[Tarantool-patches] [PATCH v1 3/4] sql: introduce cast for decimal
Safin Timur
tsafin at tarantool.org
Mon Aug 16 22:34:16 MSK 2021
Oh, now I see the reasonable implicit and explicit casts table for
decimals. I disavow my prior complain about missing directions for
CAST(exprt AS DECIMAL).
Below there is minor complain though...
On 16.08.2021 18:57, Mergen Imeev via Tarantool-patches wrote:
> This patch introduces implicit and explicit cast of the DECIMAL field in
> SQL. Implicit cast to and from DECIMAL is precise. Explicit conversion
> to decimal and back could lead to loss of precision.
>
> Part of #4415
> ---
> src/box/sql/mem.c | 334 ++++++++++++++++++++++++++++++-
> test/sql-tap/decimal.test.lua | 366 +++++++++++++++++++++++++++++++++-
> 2 files changed, 695 insertions(+), 5 deletions(-)
>
> diff --git a/src/box/sql/mem.c b/src/box/sql/mem.c
> index 016f0e80b..ff8b40d7f 100644
> --- a/src/box/sql/mem.c
> +++ b/src/box/sql/mem.c
> @@ -678,6 +678,17 @@ int_to_double_forced(struct Mem *mem)
> return CMP_OLD_NEW(i, d, int64_t);
> }
>
> +static inline int
> +int_to_dec(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_INT);
> + int64_t i = mem->u.i;
> + decimal_from_int64(&mem->u.d, i);
> + mem->type = MEM_TYPE_DEC;
> + mem->flags = 0;
> + return 0;
> +}
> +
> static inline int
> uint_to_double_precise(struct Mem *mem)
> {
> @@ -709,6 +720,17 @@ uint_to_double_forced(struct Mem *mem)
> return CMP_OLD_NEW(u, d, uint64_t);
> }
>
> +static inline int
> +uint_to_dec(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_UINT);
> + int64_t u = mem->u.u;
> + decimal_from_uint64(&mem->u.d, u);
> + mem->type = MEM_TYPE_DEC;
> + mem->flags = 0;
> + return 0;
> +}
> +
> static inline int
> int_to_str0(struct Mem *mem)
> {
> @@ -855,6 +877,19 @@ str_to_double(struct Mem *mem)
> return 0;
> }
>
> +static inline int
> +str_to_dec(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_STR);
> + decimal_t dec;
> + decimal_t *d;
> + d = decimal_from_string(&dec, mem->z);
> + if (d == NULL)
> + return -1;
> + mem_set_dec(mem, &dec);
> + return 0;
> +}
> +
> static inline int
> double_to_int(struct Mem *mem)
> {
> @@ -988,6 +1023,69 @@ double_to_uint_forced(struct Mem *mem)
> return res;
> }
>
> +static inline int
> +double_to_dec(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DOUBLE);
> + double d = mem->u.r;
> + if (d >= 1e38 || d <= -1e38)
> + return -1;
> + decimal_from_double(&mem->u.d, d);
> + mem->type = MEM_TYPE_DEC;
> + mem->flags = 0;
> + return 0;
> +}
> +
> +static inline int
> +double_to_dec_precise(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DOUBLE);
> + double d = mem->u.r;
> + if (d >= 1e38 || d <= -1e38)
> + return -1;
> + decimal_t dec;
> + decimal_from_double(&dec, d);
> + if (atof(decimal_str(&mem->u.d)) != d)
> + return -1;
> + mem->u.d = dec;
> + mem->type = MEM_TYPE_DEC;
> + mem->flags = 0;
> + return 0;
> +}
> +
> +/**
> + * Cast MEM with DOUBLE to DECIMAL. Doesn't fail. The return value is < 0
> if
> + * the original value is less than the result, > 0 if the original value
> is
> + * greater than the result, and 0 if the cast is precise.
> + */
> +static inline int
> +double_to_dec_forced(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DOUBLE);
> + double d = mem->u.r;
> + mem->type = MEM_TYPE_DEC;
> + mem->flags = 0;
> + if (d >= 1e38) {
> + const char *val =
> "99999999999999999999999999999999999999";
> + assert(strlen(val) == 38);
> + decimal_from_string(&mem->u.d, val);
> + return 1;
> + }
> + if (d <= -1e38) {
> + const char *val =
> "-99999999999999999999999999999999999999";
> + assert(strlen(val) == 39);
> + decimal_from_string(&mem->u.d, val);
> + return -1;
> + }
> + decimal_from_double(&mem->u.d, d);
> + double tmp = atof(decimal_str(&mem->u.d));
> + if (d > tmp)
> + return 1;
> + if (d < tmp)
> + return -1;
> + return 0;
> +}
> +
> static inline int
> double_to_str0(struct Mem *mem)
> {
> @@ -1001,6 +1099,173 @@ double_to_str0(struct Mem *mem)
> return 0;
> }
>
> +static inline int
> +dec_to_int(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + if (decimal_is_neg(&mem->u.d)) {
> + int64_t i;
> + if (decimal_to_int64(&mem->u.d, &i) == NULL)
> + return -1;
> + assert(i < 0);
> + mem->u.i = i;
> + mem->type = MEM_TYPE_INT;
> + mem->flags = 0;
> + return 0;
> + }
> + uint64_t u;
> + if (decimal_to_uint64(&mem->u.d, &u) == NULL)
> + return -1;
> + mem->u.u = u;
> + mem->type = MEM_TYPE_UINT;
> + mem->flags = 0;
> + return 0;
> +}
> +
> +static inline int
> +dec_to_int_precise(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + if (!decimal_is_int(&mem->u.d))
> + return -1;
> + return dec_to_int(mem);
> +}
> +
> +static inline int
> +dec_to_int_forced(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + if (decimal_is_neg(&mem->u.d)) {
> + int64_t i;
> + mem->type = MEM_TYPE_INT;
> + mem->flags = 0;
> + if (decimal_to_int64(&mem->u.d, &i) == NULL) {
> + mem->u.i = INT64_MIN;
> + return -1;
> + }
> + assert(i < 0);
> + mem->u.i = i;
> + /*
> + * Decimal is floored when cast to int, which means that
> after
> + * cast it becomes bigger if it was not integer.
