[Tarantool-patches] [PATCH v4 11/53] sql: introduce mem_is_*() functions()
Vladislav Shpilevoy
v.shpilevoy at tarantool.org
Tue Mar 30 02:01:07 MSK 2021
Thanks for working on this!
I have a general comment affecting this entire naming schema.
The names seem too long. Besides, we already had some mem_is_*()
and mem_set_*() functions before this patch, which used short
names for the types.
What I mean is 'integer' -> 'int', 'boolean' -> 'bool',
'string' -> 'str', 'binary' -> 'bin', and so on.
For the integers we have several functions because we split
unsigned, signed, and always negative integers. So we would
need more int-like names. For instance,
mem_set_uint(uint64_t) - for MEM_UInt.
mem_set_nint(int64_t) - for MEM_Int.
mem_set_int(int64_t) - for both, checks the sign inside.
mem_set_sint(int64_t, bool) - for both, takes the sign flag
in the second argument
This can be discussed. The main point - shorter is better IMO.
See 14 comments below.
On 23.03.2021 10:35, Mergen Imeev via Tarantool-patches wrote:
> This patch introduces mem_is_*() functions that allows to check current
> MEM state.
>
> Part of #5818
> ---
> diff --git a/src/box/sql/func.c b/src/box/sql/func.c
> index e600a9800..81b537d9b 100644
> --- a/src/box/sql/func.c
> +++ b/src/box/sql/func.c
> @@ -736,8 +736,8 @@ substrFunc(sql_context * context, int argc, sql_value ** argv)
> context->is_aborted = true;
> return;
> }
> - if (sql_value_is_null(argv[1])
> - || (argc == 3 && sql_value_is_null(argv[2]))
> + if (mem_is_null(argv[1])
> + || (argc == 3 && mem_is_null(argv[2]))
1. This is not movement of huge code blocks, it is rather new code
now. And in the new code better use our code style. Such as
- || goes in the end of line, not in the beginning of a next line;
- Unary operators don't have a whitespace after them;
- Comparison with NULL should be explicit, no implicit boot casts.
I mark the places below which I was able to find after a swift look.
> ) {
> return;
> }
> @@ -1578,13 +1576,13 @@ replaceFunc(sql_context * context, int argc, sql_value ** argv)
> assert(zStr == sql_value_text(argv[0])); /* No encoding change */
> zPattern = sql_value_text(argv[1]);
> if (zPattern == 0) {
> - assert(sql_value_is_null(argv[1])
> + assert(mem_is_null(argv[1])
> || sql_context_db_handle(context)->mallocFailed);
> return;
> }
> nPattern = sql_value_bytes(argv[1]);
> if (nPattern == 0) {
> - assert(! sql_value_is_null(argv[1]));
> + assert(! mem_is_null(argv[1]));
2. Whitespace after unary operator.
> sql_result_value(context, argv[0]);
> return;
> }
> @@ -2039,7 +2035,7 @@ countStep(sql_context * context, int argc, sql_value ** argv)
> return;
> }
> p = sql_aggregate_context(context, sizeof(*p));
> - if ((argc == 0 || ! sql_value_is_null(argv[0])) && p) {
> + if ((argc == 0 || ! mem_is_null(argv[0])) && p) {
3. Ditto.
> p->n++;
> }
> }
> diff --git a/src/box/sql/mem.c b/src/box/sql/mem.c
> index ec6aaab64..abc9291ef 100644
> --- a/src/box/sql/mem.c
> +++ b/src/box/sql/mem.c
> @@ -37,6 +37,142 @@
> #include "box/tuple.h"
> #include "mpstream/mpstream.h"
>
> +bool
> +mem_is_null(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Null) != 0;
> +}
4. Maybe better move them all to mem.h. These one-liners easily
can be inlined (the ones which are <= 3 lines long could be moved).
