unused);
 	if (p0type == MP_BIN) {
-		len = sql_value_bytes(argv[0]);
 		z = mem_as_bin(argv[0]);
+		len = mem_get_length(argv[0]);
 		if (z == 0)
 			return;
-		assert(len == sql_value_bytes(argv[0]));
+		assert(len == mem_get_length(argv[0]));
 	} else {
 		z = mem_as_ustr(argv[0]);
 		if (z == 0)
 			return;
 		len = 0;
 		if (p1 < 0)
-			len = sql_utf8_char_count(z, sql_value_bytes(argv[0]));
+			len = sql_utf8_char_count(z, mem_get_length(argv[0]));
 	}
 	if (argc == 3) {
 		mem_get_int(argv[2], &p2, &unused);
@@ -507,7 +516,7 @@ substrFunc(sql_context * context, int argc, sql_value ** argv)
 		 * used because '\0' is not supposed to be
 		 * end-of-string symbol.
 		 */
-		int byte_size = sql_value_bytes(argv[0]);
+		int byte_size = mem_get_length(argv[0]);
 		int n_chars = sql_utf8_char_count(z, byte_size);
 		int cnt = 0;
 		int i = 0;
@@ -627,7 +636,7 @@ case_type##ICUFunc(sql_context *context, int argc, sql_value **argv)   \
 		return;                                                        \
 	}                                                                      \
 	z2 = mem_as_str0(argv[0]);                                             \
-	n = sql_value_bytes(argv[0]);                                      \
+	n = mem_get_length(argv[0]);                                           \
 	/*                                                                     \
 	 * Verify that the call to _bytes()                                    \
 	 * does not invalidate the _text() pointer.                            \
@@ -965,15 +974,15 @@ likeFunc(sql_context *context, int argc, sql_value **argv)
 	}
 	const char *zB = mem_as_str0(argv[0]);
 	const char *zA = mem_as_str0(argv[1]);
-	const char *zB_end = zB + sql_value_bytes(argv[0]);
-	const char *zA_end = zA + sql_value_bytes(argv[1]);
+	const char *zB_end = zB + mem_get_length(argv[0]);
+	const char *zA_end = zA + mem_get_length(argv[1]);
 