> + */
> + return decimal_is_int(&mem->u.d) ? 0 : -1;
> + }
> + uint64_t u;
> + mem->type = MEM_TYPE_UINT;
> + mem->flags = 0;
> + if (decimal_to_uint64(&mem->u.d, &u) == NULL) {
> + mem->u.u = UINT64_MAX;
> + return 1;
> + }
> + mem->u.u = u;
> + /*
> + * Decimal is floored when cast to uint, which means that after
> cast it
> + * becomes less if it was not integer.
> + */
> + return decimal_is_int(&mem->u.d) ? 0 : 1;
> +}
> +
> +static inline int
> +dec_to_uint(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + uint64_t u;
> + if (decimal_to_uint64(&mem->u.d, &u) == NULL)
> + return -1;
> + mem->u.u = u;
> + mem->type = MEM_TYPE_UINT;
> + mem->flags = 0;
> + return 0;
> +}
> +
> +static inline int
> +dec_to_uint_precise(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + if (!decimal_is_int(&mem->u.d))
> + return -1;
> + return dec_to_uint(mem);
> +}
> +
> +static inline int
> +dec_to_uint_forced(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + uint64_t u;
> + mem->type = MEM_TYPE_UINT;
> + mem->flags = 0;
> + if (decimal_to_uint64(&mem->u.d, &u) == NULL) {
> + if (decimal_is_neg(&mem->u.d)) {
> + mem->u.u = 0;
> + return -1;
> + }
> + mem->u.u = UINT64_MAX;
> + return 1;
> + }
> + mem->u.u = u;
> + /*
> + * Decimal is floored when cast to uint, which means that after
> cast if
> + * if was not integer it becomes less if it was positive, and move
> if it
> + * was negative.
> + */
> + if (decimal_is_int(&mem->u.d))
> + return 0;
> + return decimal_is_neg(&mem->u.d) ? -1 : 1;
> +}
> +
> +static inline int
> +dec_to_double(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + double r = atof(decimal_str(&mem->u.d));
> + mem->u.r = r;
> + mem->type = MEM_TYPE_DOUBLE;
> + mem->flags = 0;
> + return 0;
> +}
> +
> +static inline int
> +dec_to_double_precise(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + double r = atof(decimal_str(&mem->u.d));
> + decimal_t d;
> + decimal_t *dec = decimal_from_double(&d, r);
> + if (dec == NULL || decimal_compare(dec, &mem->u.d) != 0)
> + return -1;
> + mem->u.r = r;
> + mem->type = MEM_TYPE_DOUBLE;
> + mem->flags = 0;
> + return 0;
> +}
> +
> +static inline int
> +dec_to_double_forced(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + mem->type = MEM_TYPE_DOUBLE;
> + mem->flags = 0;
> + double r = atof(decimal_str(&mem->u.d));
> + int res;
> + decimal_t d;
> + if (r <= -1e38)
> + res = 1;
> + else if (r >= 1e38)
> + res = -1;
> + else
> + res = decimal_compare(&mem->u.d, decimal_from_double(&d,
> r));
> + mem->u.r = r;
> + return res;
> +}
> +
> +static inline int
> +dec_to_str0(struct Mem *mem)
> +{
> + assert(mem->type == MEM_TYPE_DEC);
> + return mem_copy_str0(mem, decimal_str(&mem->u.d));
> +}
> +
> static inline int
> bool_to_str0(struct Mem *mem)
> {
> @@ -1053,6 +1318,8 @@ mem_to_int(struct Mem *mem)
> return str_to_int(mem);
> if (mem->type == MEM_TYPE_DOUBLE)
> return double_to_int(mem);
> + if (mem->type == MEM_TYPE_DEC)
> + return dec_to_int(mem);
> return -1;
> }
>
> @@ -1068,6 +1335,8 @@ mem_to_int_precise(struct Mem *mem)
> return str_to_int(mem);
> if (mem->type == MEM_TYPE_DOUBLE)
> return double_to_int_precise(mem);
> + if (mem->type == MEM_TYPE_DEC)
> + return dec_to_int(mem);
> return -1;
> }
>
> @@ -1083,6 +1352,8 @@ mem_to_double(struct Mem *mem)
> return int_to_double(mem);
> if (mem->type == MEM_TYPE_STR)
> return str_to_double(mem);
