Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Skip to content

Commit 7538708

Browse files
committed
Avoid gratuitous inaccuracy in numeric width_bucket().
Multiply before dividing, not the reverse, so that cases that should produce exact results do produce exact results. (width_bucket_float8 got this right already.) Even when the result is inexact, this avoids making it more inexact, since only the division step introduces any imprecision. While at it, fix compute_bucket() to not uselessly repeat the sign check already done by its caller, and avoid duplicating the multiply/divide steps by adjusting variable usage. Per complaint from Martin Visser. Although this seems like a bug fix, I'm hesitant to risk changing width_bucket()'s results in stable branches, so no back-patch. Discussion: https://postgr.es/m/6FA5117D-6AED-4656-8FEF-B74AC18FAD85@brytlyt.com
1 parent 8ce423b commit 7538708

File tree

3 files changed

+65
-15
lines changed

3 files changed

+65
-15
lines changed

src/backend/utils/adt/numeric.c

+16-15
Original file line numberDiff line numberDiff line change
@@ -592,7 +592,8 @@ static void round_var(NumericVar *var, int rscale);
592592
static void trunc_var(NumericVar *var, int rscale);
593593
static void strip_var(NumericVar *var);
594594
static void compute_bucket(Numeric operand, Numeric bound1, Numeric bound2,
595-
const NumericVar *count_var, NumericVar *result_var);
595+
const NumericVar *count_var, bool reversed_bounds,
596+
NumericVar *result_var);
596597

597598
static void accum_sum_add(NumericSumAccum *accum, const NumericVar *var1);
598599
static void accum_sum_rescale(NumericSumAccum *accum, const NumericVar *val);
@@ -1752,8 +1753,8 @@ width_bucket_numeric(PG_FUNCTION_ARGS)
17521753
else if (cmp_numerics(operand, bound2) >= 0)
17531754
add_var(&count_var, &const_one, &result_var);
17541755
else
1755-
compute_bucket(operand, bound1, bound2,
1756-
&count_var, &result_var);
1756+
compute_bucket(operand, bound1, bound2, &count_var, false,
1757+
&result_var);
17571758
break;
17581759

17591760
/* bound1 > bound2 */
@@ -1763,8 +1764,8 @@ width_bucket_numeric(PG_FUNCTION_ARGS)
17631764
else if (cmp_numerics(operand, bound2) <= 0)
17641765
add_var(&count_var, &const_one, &result_var);
17651766
else
1766-
compute_bucket(operand, bound1, bound2,
1767-
&count_var, &result_var);
1767+
compute_bucket(operand, bound1, bound2, &count_var, true,
1768+
&result_var);
17681769
break;
17691770
}
17701771

@@ -1783,11 +1784,13 @@ width_bucket_numeric(PG_FUNCTION_ARGS)
17831784
/*
17841785
* If 'operand' is not outside the bucket range, determine the correct
17851786
* bucket for it to go. The calculations performed by this function
1786-
* are derived directly from the SQL2003 spec.
1787+
* are derived directly from the SQL2003 spec. Note however that we
1788+
* multiply by count before dividing, to avoid unnecessary roundoff error.
17871789
*/
17881790
static void
17891791
compute_bucket(Numeric operand, Numeric bound1, Numeric bound2,
1790-
const NumericVar *count_var, NumericVar *result_var)
1792+
const NumericVar *count_var, bool reversed_bounds,
1793+
NumericVar *result_var)
17911794
{
17921795
NumericVar bound1_var;
17931796
NumericVar bound2_var;
@@ -1797,23 +1800,21 @@ compute_bucket(Numeric operand, Numeric bound1, Numeric bound2,
17971800
init_var_from_num(bound2, &bound2_var);
17981801
init_var_from_num(operand, &operand_var);
17991802

