Github user paul-rogers commented on the issue:
https://github.com/apache/drill/pull/1085
### Error Handling
The column writer work introduces a new scan operator mechanism. Part of
that mechanism is to standardize error handling. Part of that work is, in turn,
to add a a couple new error types to UserError to better categorize errors.
### Star Column
Organized our many references to the "star column" to make it more readily
available to the new scan framework.
### Date/Time Utilities
The various date/time vectors have very handy methods to convert the vector
representation into Joda objects. This PR pulls that code out into a utility
class that can be used in other places. This change uncovered a flaw in JDBC:
that the existing date/time utilities were not included in the JDBC jar,
causing failures. This PR fixes that.
### Deprecate OptionSet
In a previous set of PRs we've been assembling a sub-operator test
framework. This PR continues that work by deprecating the prior OptionSet
class. (OptionSet was meant to work around limitations in the original
OptionManager, but those issues have since been resolved by all the great work
that Jyothsna and Tim have done. Since OptionSet now no longer serves a useful
purpose, this PR removes it.)
### Deprecate StatsWriter
Similarly, the StatsWriter class added a stats mechanism that can be used
in sub-operator tests. Further experience has shown we can make the full
OperatorStats work in this context, so StatsWriter is deprecated in this PR.
### Fix for Close of Sort Operator
A bit of code was reshuffled to close the operator context in the correct
order in the sort operator in both production and unit tests.
### Swap Containers
Added code to swap buffers between two containers. Builds on code added
earlier to exchange buffers between two vectors. (Used in scan operator, to be
added later.)
### CSV Column Header Error Handling
Changed CSV to use UserException to report errors instead of an ad-hoc
HeaderError exception.
### Test Tools
Add a number of services to the ClusterFixture family of test classes. The
key new feature is the ability to read the results of a multi-batch query as a
sequence of RowSet objects so we can use the row set tools to verify results.
### Union Vector
The union vector is a rather odd duck: it works, but is not widely
supported in Drill. It stores its "type" members in two ways: in an internal
map and in member variables. The member variable format makes it hard to treat
the types generically. This PR keeps the serialized format (storing vectors in
an internal map) but changes the member variable format to use an array of type
vectors rather than individual variables. The result is that generic access
(for the new Union column writer) is much easier.
### Materialized Field in Vectors
Vectors provide a MaterializedField to hold their metadata. The original
idea seems to be that a MaterializedField is immutable. This worked great for
simple scalar types. But, code evolved to change the MaterializedField for maps
(to add map members), for Unions (to add union members) and so on.
For container types (Map, Repeated Map, Union, List, Repeated List), the
"parent" MaterializedField holds onto a copy of the child field. But, some
parents, when they receive new members, throw away their old MaterializedField
and create a new one. The result is that, if we have the structure map(map(a,
b)), when we add b to the inner map, the parent ends up pointing to an older
version of the child schema, not the updated one.
This PR fixes at least some of those issues in the Union vector.
### JUnit Upgrade
Upgrades JUnit from 4.11 to 4.12 to make use of a handy annotation that
disables timeouts when debugging.
### Generic Cleanup
Added comments, fixed indentation and other minor changes.
---