> + if (mem->type == MEM_TYPE_DEC)
> + return dec_to_double(mem);
> return -1;
> }
>
> @@ -1129,6 +1400,8 @@ mem_to_str0(struct Mem *mem)
> return array_to_str0(mem);
> case MEM_TYPE_UUID:
> return uuid_to_str0(mem);
> + case MEM_TYPE_DEC:
> + return dec_to_str0(mem);
> default:
> return -1;
> }
> @@ -1157,6 +1430,8 @@ mem_to_str(struct Mem *mem)
> return array_to_str0(mem);
> case MEM_TYPE_UUID:
> return uuid_to_str0(mem);
> + case MEM_TYPE_DEC:
> + return dec_to_str0(mem);
> default:
> return -1;
> }
> @@ -1177,6 +1452,8 @@ mem_cast_explicit(struct Mem *mem, enum field_type
> type)
> return str_to_uint(mem);
> case MEM_TYPE_DOUBLE:
> return double_to_uint(mem);
> + case MEM_TYPE_DEC:
> + return dec_to_uint(mem);
> default:
> return -1;
> }
> @@ -1209,9 +1486,21 @@ mem_cast_explicit(struct Mem *mem, enum field_type
> type)
> case FIELD_TYPE_NUMBER:
> return mem_to_number(mem);
> case FIELD_TYPE_DECIMAL:
> - if (mem->type == MEM_TYPE_DEC)
> + switch (mem->type) {
> + case MEM_TYPE_INT:
> + return int_to_dec(mem);
> + case MEM_TYPE_UINT:
> + return uint_to_dec(mem);
> + case MEM_TYPE_STR:
> + return str_to_dec(mem);
> + case MEM_TYPE_DOUBLE:
> + return double_to_dec(mem);
> + case MEM_TYPE_DEC:
> + mem->flags = 0;
> return 0;
> - return -1;
> + default:
> + return -1;
> + }
For consistency with majority of other numeric directions (mem_to_int(),
mem_to_number(), mem_to_double()) I'd extract this code similarly to
mem_to_decimal(). But this is up to you...
> case FIELD_TYPE_UUID:
> if (mem->type == MEM_TYPE_UUID) {
> mem->flags = 0;
...
> @@ -1296,9 +1591,19 @@ mem_cast_implicit(struct Mem *mem, enum field_type
> type)
> mem->flags = MEM_Number;
> return 0;
> case FIELD_TYPE_DECIMAL:
> - if (mem->type == MEM_TYPE_DEC)
> + switch (mem->type) {
> + case MEM_TYPE_INT:
> + return int_to_dec(mem);
> + case MEM_TYPE_UINT:
> + return uint_to_dec(mem);
> + case MEM_TYPE_DOUBLE:
> + return double_to_dec_precise(mem);
> + case MEM_TYPE_DEC:
> + mem->flags = 0;
> return 0;
> - return -1;
> + default:
> + return -1;
> + }
> case FIELD_TYPE_MAP:
> if (mem->type == MEM_TYPE_MAP)
> return 0;
> @@ -1343,6 +1648,8 @@ mem_cast_implicit_number(struct Mem *mem, enum
> field_type type)
> return -1;
> case MEM_TYPE_DOUBLE:
> return double_to_uint_forced(mem);
> + case MEM_TYPE_DEC:
> + return dec_to_uint_forced(mem);
> default:
> unreachable();
> }
> @@ -1353,6 +1660,8 @@ mem_cast_implicit_number(struct Mem *mem, enum
> field_type type)
> return int_to_double_forced(mem);
> case MEM_TYPE_UINT:
> return uint_to_double_forced(mem);
> + case MEM_TYPE_DEC:
> + return dec_to_double_forced(mem);
> case MEM_TYPE_DOUBLE:
> mem->flags = 0;
> return 0;
> @@ -1368,6 +1677,23 @@ mem_cast_implicit_number(struct Mem *mem, enum
> field_type type)
> return 0;
> case MEM_TYPE_DOUBLE:
> return double_to_int_forced(mem);
> + case MEM_TYPE_DEC:
> + return dec_to_int_forced(mem);
> + default:
> + unreachable();
> + }
> + break;
> + case FIELD_TYPE_DECIMAL:
> + switch (mem->type) {
> + case MEM_TYPE_INT:
> + return int_to_dec(mem);
> + case MEM_TYPE_UINT:
> + return uint_to_dec(mem);
> + case MEM_TYPE_DEC:
> + mem->flags = 0;
> + return 0;
> + case MEM_TYPE_DOUBLE:
> + return double_to_dec_forced(mem);
> default:
> unreachable();
> }
Otherwise great patch, with the expected extension to explicit/implicit
casts tables.
Thanks,
Timur
More information about the Tarantool-patches
mailing list