 	/*
 	 * Limit the length of the LIKE pattern to avoid problems
 	 * of deep recursion and N*N behavior in
 	 * sql_utf8_pattern_compare().
 	 */
-	nPat = sql_value_bytes(argv[0]);
+	nPat = mem_get_length(argv[0]);
 	testcase(nPat == db->aLimit[SQL_LIMIT_LIKE_PATTERN_LENGTH]);
 	testcase(nPat == db->aLimit[SQL_LIMIT_LIKE_PATTERN_LENGTH] + 1);
 	if (nPat > db->aLimit[SQL_LIMIT_LIKE_PATTERN_LENGTH]) {
@@ -994,7 +1003,7 @@ likeFunc(sql_context *context, int argc, sql_value **argv)
 		const unsigned char *zEsc = mem_as_ustr(argv[2]);
 		if (zEsc == 0)
 			return;
-		if (sql_utf8_char_count(zEsc, sql_value_bytes(argv[2])) != 1) {
+		if (sql_utf8_char_count(zEsc, mem_get_length(argv[2])) != 1) {
 			diag_set(ClientError, ER_SQL_EXECUTE, "ESCAPE "\
 				 "expression must be a single character");
 			context->is_aborted = true;
@@ -1094,7 +1103,7 @@ quoteFunc(sql_context * context, int argc, sql_value ** argv)
 	case MP_MAP: {
 			char *zText = 0;
 			char const *zBlob = mem_as_bin(argv[0]);
-			int nBlob = sql_value_bytes(argv[0]);
+			int nBlob = mem_get_length(argv[0]);
 			assert(zBlob == mem_as_bin(argv[0]));	/* No encoding change */
 			zText =
 			    (char *)contextMalloc(context,
@@ -1230,7 +1239,7 @@ hexFunc(sql_context * context, int argc, sql_value ** argv)
 	assert(argc == 1);
 	UNUSED_PARAMETER(argc);
 	pBlob = mem_as_bin(argv[0]);
-	n = sql_value_bytes(argv[0]);
+	n = mem_get_length(argv[0]);
 	assert(pBlob == mem_as_bin(argv[0]));	/* No encoding change */
 	z = zHex = contextMalloc(context, ((i64) n) * 2 + 1);
 	if (zHex) {
@@ -1289,7 +1298,7 @@ replaceFunc(sql_context * context, int argc, sql_value ** argv)
 	zStr = mem_as_ustr(argv[0]);
 	if (zStr == 0)
 		return;
-	nStr = sql_value_bytes(argv[0]);
+	nStr = mem_get_length(argv[0]);
 	assert(zStr == mem_as_ustr(argv[0]));	/* No encoding change */
 	zPattern = mem_as_ustr(argv[1]);
 	if (zPattern == 0) {
@@ -1297,7 +1306,7 @@ replaceFunc(sql_context * context, int argc, sql_value ** argv)
 		       || sql_context_db_handle(context)->mallocFailed);
 		return;
 	}
-	nPattern = sql_value_bytes(argv[1]);
+	nPattern = mem_get_length(argv[1]);
 	if (nPattern == 0) {
 		assert(!mem_is_null(argv[1]));
 		sql_result_value(context, argv[0]);
@@ -1307,7 +1316,7 @@ replaceFunc(sql_context * context, int argc, sql_value ** argv)
 	zRep = mem_as_ustr(argv[2]);
 	if (zRep == 0)
 		return;
-	nRep = sql_value_bytes(argv[2]);
+	nRep = mem_get_length(argv[2]);
 	assert(zRep == mem_as_ustr(argv[2]));
 	nOut = nStr + 1;
 	assert(nOut < SQL_MAX_LENGTH);
@@ -1469,7 +1478,7 @@ trim_func_one_arg(struct sql_context *context, sql_value *arg)
 		default_trim = (const unsigned char *) "\0";
 	else
 		default_trim = (const unsigned char *) " ";
-	int input_str_sz = sql_value_bytes(arg);
+	int input_str_sz = mem_get_length(arg);
 	const unsigned char *input_str = mem_as_ustr(arg);
 	uint8_t trim_char_len[1] = { 1 };
 	trim_procedure(context, TRIM_BOTH, default_trim, trim_char_len, 1,
@@ -1495,7 +1504,7 @@ trim_func_two_args(struct sql_context *context, sql_value *arg1,
 	if ((input_str = mem_as_ustr(arg2)) == NULL)
 		return;
 
-	int input_str_sz = sql_value_bytes(arg2);
+	int input_str_sz = mem_get_length(arg2);
 	if (sql_value_type(arg1) == MP_INT || sql_value_type(arg1) == MP_UINT) {
 		uint8_t len_one = 1;
 		uint64_t n;
@@ -1503,7 +1512,7 @@ trim_func_two_args(struct sql_context *context, sql_value *arg1,
 		trim_procedure(context, n, (const unsigned char *) " ",
 			       &len_one, 1, input_str, input_str_sz);
 	} else if ((trim_set = mem_as_ustr(arg1)) != NULL) {
-		int trim_set_sz = sql_value_bytes(arg1);
+		int trim_set_sz = mem_get_length(arg1);
 		uint8_t *char_len;
 		int char_cnt = trim_prepare_char_len(context, trim_set,
 						     trim_set_sz, &char_len);
@@ -1532,8 +1541,8 @@ trim_func_three_args(struct sql_context *context, sql_value *arg1,
 	    (trim_set = mem_as_ustr(arg2)) == NULL)
 		return;
 
