[Tarantool-patches] [PATCH v5 39/52] sql: introduce mem_to_int*() functions

[Mergen Imeev via Tarantool-patches <tarantool-patches@dev.tarantool.org>]

Thank you for the review! My answers and new patch below.


On 30.03.2021 02:07, Vladislav Shpilevoy wrote:
> Thanks for the patch!
>
> See 2 comments below.
>
> On 23.03.2021 10:36, imeevma@tarantool.org wrote:
>> This patch introduces mem_convert_to_integer() which is used to convert
>> a MEM to a MEM that contains integer value.
>>
>> Part of #5818
>> ---
>>  src/box/sql/mem.c  | 114 +++++++++++++++++++++++++++++----------------
>>  src/box/sql/mem.h  |   7 ++-
>>  src/box/sql/vdbe.c |  19 ++++----
>>  3 files changed, 88 insertions(+), 52 deletions(-)
>>
>> diff --git a/src/box/sql/mem.c b/src/box/sql/mem.c
>> index 1209df1ce..b9bcd3d3a 100644
>> --- a/src/box/sql/mem.c
>> +++ b/src/box/sql/mem.c
>> @@ -695,6 +695,80 @@ mem_set_cleared(struct Mem *mem)
>
> <...>
>
>> +
>> +static inline int
>> +mem_convert_double_to_integer_lossless(struct Mem *mem)
>
> 1. Lets use 'precise'. Here and in other places. And shorten
> the names. So it would be mem_convert_bool_to_int(),
> mem_convert_to_int(), etc.
>
Thanks. Fixed. I shortened names of the functions to mem_to_int(),
mem_to_double(), etc. Also, I shortened names of static functions to
double_to_int(), int_to_string(), etc. Replaces "lossless" by "precise".

>> +{
>> +	double d = mem->u.r;
>> +	if (d < 0 && d >= (double)INT64_MIN && (double)(int64_t)d == d) {
>> +		mem_set_integer(mem, (int64_t)d, true);
>> +		return 0;
>> +	}
>> +	if (d >= 0 && d < (double)UINT64_MAX && (double)(uint64_t)d == d) {
>> +		mem_set_integer(mem, (int64_t)(uint64_t)d, false);
>
> 2. Isn't mem_set_unsigned() faster? Or we can't set FIELD_TYPE_UNSIGNED?
Inlined setting of MEM here and in some other places.


New patch:

commit 8b70c648c1b907b04859cfc8bcef94870057bf9f
Author: Mergen Imeev <imeevma@gmail.com>
Date:   Wed Mar 17 10:46:29 2021 +0300

    sql: introduce mem_to_int*() functions
    
    This patch introduces mem_to_int() and mem_to_int_precise() functions.
    These functions are used to convert a MEM to a MEM that contains
    integer value. These functions defines the rules that are used during
    convertion from values of all other types to INTEGER.
    
    Part of #5818

diff --git a/src/box/sql/mem.c b/src/box/sql/mem.c
index 52b1891aa..d3a3215bc 100644
--- a/src/box/sql/mem.c
+++ b/src/box/sql/mem.c
@@ -687,6 +687,90 @@ mem_set_null_clear(struct Mem *mem)
 	mem->flags = MEM_Null | MEM_Cleared;
 }
 
+static inline int
+bytes_to_int(struct Mem *mem)
+{
+	bool is_neg;
+	int64_t i;
+	if (sql_atoi64(mem->z, &i, &is_neg, mem->n) != 0)
+		return -1;
+	mem_set_int(mem, i, is_neg);
+	return 0;
+}
+
+static inline int
+double_to_int(struct Mem *mem)
+{
+	double d = mem->u.r;
+	if (d < 0 && d >= (double)INT64_MIN) {
+		mem->u.i = (int64_t)d;
+		mem->flags = MEM_Int;
+		mem->field_type = FIELD_TYPE_INTEGER;
+		return 0;
+	}
+	if (d >= 0 && d < (double)UINT64_MAX) {
+		mem->u.u = (uint64_t)d;
+		mem->flags = MEM_UInt;
+		mem->field_type = FIELD_TYPE_UNSIGNED;
+		return 0;
+	}
+	return -1;
+}
+
+static inline int
+double_to_int_precise(struct Mem *mem)
+{
+	double d = mem->u.r;
+	if (d < 0 && d >= (double)INT64_MIN && (double)(int64_t)d == d) {
+		mem->u.i = (int64_t)d;
+		mem->flags = MEM_Int;
+		mem->field_type = FIELD_TYPE_INTEGER;
+		return 0;
+	}
+	if (d >= 0 && d < (double)UINT64_MAX && (double)(uint64_t)d == d) {
+		mem->u.u = (uint64_t)d;
+		mem->flags = MEM_UInt;
+		mem->field_type = FIELD_TYPE_UNSIGNED;
+		return 0;
+	}
+	return -1;
+}
+
+static inline int
+bool_to_int(struct Mem *mem)
+{
+	mem->u.u = (uint64_t)mem->u.b;
+	mem->flags = MEM_UInt;
+	mem->field_type = FIELD_TYPE_UNSIGNED;
+	return 0;
+}
+
+int
+mem_to_int(struct Mem *mem)
+{
+	if ((mem->flags & (MEM_Int | MEM_UInt)) != 0)
+		return 0;
+	if ((mem->flags & (MEM_Str | MEM_Blob)) != 0)
+		return bytes_to_int(mem);
+	if ((mem->flags & MEM_Real) != 0)
+		return double_to_int(mem);
+	if ((mem->flags & MEM_Bool) != 0)
+		return bool_to_int(mem);
+	return -1;
+}
+
+int
+mem_to_int_precise(struct Mem *mem)
+{
+	if ((mem->flags & (MEM_Int | MEM_UInt)) != 0)
+		return 0;
+	if ((mem->flags & MEM_Str) != 0)
+		return bytes_to_int(mem);
+	if ((mem->flags & MEM_Real) != 0)
+		return double_to_int_precise(mem);
+	return -1;
+}
+
 int
 mem_copy(struct Mem *to, const struct Mem *from)
 {
@@ -1924,26 +2008,6 @@ sqlVdbeMemCast(Mem * pMem, enum field_type type)
 	}
 }
 
