[
https://issues.apache.org/jira/browse/PHOENIX-2199?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15174620#comment-15174620
]
James Taylor commented on PHOENIX-2199:
---------------------------------------
Welcome, @Kaide Mu, and yes, your understanding is correct. The new JavaCC
templates can create the ParseNode constructs we need for the MetaDataClient
method calls. There are no expressions involved in CREATE/DROP SEQUENCE and
CREATE/DROP TABLE, so I'd start with those first. Both DROP VIEW and DROP INDEX
also don't use any expression. The CREATE INDEX call allows an expression to be
indexed and CREATE VIEW allows an expression in the WHERE clause (this is the
only construct we support currently in our CREATE VIEW) syntax. However, even
for those, we can convert from an Expression to a RexNode, so I think we can
likely keep the MetaDataClient and ParseNode APIs as they are for now.
> Support DDL in Phoenix-Calcite integration
> ------------------------------------------
>
> Key: PHOENIX-2199
> URL: https://issues.apache.org/jira/browse/PHOENIX-2199
> Project: Phoenix
> Issue Type: Task
> Reporter: James Taylor
> Labels: calcite, gsoc2016
>
> The existing Phoenix compiler classes are of the form Create<SQL
> type>Compiler. For example CreateTableCompiler handles both VIEW and TABLE
> compilation, CreateSequenceCompiler is for CREATE SEQEUNCE, and
> CreateIndexCompiler is for CREATE INDEX. The compiler class does most of the
> validation and then calls a MetaDataClient method to update the meta data
> repository.
> As a general implementation strategy, we'd transfer over the validation code
> from the compiler methods to new classes plugged into the JavaCC compilation
> of Calcite and then have these new classes call the same, somewhat modified
> MetaDataClient APIs. A slightly more detailed breakdown of the work includes:
> - Creating new JavaCC rules using the same templating technique as we did for
> CREATE VIEW (see https://github.com/apache/phoenix/pull/113/commits), marking
> the parse node produced as DDL.
> - Producing the information required while the Calcite parser is running to
> enable the call to the appropriate MetaDataClient API to update the metadata
> repository.
> - Tweaking the MetaDataClient APIs to not depend on the objects created at
> parse time, but instead pass through the constituent parts. For example,
> instead of passing through a CreateTableStatement object in the
> MetaDataClient.createTable() call, only the necessary member variables would
> be passed through. The idea would be to break the dependency on parse-time
> object after compilation is complete.
> - Changing ParseNode references used in MetaDataClient, as we don't want to
> continue maintaining these. By ParseNode, I mean the parts in the grammar
> that match an expression and produce a ParseNode instance. For DDL, this is
> the CreateIndexStatement object, which stores a list of ParseNodes as the
> indexed expressions. These are currently processed in MetaDataClient, so
> these could be changed to either Expressions or RexNodes.
> - Collecting at parse time the list of UDFs that need to be resolved. I see
> this in CreateIndexStatement, but not CreateTableStatement (which I believe
> is an oversight - probably wouldn't work to reference a UDF in the WHERE
> clause of a view).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
