Skip to content

H
h2database
Unverified 提交 13796c1a authored 作者: Noel Grandin's avatar Noel Grandin 提交者： GitHub
浏览文件
操作

Merge pull request #841 from katzyn/median 

Support indexes with nulls last for MEDIAN aggregate
上级 afb7546e 3a460ebb
隐藏空白字符变更
内嵌 并排
正在显示 包含 143 行增加 和 38 行删除
h2/src/main/org/h2/expression/AggregateDataMedian.java
浏览文件 @ 13796c1a
......@@ -42,6 +42,13 @@ import org.h2.value.ValueTimestampTimeZone;
class AggregateDataMedian extends AggregateData {
private Collection<Value> values;
private static boolean isNullsLast(Index index) {
IndexColumn ic = index.getIndexColumns()[0];
int sortType = ic.sortType;
return (sortType & SortOrder.NULLS_LAST) != 0
|| (sortType & SortOrder.DESCENDING) != 0 && (sortType & SortOrder.NULLS_FIRST) == 0;
}
static Index getMedianColumnIndex(Expression on) {
if (on instanceof ExpressionColumn) {
ExpressionColumn col = (ExpressionColumn) on;
......@@ -52,6 +59,7 @@ class AggregateDataMedian extends AggregateData {
ArrayList<Index> indexes = table.getIndexes();
Index result = null;
if (indexes != null) {
boolean nullable = column.isNullable();
for (int i = 1, size = indexes.size(); i < size; i++) {
Index index = indexes.get(i);
if (!index.canFindNext()) {
......@@ -60,18 +68,9 @@ class AggregateDataMedian extends AggregateData {
if (!index.isFirstColumn(column)) {
continue;
}
IndexColumn ic = index.getIndexColumns()[0];
if (column.isNullable()) {
int sortType = ic.sortType;
// Nulls last is not supported
if ((sortType & SortOrder.NULLS_LAST) != 0)
continue;
// Descending without nulls first is not supported
if ((sortType & SortOrder.DESCENDING) != 0 && (sortType & SortOrder.NULLS_FIRST) == 0) {
continue;
}
}
if (result == null || result.getColumns().length > index.getColumns().length) {
if (result == null || result.getColumns().length > index.getColumns().length
// Prefer index without nulls last for nullable columns
|| nullable && isNullsLast(result) && !isNullsLast(index)) {
result = index;
}
}
......@@ -90,32 +89,58 @@ class AggregateDataMedian extends AggregateData {
}
Cursor cursor = index.find(session, null, null);
cursor.next();
// Skip nulls
SearchRow row;
while (count > 0) {
row = cursor.getSearchRow();
if (row == null) {
int columnId = index.getColumns()[0].getColumnId();
ExpressionColumn expr = (ExpressionColumn) on;
if (expr.getColumn().isNullable()) {
boolean hasNulls = false;
SearchRow row;
/*
* Try to skip nulls from the start first with the same cursor that will be used
* to read values.
*/
while (count > 0) {
row = cursor.getSearchRow();
if (row == null) {
return ValueNull.INSTANCE;
}
if (row.getValue(columnId) == ValueNull.INSTANCE) {
count--;
cursor.next();
hasNulls = true;
} else
break;
}
if (count == 0) {
return ValueNull.INSTANCE;
}
if (row.getValue(index.getColumns()[0].getColumnId()) == ValueNull.INSTANCE) {
count--;
cursor.next();
} else
break;
}
if (count == 0) {
return ValueNull.INSTANCE;
/*
* If no nulls found and if index orders nulls last create a second cursor to
* count nulls at the end.
*/
if (!hasNulls && isNullsLast(index)) {
TableFilter tableFilter = expr.getTableFilter();
SearchRow check = tableFilter.getTable().getTemplateSimpleRow(true);
check.setValue(columnId, ValueNull.INSTANCE);
Cursor nullsCursor = index.find(session, check, check);
while (nullsCursor.next()) {
count--;
}
if (count <= 0) {
return ValueNull.INSTANCE;
}
}
}
long skip = (count - 1) / 2;
for (int i = 0; i < skip; i++) {
cursor.next();
}
row = cursor.getSearchRow();