-/*
- * The MEM structure is already a MEM_Real.  Try to also make it a
- * MEM_Int if we can.
- */
-int
-mem_apply_integer_type(Mem *pMem)
-{
-	int rc;
-	i64 ix;
-	assert(pMem->flags & MEM_Real);
-	assert(EIGHT_BYTE_ALIGNMENT(pMem));
-
-	if ((rc = doubleToInt64(pMem->u.r, (int64_t *) &ix)) == 0) {
-		pMem->u.i = ix;
-		pMem->flags = pMem->u.r <= -1 ? MEM_Int : MEM_UInt;
-		pMem->field_type = FIELD_TYPE_INTEGER;
-	}
-	return rc;
-}
-
 /*
  * Add MEM_Str to the set of representations for the given Mem.  Numbers
  * are converted using sql_snprintf().  Converting a BLOB to a string
@@ -2215,34 +2279,6 @@ mem_convert_to_unsigned(struct Mem *mem)
 	return 0;
 }
 
-/**
- * Convert the numeric value contained in MEM to integer.
- *
- * @param mem The MEM that contains the numeric value.
- * @retval 0 if the conversion was successful, -1 otherwise.
- */
-static int
-mem_convert_to_integer(struct Mem *mem)
-{
-	if ((mem->flags & (MEM_UInt | MEM_Int)) != 0)
-		return 0;
-	if ((mem->flags & MEM_Real) == 0)
-		return -1;
-	double d = mem->u.r;
-	if (d >= (double)UINT64_MAX || d < (double)INT64_MIN)
-		return -1;
-	if (d < 0.) {
-		mem->u.i = (int64_t)d;
-		mem->flags = MEM_Int;
-		mem->field_type = FIELD_TYPE_INTEGER;
-	} else {
-		mem->u.u = (uint64_t)d;
-		mem->flags = MEM_UInt;
-		mem->field_type = FIELD_TYPE_UNSIGNED;
-	}
-	return 0;
-}
-
 int
 mem_convert_to_numeric(struct Mem *mem, enum field_type type)
 {
@@ -2253,7 +2289,7 @@ mem_convert_to_numeric(struct Mem *mem, enum field_type type)
 	if (type == FIELD_TYPE_UNSIGNED)
 		return mem_convert_to_unsigned(mem);
 	assert(type == FIELD_TYPE_INTEGER);
-	return mem_convert_to_integer(mem);
+	return mem_to_int(mem);
 }
 
 static int
diff --git a/src/box/sql/mem.h b/src/box/sql/mem.h
index ce5076361..d3eb04c44 100644
--- a/src/box/sql/mem.h
+++ b/src/box/sql/mem.h
@@ -473,6 +473,24 @@ int
 mem_compare(const struct Mem *left, const struct Mem *right, int *result,
 	    enum field_type type, struct coll *coll);
 
+/**
+ * Convert the given MEM to INTEGER. This function and the function below define
+ * the rules that are used to convert values of all other types to INTEGER. In
+ * this function, the conversion from double to integer may result in loss of
+ * precision.
+ */
+int
+mem_to_int(struct Mem *mem);
+
+/**
+ * Convert the given MEM to INTEGER. This function and the function above define
+ * the rules that are used to convert values of all other types to INTEGER. In
+ * this function, the conversion from double to integer is only possible if it
+ * is lossless.
+ */
+int
+mem_to_int_precise(struct Mem *mem);
+
 /**
  * Simple type to str convertor. It is used to simplify
  * error reporting.
@@ -533,7 +551,6 @@ int
 vdbe_mem_numerify(struct Mem *mem);
 
 int sqlVdbeMemCast(struct Mem *, enum field_type type);
-int mem_apply_integer_type(struct Mem *);
 int sqlVdbeMemStringify(struct Mem *);
 int sqlVdbeMemNulTerminate(struct Mem *);
 int sqlVdbeMemExpandBlob(struct Mem *);
diff --git a/src/box/sql/vdbe.c b/src/box/sql/vdbe.c
index 71a827034..e61ad4251 100644
--- a/src/box/sql/vdbe.c
+++ b/src/box/sql/vdbe.c
@@ -1432,17 +1432,14 @@ case OP_AddImm: {            /* in1 */
  */
 case OP_MustBeInt: {            /* jump, in1 */
 	pIn1 = &aMem[pOp->p1];
-	if (!mem_is_int(pIn1)) {
-		mem_apply_type(pIn1, FIELD_TYPE_INTEGER);
-		if (!mem_is_int(pIn1)) {
-			if (pOp->p2==0) {
-				diag_set(ClientError, ER_SQL_TYPE_MISMATCH,
-					 mem_str(pIn1), "integer");
-				goto abort_due_to_error;
-			} else {
-				goto jump_to_p2;
-			}
-		}
+	if (mem_is_int(pIn1))
+		break;
+	if (mem_to_int_precise(pIn1) != 0) {
+		if (pOp->p2 !=0 )
+			goto jump_to_p2;
+		diag_set(ClientError, ER_SQL_TYPE_MISMATCH,
+			 mem_str(pIn1), "integer");
+		goto abort_due_to_error;
 	}
 	break;
 }