> +
> +bool
> +mem_is_unsigned(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_UInt) != 0;
> +}
> +
> +bool
> +mem_is_string(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Str) != 0;
> +}
> +
> +bool
> +mem_is_number(const struct Mem *mem)
> +{
> + return (mem->flags & (MEM_Real | MEM_Int |MEM_UInt)) != 0;
5. Missed whitespace after the last '|'.
> +}
> +
> +bool
> +mem_is_double(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Real) != 0;
> +}
> +
> +bool
> +mem_is_integer(const struct Mem *mem)
> +{
> + return (mem->flags & (MEM_Int | MEM_UInt)) != 0;
> +}
> +
> +bool
> +mem_is_boolean(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Bool) != 0;
> +}
> +
> +bool
> +mem_is_binary(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Blob) != 0;
> +}
> +
> +bool
> +mem_is_map(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Blob) != 0 &&
> + (mem->flags & MEM_Subtype) != 0 &&
> + mem->subtype == SQL_SUBTYPE_MSGPACK &&
> + mp_typeof(*mem->z) == MP_MAP;
> +}
> +
> +bool
> +mem_is_array(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Blob) != 0 &&
> + (mem->flags & MEM_Subtype) != 0 &&
> + mem->subtype == SQL_SUBTYPE_MSGPACK &&
> + mp_typeof(*mem->z) == MP_ARRAY;
> +}
> +
> +bool
> +mem_is_aggregate(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Agg) != 0;
> +}
> +
> +bool
> +mem_is_varstring(const struct Mem *mem)
> +{
> + return (mem->flags & (MEM_Blob | MEM_Str)) != 0;
6. It does not look right to call it varstring if it includes
binary. A string is always binary, but not each binary object
is a string.
Maybe mem_is_bytes()? mem_is_bytearray()?
> +}
> +
> +bool
> +mem_is_frame(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Frame) != 0;
> +}
> +
> +bool
> +mem_is_undefined(const struct Mem *mem)
> +{
> + return (mem->flags & MEM_Undefined) != 0;
> +}
> +
> +bool
> +mem_is_static(const struct Mem *mem)
> +{
> + return (mem->flags & (MEM_Str | MEM_Blob)) != 0 &&
> + (mem->flags & MEM_Static) != 0;
> +}
> +
> +bool
> +mem_is_ephemeral(const struct Mem *mem)
> +{
> + return (mem->flags & (MEM_Str | MEM_Blob)) != 0 &&
> + (mem->flags & MEM_Ephem) != 0;
7. How can it be that MEM_Ephem is set, but Str/Blob are not?
> +}
> +
> +bool
> +mem_is_dynamic(const struct Mem *mem)
> +{
> + return (mem->flags & (MEM_Str | MEM_Blob)) != 0 &&
> + (mem->flags & MEM_Dyn) != 0;
> +}
> diff --git a/src/box/sql/vdbe.c b/src/box/sql/vdbe.c
> index 12712efb4..05e0f78c1 100644
> --- a/src/box/sql/vdbe.c
> +++ b/src/box/sql/vdbe.c
> @@ -1088,10 +1086,10 @@ case OP_Concat: { /* same as TK_CONCAT, in1, in2, out3 */
> * Concatenation operation can be applied only to
> * strings and blobs.
> */
> - uint32_t str_type_p1 = pIn1->flags & (MEM_Blob | MEM_Str);
> - uint32_t str_type_p2 = pIn2->flags & (MEM_Blob | MEM_Str);
> - if (str_type_p1 == 0 || str_type_p2 == 0) {
> - char *inconsistent_type = str_type_p1 == 0 ?
> + bool str_type_p1 = mem_is_varstring(pIn1);
> + bool str_type_p2 = mem_is_varstring(pIn2);
8. They are not types now. Only flags. Should be renamed to something
more appropriate.
> + if (!str_type_p1 || !str_type_p2) {
> + char *inconsistent_type = !str_type_p1 ?
> mem_type_to_str(pIn1) :
> mem_type_to_str(pIn2);
> diag_set(ClientError, ER_INCONSISTENT_TYPES,
> @@ -1100,7 +1098,7 @@ case OP_Concat: { /* same as TK_CONCAT, in1, in2, out3 */