Value v;
SearchRow row = cursor.getSearchRow();
if (row == null) {
v = ValueNull.INSTANCE;
} else {
v = row.getValue(index.getColumns()[0].getColumnId());
return ValueNull.INSTANCE;
}
Value v = row.getValue(columnId);
if (v == ValueNull.INSTANCE) {
return v;
}
if ((count & 1) == 0) {
cursor.next();
......@@ -123,7 +148,10 @@ class AggregateDataMedian extends AggregateData {
if (row == null) {
return v;
}
Value v2 = row.getValue(index.getColumns()[0].getColumnId());
Value v2 = row.getValue(columnId);
if (v2 == ValueNull.INSTANCE) {
return v;
}
return getMedian(v, v2, dataType, session.getDatabase().getCompareMode());
}
return v;
......@@ -144,6 +172,7 @@ class AggregateDataMedian extends AggregateData {
@Override
Value getValue(Database database, int dataType, boolean distinct) {
Collection<Value> c = values;
// Non-null collection cannot be empty here
if (c == null) {
return ValueNull.INSTANCE;
}
......@@ -214,15 +243,19 @@ class AggregateDataMedian extends AggregateData {
long dateSum = DateTimeUtils.absoluteDayFromDateValue(ts0.getDateValue())
+ DateTimeUtils.absoluteDayFromDateValue(ts1.getDateValue());
long nanos = (ts0.getTimeNanos() + ts1.getTimeNanos()) / 2;
int offset = ts0.getTimeZoneOffsetMins() + ts1.getTimeZoneOffsetMins();
if ((dateSum & 1) != 0) {
nanos += DateTimeUtils.NANOS_PER_DAY / 2;
if (nanos >= DateTimeUtils.NANOS_PER_DAY) {
nanos -= DateTimeUtils.NANOS_PER_DAY;
dateSum++;
}
}
if ((offset & 1) != 0) {
nanos += 30L * 1000000000;
}
if (nanos >= DateTimeUtils.NANOS_PER_DAY) {
nanos -= DateTimeUtils.NANOS_PER_DAY;
dateSum++;
}
return ValueTimestampTimeZone.fromDateValueAndNanos(DateTimeUtils.dateValueFromAbsoluteDay(dateSum / 2),
nanos, (short) ((ts0.getTimeZoneOffsetMins() + ts1.getTimeZoneOffsetMins()) / 2));
nanos, (short) (offset / 2));
}
default:
// Just return first
......
h2/src/test/org/h2/test/scripts/functions/aggregate/median.sql
浏览文件 @ 13796c1a
......@@ -633,13 +633,13 @@ select median(v) from test;
delete from test;
> update count: 5
insert into test values ('2000-01-20 20:00:00+10'), ('2000-01-21 20:00:00-09');
insert into test values ('2000-01-20 20:00:00+10:15'), ('2000-01-21 20:00:00-09');
> update count: 2
select median(v) from test;
> MEDIAN(V)
> ---------------------------
> 2000-01-21 08:00:00.0+00:30
> 2000-01-21 08:00:30.0+00:37
drop table test;
> ok
......@@ -662,3 +662,75 @@ select name, median(value) from test group by name order by name;
drop table test;
> ok
-- with filter
create table test(v int);
> ok
insert into test values (20), (20), (10);
> update count: 3
select median(v) from test where v <> 20;
> MEDIAN(V)
> ---------
> 10
create index test_idx on test(v asc);
> ok
select median(v) from test where v <> 20;
> MEDIAN(V)
> ---------
> 10
drop table test;
> ok
-- two-column index
create table test(v int, v2 int);
> ok
create index test_idx on test(v, v2);
> ok
insert into test values (20, 1), (10, 2), (20, 3);
> update count: 3
select median(v) from test;
> MEDIAN(V)
> ---------
> 20
drop table test;
> ok
-- not null column
create table test (v int not null);
> ok
create index test_idx on test(v desc);
> ok
select median(v) from test;
> MEDIAN(V)
> ---------
> null
insert into test values (10), (20);
> update count: 2
select median(v) from test;
> MEDIAN(V)
> ---------
> 15
insert into test values (20), (10), (20);
> update count: 3
select median(v) from test;
> MEDIAN(V)
> ---------
> 20
drop table test;
> ok
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论