1800-
if (cmp_var(&bound1_var, &bound2_var) < 0)
1803+
if (!reversed_bounds)
18011804
{
18021805
sub_var(&operand_var, &bound1_var, &operand_var);
18031806
sub_var(&bound2_var, &bound1_var, &bound2_var);
1804-
div_var(&operand_var, &bound2_var, result_var,
1805-
select_div_scale(&operand_var, &bound2_var), true);
18061807
}
18071808
else
18081809
{
18091810
sub_var(&bound1_var, &operand_var, &operand_var);
1810-
sub_var(&bound1_var, &bound2_var, &bound1_var);
1811-
div_var(&operand_var, &bound1_var, result_var,
1812-
select_div_scale(&operand_var, &bound1_var), true);
1811+
sub_var(&bound1_var, &bound2_var, &bound2_var);
18131812
}
18141813

1815-
mul_var(result_var, count_var, result_var,
1816-
result_var->dscale + count_var->dscale);
1814+
mul_var(&operand_var, count_var, &operand_var,
1815+
operand_var.dscale + count_var->dscale);
1816+
div_var(&operand_var, &bound2_var, result_var,
1817+
select_div_scale(&operand_var, &bound2_var), true);
18171818
add_var(result_var, &const_one, result_var);
18181819
floor_var(result_var, result_var);
18191820

src/test/regress/expected/numeric.out

+40
Original file line numberDiff line numberDiff line change
@@ -1385,6 +1385,46 @@ SELECT width_bucket('Infinity'::float8, 1, 10, 10),
13851385
(1 row)
13861386

13871387
DROP TABLE width_bucket_test;
1388+
-- Simple test for roundoff error when results should be exact
1389+
SELECT x, width_bucket(x::float8, 10, 100, 9) as flt,
1390+
width_bucket(x::numeric, 10, 100, 9) as num
1391+
FROM generate_series(0, 110, 10) x;
1392+
x | flt | num
1393+
-----+-----+-----
1394+
0 | 0 | 0
1395+
10 | 1 | 1
1396+
20 | 2 | 2
1397+
30 | 3 | 3
1398+
40 | 4 | 4
1399+
50 | 5 | 5
1400+
60 | 6 | 6
1401+
70 | 7 | 7
1402+
80 | 8 | 8
1403+
90 | 9 | 9
1404+
100 | 10 | 10
1405+
110 | 10 | 10
1406+
(12 rows)
1407+
1408+
SELECT x, width_bucket(x::float8, 100, 10, 9) as flt,
1409+
width_bucket(x::numeric, 100, 10, 9) as num
1410+
FROM generate_series(0, 110, 10) x;
1411+
x | flt | num
1412+
-----+-----+-----
1413+
0 | 10 | 10
1414+
10 | 10 | 10
1415+
20 | 9 | 9
1416+
30 | 8 | 8
1417+
40 | 7 | 7
1418+
50 | 6 | 6
1419+
60 | 5 | 5
1420+
70 | 4 | 4
1421+
80 | 3 | 3
1422+
90 | 2 | 2
1423+
100 | 1 | 1
1424+
110 | 0 | 0
1425+
(12 rows)
1426+
1427+
--
13881428
-- TO_CHAR()
13891429
--
13901430
SELECT '' AS to_char_1, to_char(val, '9G999G999G999G999G999')

src/test/regress/sql/numeric.sql

+9
Original file line numberDiff line numberDiff line change
@@ -888,6 +888,15 @@ SELECT width_bucket('Infinity'::float8, 1, 10, 10),
888888

889889
DROP TABLE width_bucket_test;
890890

891+
-- Simple test for roundoff error when results should be exact
892+
SELECT x, width_bucket(x::float8, 10, 100, 9) as flt,
893+
width_bucket(x::numeric, 10, 100, 9) as num
894+
FROM generate_series(0, 110, 10) x;
895+
SELECT x, width_bucket(x::float8, 100, 10, 9) as flt,
896+
width_bucket(x::numeric, 100, 10, 9) as num
897+
FROM generate_series(0, 110, 10) x;
898+
899+
--
891900
-- TO_CHAR()
892901
--
893902
SELECT '' AS to_char_1, to_char(val, '9G999G999G999G999G999')

0 commit comments

Comments
 (0)