> }
>
> /* Moreover, both operands must be of the same type. */
> - if (str_type_p1 != str_type_p2) {
> + if (mem_is_string(pIn1) != mem_is_string(pIn2)) {
9. I would recommend mem_is_same_type(). Up to you.
> diag_set(ClientError, ER_INCONSISTENT_TYPES,
> mem_type_to_str(pIn2), mem_type_to_str(pIn1));
> goto abort_due_to_error;
> @@ -1186,14 +1183,16 @@ case OP_Remainder: { /* same as TK_REM, in1, in2, out3 */
> pIn2 = &aMem[pOp->p2];
> type2 = numericType(pIn2);
> pOut = vdbe_prepare_null_out(p, pOp->p3);
> - flags = pIn1->flags | pIn2->flags;
> - if ((flags & MEM_Null)!=0) goto arithmetic_result_is_null;
> + if (mem_is_null(pIn1) || mem_is_null(pIn2))
> + goto arithmetic_result_is_null;
> if ((type1 & (MEM_Int | MEM_UInt)) != 0 &&
> (type2 & (MEM_Int | MEM_UInt)) != 0) {
> iA = pIn1->u.i;
> iB = pIn2->u.i;
> - bool is_lhs_neg = pIn1->flags & MEM_Int;
> - bool is_rhs_neg = pIn2->flags & MEM_Int;
> + bool is_lhs_neg = mem_is_integer(pIn1) &&
> + !mem_is_unsigned(pIn1);
> + bool is_rhs_neg = mem_is_integer(pIn2) &&
> + !mem_is_unsigned(pIn2);
10. The checks look overcomplicated. Worth adding mem_is_nint()?
> bool is_res_neg;
> switch( pOp->opcode) {
> case OP_Add: {
> @@ -1509,7 +1508,7 @@ case OP_ShiftRight: { /* same as TK_RSHIFT, in1, in2, out3 */
> pIn1 = &aMem[pOp->p1];
> pIn2 = &aMem[pOp->p2];
> pOut = vdbe_prepare_null_out(p, pOp->p3);
> - if ((pIn1->flags | pIn2->flags) & MEM_Null) {
> + if (mem_is_null(pIn1) || mem_is_null(pIn2)) {
11. This is at least third time I see the check of kind "one of them is null".
And I see more below. Probably worth adding a function which would do it more
efficient, in one check: mem_is_any_null(mem1, mem2) or something. Up to you.
> @@ -1757,11 +1756,10 @@ case OP_Ge: { /* same as TK_GE, jump, in1, in3 */
> * or not both operands are null.
> */
> assert(pOp->opcode==OP_Eq || pOp->opcode==OP_Ne);
> - assert((flags1 & MEM_Cleared)==0);
> + assert(!mem_is_cleared(pIn1));
> assert((pOp->p5 & SQL_JUMPIFNULL)==0);
> - if ((flags1&flags3&MEM_Null)!=0
> - && (flags3&MEM_Cleared)==0
> - ) {
> + if (mem_is_null(pIn1) && mem_is_null(pIn3) &&
12. You already know pIn1 or pIn3 is NULL from the 'if' above. So it
would be just a bit faster and easier to do mem_is_same_type(). Only
one branch.
> + !mem_is_cleared(pIn3)) {
> res = 0; /* Operands are equal */
> } else {
> res = 1; /* Operands are not equal */
> @@ -2982,18 +2965,18 @@ case OP_SeekGT: { /* jump, in3 */
> * the seek, so convert it.
> */
> pIn3 = &aMem[int_field];
> - if ((pIn3->flags & MEM_Null) != 0)
> + if (mem_is_null(pIn3))
> goto skip_truncate;
> - if ((pIn3->flags & MEM_Str) != 0)
> + if (mem_is_string(pIn3))
> mem_apply_numeric_type(pIn3);
> int64_t i;
> - if ((pIn3->flags & MEM_Int) == MEM_Int) {
> + if (mem_is_integer(pIn3) && !mem_is_unsigned(pIn3)) {
13. Better be mem_is_nint(), mentioned in one of the previous
comments.
> @@ -3352,11 +3336,11 @@ case OP_FCopy: { /* out2 */
> pIn1 = &aMem[pOp->p1];
> }
>
> - if ((pOp->p3 & OPFLAG_NOOP_IF_NULL) && (pIn1->flags & MEM_Null)) {
> + if ((pOp->p3 & OPFLAG_NOOP_IF_NULL) && mem_is_null(pIn1)) {
14. Should be explicit != 0.
> pOut = vdbe_prepare_null_out(p, pOp->p2);
> } else {
> assert(memIsValid(pIn1));
> - assert((pIn1->flags & (MEM_Int | MEM_UInt)) != 0);
> + assert(mem_is_integer(pIn1));
More information about the Tarantool-patches
mailing list