-	int trim_set_sz = sql_value_bytes(arg2);
-	int input_str_sz = sql_value_bytes(arg3);
+	int trim_set_sz = mem_get_length(arg2);
+	int input_str_sz = mem_get_length(arg3);
 	uint8_t *char_len;
 	int char_cnt = trim_prepare_char_len(context, trim_set, trim_set_sz,
 					     &char_len);
@@ -1853,7 +1862,7 @@ groupConcatStep(sql_context * context, int argc, sql_value ** argv)
 		if (!firstTerm) {
 			if (argc == 2) {
 				zSep = mem_as_str0(argv[1]);
-				nSep = sql_value_bytes(argv[1]);
+				nSep = mem_get_length(argv[1]);
 			} else {
 				zSep = ",";
 				nSep = 1;
@@ -1862,7 +1871,7 @@ groupConcatStep(sql_context * context, int argc, sql_value ** argv)
 				sqlStrAccumAppend(pAccum, zSep, nSep);
 		}
 		zVal = mem_as_str0(argv[0]);
-		nVal = sql_value_bytes(argv[0]);
+		nVal = mem_get_length(argv[0]);
 		if (zVal)
 			sqlStrAccumAppend(pAccum, zVal, nVal);
 	}
diff --git a/src/box/sql/mem.c b/src/box/sql/mem.c
index 2e69f4a80..3e9544cae 100644
--- a/src/box/sql/mem.c
+++ b/src/box/sql/mem.c
@@ -1345,6 +1345,18 @@ mem_get_bin(const struct Mem *mem, const char **s)
 	return 0;
 }
 
+int
+mem_get_bytes_len(const struct Mem *mem, uint32_t *len)
+{
+	if ((mem->flags & (MEM_Str | MEM_Blob)) == 0)
+		return -1;
+	if ((mem->flags & MEM_Blob) !=0 && (mem->flags & MEM_Zero) != 0)
+		*len = mem->n + mem->u.nZero;
+	else
+		*len = mem->n;
+	return 0;
+}
+
 int
 mem_copy(struct Mem *to, const struct Mem *from)
 {
@@ -2116,41 +2128,6 @@ sql_value_type(sql_value *pVal)
 	return mem_mp_type(mem);
 }
 
-/*
- * The sqlValueBytes() routine returns the number of bytes in the
- * sql_value object assuming that it uses the encoding "enc".
- * The valueBytes() routine is a helper function.
- */
-static SQL_NOINLINE int
-valueBytes(sql_value * pVal)
-{
-	if (mem_to_str(pVal) != 0)
-		return 0;
-	return pVal->n;
-}
-
-int
-sqlValueBytes(sql_value * pVal)
-{
-	Mem *p = (Mem *) pVal;
-	assert((p->flags & MEM_Null) == 0
-	       || (p->flags & (MEM_Str | MEM_Blob)) == 0);
-	if ((p->flags & MEM_Str) != 0) {
-		return p->n;
-	}
-	if ((p->flags & MEM_Blob) != 0) {
-		if (p->flags & MEM_Zero) {
-			return p->n + p->u.nZero;
-		} else {
-			return p->n;
-		}
-	}
-	if (p->flags & MEM_Null)
-		return 0;
-	return valueBytes(pVal);
-}
-
-
 #ifdef SQL_DEBUG
 /*
  * Check invariants on a Mem object.
@@ -2457,53 +2434,6 @@ releaseMemArray(Mem * p, int N)
 	}
 }
 
-int
-sql_value_bytes(sql_value * pVal)
-{
-	return sqlValueBytes(pVal);
-}
-
-/*
- * Return a pointer to static memory containing an SQL NULL value.
- */
-const Mem *
-columnNullValue(void)
-{
-	/* Even though the Mem structure contains an element
-	 * of type i64, on certain architectures (x86) with certain compiler
-	 * switches (-Os), gcc may align this Mem object on a 4-byte boundary
-	 * instead of an 8-byte one. This all works fine, except that when
-	 * running with SQL_DEBUG defined the sql code sometimes assert()s
-	 * that a Mem structure is located on an 8-byte boundary. To prevent
-	 * these assert()s from failing, when building with SQL_DEBUG defined
-	 * using gcc, we force nullMem to be 8-byte aligned using the magical
-	 * __attribute__((aligned(8))) macro.
-	 */
-	static const Mem nullMem
-#if defined(SQL_DEBUG) && defined(__GNUC__)
-	    __attribute__ ((aligned(8)))
-#endif
-	    = {
-		/* .u          = */  {
-		0},
-		    /* .flags      = */ (u16) MEM_Null,
-		    /* .eSubtype   = */ (u8) 0,
-		    /* .field_type = */ field_type_MAX,
-		    /* .n          = */ (int)0,
-		    /* .z          = */ (char *)0,
-		    /* .zMalloc    = */ (char *)0,
-		    /* .szMalloc   = */ (int)0,
-		    /* .uTemp      = */ (u32) 0,
-		    /* .db         = */ (sql *) 0,
-		    /* .xDel       = */ (void (*)(void *))0,
-#ifdef SQL_DEBUG
-		    /* .pScopyFrom = */ (Mem *) 0,
-		    /* .pFiller    = */ (void *)0,
-#endif
-	};
-	return &nullMem;
-}
-
 /*
  * Return true if the Mem object contains a TEXT or BLOB that is
  * too large - whose size exceeds SQL_MAX_LENGTH.
diff --git a/src/box/sql/mem.h b/src/box/sql/mem.h
index e2d0f343e..9c6d18e1f 100644
--- a/src/box/sql/mem.h
+++ b/src/box/sql/mem.h
@@ -598,6 +598,13 @@ mem_as_str0(struct Mem *mem)
 int
 mem_get_bin(const struct Mem *mem, const char **s);
 
