[Tarantool-patches] [PATCH v1 1/1] sql: fix comparison between DECIMAL and big DOUBLE

Mergen Imeev imeevma at tarantool.org
Wed Sep 1 11:52:35 MSK 2021


Thank you for the review! My answer below.

On Tue, Aug 31, 2021 at 10:46:40PM +0300, Timur Safin wrote:
> > From: imeevma at tarantool.org <imeevma at tarantool.org>
> > Subject: [PATCH v1 1/1] sql: fix comparison between DECIMAL and big
> > DOUBLE
> > 
> > This patch fixes comparison between DECIMAL value and DOUBLE values
> > greater or equal to 1e38 or less or equal to -1e38. Now any DOUBLE
> > value
> > greater or equal to 1e38 is more than any DECIMAL value and DOUBLE
> > value less or equal to -1e38 is less than any DECIMAL value.
> > 
> > Closes #6376
> ...
> > diff --git a/changelogs/unreleased/gh-6376-fix-incorrect-dec-inf-
> > cmp.md b/changelogs/unreleased/gh-6376-fix-incorrect-dec-inf-cmp.md
> > new file mode 100644
> > index 000000000..70de655f1
> > --- /dev/null
> > +++ b/changelogs/unreleased/gh-6376-fix-incorrect-dec-inf-cmp.md
> > @@ -0,0 +1,3 @@
> > +## bugfix/sql
> > +
> > +* Fixed wrong comparison between DECIMAL and large DOUBLE values
> > (gh-6376).
> > diff --git a/src/box/sql/mem.c b/src/box/sql/mem.c
> > index 4c40f15dc..a3ab31af5 100644
> > --- a/src/box/sql/mem.c
> > +++ b/src/box/sql/mem.c
> > @@ -2451,9 +2451,9 @@ mem_cmp_num(const struct Mem *a, const struct
> > Mem *b)
> >  		}
> >  		case MEM_TYPE_DOUBLE: {
> >  			if (b->u.r >= 1e38)
> > -				return 1;
> > -			if (b->u.r <= -1e38)
> >  				return -1;
> > +			if (b->u.r <= -1e38)
> > +				return 1;
> 
> Well, while we are here. I do understand that these kind of constants
> already spreading all corners of mem.c when you deal with decimals,
> but it's not entirely clear that this all about DECIMAL_MAX_DIGITS
> limitation in our decimal implementation. 
> And beyond all of that - hardcoded constant are evil. Could you please
> introduce any symbolic constant defines for such DECIMAL_MAX_DIGITS-
> derivative values? And use them wherever possible.
> 
> (I'm not insisting on fixing whole mem.c, that might be done separately,
> but at least this tricky case worth it)
> 
I agree that this is not good, but I think that addition of double constant for
decimal in SQL is not good idea. I think it is better to move all these
operations to decimal.c/.h. Could you fill an issue? If you cannot, I will do
it myself a bit later.

I also agree that I was wrong when I didn't introduce new functions for decimal,
but at that time I feared that I will spend too much time on tests for these
functions, and decided to leave this for later.

In general, I have quite a few questions about functions in decimal.c/.h and
I plan to ask them in the mentioned issue.

> >  			decimal_t dec;
> >  			decimal_t *d = decimal_from_double(&dec, b->u.r);
> >  			assert(d != NULL && d == &dec);
> 
> Thanks,
> Timur
> 


More information about the Tarantool-patches mailing list