[Tarantool-patches] [PATCH 1/1] sql: introduce UUID field type
Mergen Imeev
imeevma at tarantool.org
Thu May 27 18:33:09 MSK 2021
Hi! Thank you for suggestions. My answers below.
On Wed, May 26, 2021 at 01:58:41AM +0300, Timur Safin wrote:
> There is one major issue I'd like to discuss, please see below...
>
> : From: Mergen Imeev
> : Subject: Re: [Tarantool-patches] [PATCH 1/1] sql: introduce UUID field type
> :
> ...
>
> : +static inline int
> : +str_to_uuid(struct Mem *mem)
> : +{
> : + assert(mem->type == MEM_TYPE_STR);
> : + struct tt_uuid uuid;
> : + if (mem->n != UUID_STR_LEN ||
> : + tt_uuid_from_string(tt_cstr(mem->z, mem->n), &uuid) != 0)
>
> Here lies the limitation I've specifically wanted to avoid and put
> below to RFC:
>
> "The most complete implementation seemingly is in PostgreSQL,
> which allows various relaxed formats for string literals
> which may be accepted as UUID - the plan is to be as close
> as possible to PostgreSQL here."
>
> Checking for UUID_STR_LEN restricts only to 1 specific form with
> dashes in specific positions, but we would rather want to allow
> relaxed formats also, with some dashes omitted. Also with optional
> braces around. i.e. putting this as regexp
>
> \{?[0-9a-f]{8}-?[0-9a-f]{4}-?[0-9a-f]{4}-?[0-9a-f]{4}-?[0-9a-f]{12}\}?
>
> And yes, I mean we need to reimplement tt_uuid_from_string() differently.
> (If you were asking me then I'd use RE2C for that purpose, but that up
> to you)
>
Addition of support of other versions of UUID should not be done as part of this
issue, since it is problem of BOX, not just SQL.
> : + return -1;
> : + mem_set_uuid(mem, &uuid);
> : + return 0;
> : +}
> : +
> : static inline int
> : str_to_bool(struct Mem *mem)
> : {
> ...
> : diff --git a/src/box/sql/mem.h b/src/box/sql/mem.h
> : index 15d97da0e..b3cd5c545 100644
> : --- a/src/box/sql/mem.h
> : +++ b/src/box/sql/mem.h
> : @@ -30,6 +30,7 @@
> : * SUCH DAMAGE.
> : */
> : #include "box/field_def.h"
> : +#include "uuid/tt_uuid.h"
> :
> : struct sql;
> : struct Vdbe;
> : @@ -47,10 +48,11 @@ enum mem_type {
> : MEM_TYPE_MAP = 1 << 6,
> : MEM_TYPE_BOOL = 1 << 7,
> : MEM_TYPE_DOUBLE = 1 << 8,
> : - MEM_TYPE_INVALID = 1 << 9,
> : - MEM_TYPE_FRAME = 1 << 10,
> : - MEM_TYPE_PTR = 1 << 11,
> : - MEM_TYPE_AGG = 1 << 12,
> : + MEM_TYPE_UUID = 1 << 9,
> : + MEM_TYPE_INVALID = 1 << 10,
> : + MEM_TYPE_FRAME = 1 << 11,
> : + MEM_TYPE_PTR = 1 << 12,
> : + MEM_TYPE_AGG = 1 << 13,
>
> I guess there should be no incompatibility problems here, but just
> in case I ask - why did you insert new constant to the middle of prior list?
>
Before MEM_TYPE_INVALID are types that are accesisble to user. Starting with
MEM_TYPE_INVALID are for internal use.
> Also I see there is another omission from RFC - there is nothing added
> for SQL built-in support to generate UUID from within SQL mode.
>
> "Introduce UUID([version#]) functions which would allow generating
> any particular type of GUID. If version argument is omitted then
> we generate UUID v4 (as used by default in box);"
>
I will add this function.
> Regards,
> Timur
>
More information about the Tarantool-patches
mailing list