spark-instrumented-optimizer/sql/core
Herman van Hovell f14ae4900a [SPARK-18300][SQL] Do not apply foldable propagation with expand as a child.
## What changes were proposed in this pull request?
The `FoldablePropagation` optimizer rule, pulls foldable values out from under an `Expand`. This breaks the `Expand` in two ways:

- It rewrites the output attributes of the `Expand`. We explicitly define output attributes for `Expand`, these are (unfortunately) considered as part of the expressions of the `Expand` and can be rewritten.
- Expand can actually change the column (it will typically re-use the attributes or the underlying plan). This means that we cannot safely propagate the expressions from under an `Expand`.

This PR fixes this and (hopefully) other issues by explicitly whitelisting allowed operators.

## How was this patch tested?
Added tests to `FoldablePropagationSuite` and to `SQLQueryTestSuite`.

Author: Herman van Hovell <hvanhovell@databricks.com>

Closes #15857 from hvanhovell/SPARK-18300.
2016-11-15 06:59:25 -08:00
..
benchmarks [SPARK-17335][SQL] Fix ArrayType and MapType CatalogString. 2016-09-03 19:02:20 +02:00
src [SPARK-18300][SQL] Do not apply foldable propagation with expand as a child. 2016-11-15 06:59:25 -08:00
pom.xml [SPARK-17346][SQL][TEST-MAVEN] Generate the sql test jar to fix the maven build 2016-10-05 18:11:31 -07:00