+/**
+ * Return length of value for MEM of STRING or VARBINARY type. Original MEM is
+ * not changed.
+ */
+int
+mem_get_bytes_len(const struct Mem *mem, uint32_t *len);
+
 /**
  * Simple type to str convertor. It is used to simplify
  * error reporting.
@@ -616,8 +623,6 @@ mem_mp_type(struct Mem *mem);
 enum mp_type
 sql_value_type(struct Mem *);
 
-int sqlValueBytes(struct Mem *);
-
 #ifdef SQL_DEBUG
 int sqlVdbeCheckMemInvariants(struct Mem *);
 void sqlVdbeMemPrettyPrint(Mem * pMem, char *zBuf);
@@ -647,18 +652,8 @@ struct Mem *sqlValueNew(struct sql *);
 void
 releaseMemArray(Mem * p, int N);
 
-/** Getters. */
-
-int
-sql_value_bytes(struct Mem *);
-
 #define VdbeFrameMem(p) ((Mem *)&((u8 *)p)[ROUND8(sizeof(VdbeFrame))])
 
-const Mem *
-columnNullValue(void);
-
-/** Checkers. */
-
 int sqlVdbeMemTooBig(Mem *);
 
 int sqlMemCompare(const Mem *, const Mem *, const struct coll *);
diff --git a/src/box/sql/sqlInt.h b/src/box/sql/sqlInt.h
index 8a7a87b49..8feb112f5 100644
--- a/src/box/sql/sqlInt.h
+++ b/src/box/sql/sqlInt.h
@@ -433,9 +433,6 @@ sql_stmt_compile(const char *sql, int bytes_count, struct Vdbe *re_prepared,
 int
 sql_step(sql_stmt *);
 
-int
-sql_column_bytes(sql_stmt *, int iCol);
-
 int
 sql_column_bytes16(sql_stmt *, int iCol);
 
diff --git a/src/box/sql/vdbeapi.c b/src/box/sql/vdbeapi.c
index 19b70a55b..b0334f3ed 100644
--- a/src/box/sql/vdbeapi.c
+++ b/src/box/sql/vdbeapi.c
@@ -442,41 +442,6 @@ sql_data_count(sql_stmt * pStmt)
 	return pVm->nResColumn;
 }
 
-/*
- * Check to see if column iCol of the given statement is valid.  If
- * it is, return a pointer to the Mem for the value of that column.
- * If iCol is not valid, return a pointer to a Mem which has a value
- * of NULL.
- */
-static Mem *
-columnMem(sql_stmt * pStmt, int i)
-{
-	Vdbe *pVm;
-	Mem *pOut;
-
-	pVm = (Vdbe *) pStmt;
-	if (pVm == 0)
-		return (Mem *) columnNullValue();
-	assert(pVm->db);
-	if (pVm->pResultSet != 0 && i < pVm->nResColumn && i >= 0) {
-		pOut = &pVm->pResultSet[i];
-	} else {
-		pOut = (Mem *) columnNullValue();
-	}
-	return pOut;
-}
-
-/**************************** sql_column_  ******************************
- * The following routines are used to access elements of the current row
- * in the result set.
- */
-
-int
-sql_column_bytes(sql_stmt * pStmt, int i)
-{
-	return sql_value_bytes(columnMem(pStmt, i));
-}
-
 char *
 sql_result_to_msgpack(struct sql_stmt *stmt, uint32_t *tuple_size,
 		      struct region *region)


More information about the Tarantool-patches mailing list