[
https://issues.apache.org/jira/browse/FLINK-39632?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ramin Gharib updated FLINK-39632:
---------------------------------
Description:
h2. Background
{\{Table}} already exposes ergonomic shortcuts for the built-in changelog PTFs:
{code:java}
Table result = cdcStream.fromChangelog();
Table appendOnly = updatingTable.toChangelog();
{code}
These are equivalent to calling \{{.process("FROM_CHANGELOG", ...)}} /
\{{.process("TO_CHANGELOG", ...)}}, but read more naturally and don't require
the user to remember the function name as a string.
h2. Problem
When the user wants set semantics (\{{PARTITION BY}}), the equivalent path
today goes through the generic \{{PartitionedTable#process(String, Object...)}}:
{code:java}
Table result = cdcStream
.partitionBy($("id"))
.process("FROM_CHANGELOG"); // string-based, no compile-time check
{code}
This is awkward and inconsistent with the row-semantics API.
h2. Proposal
Add \{{fromChangelog(Expression...)}} and \{{toChangelog(Expression...)}} to
\{{PartitionedTable}} (alongside the existing \{{orderBy}}, \{{process}},
\{{asArgument}}):
{code:java}
Table result = cdcStream
.partitionBy($("id"))
.fromChangelog(); // convenience method, mirrors
Table#fromChangelog
Table appendOnly = updatingTable
.partitionBy($("id"))
.toChangelog();
{code}
These would internally delegate to
\{{process(BuiltInFunctionDefinitions.FROM_CHANGELOG.getName(), arguments)}} /
\{{...TO_CHANGELOG...}}, exactly as the row-semantics \{{Table#fromChangelog}}
/ \{{Table#toChangelog}} already do.
> Add fromChangelog() / toChangelog() convenience methods to PartitionedTable
> ---------------------------------------------------------------------------
>
> Key: FLINK-39632
> URL: https://issues.apache.org/jira/browse/FLINK-39632
> Project: Flink
> Issue Type: Sub-task
> Components: Table SQL / API
> Reporter: Ramin Gharib
> Priority: Major
>
> h2. Background
> {\{Table}} already exposes ergonomic shortcuts for the built-in changelog
> PTFs:
> {code:java}
> Table result = cdcStream.fromChangelog();
> Table appendOnly = updatingTable.toChangelog();
> {code}
> These are equivalent to calling \{{.process("FROM_CHANGELOG", ...)}} /
> \{{.process("TO_CHANGELOG", ...)}}, but read more naturally and don't require
> the user to remember the function name as a string.
> h2. Problem
> When the user wants set semantics (\{{PARTITION BY}}), the equivalent path
> today goes through the generic \{{PartitionedTable#process(String,
> Object...)}}:
> {code:java}
> Table result = cdcStream
> .partitionBy($("id"))
> .process("FROM_CHANGELOG"); // string-based, no compile-time check
> {code}
> This is awkward and inconsistent with the row-semantics API.
> h2. Proposal
> Add \{{fromChangelog(Expression...)}} and \{{toChangelog(Expression...)}} to
> \{{PartitionedTable}} (alongside the existing \{{orderBy}}, \{{process}},
> \{{asArgument}}):
> {code:java}
> Table result = cdcStream
> .partitionBy($("id"))
> .fromChangelog(); // convenience method, mirrors
> Table#fromChangelog
> Table appendOnly = updatingTable
> .partitionBy($("id"))
> .toChangelog();
> {code}
> These would internally delegate to
> \{{process(BuiltInFunctionDefinitions.FROM_CHANGELOG.getName(), arguments)}}
> / \{{...TO_CHANGELOG...}}, exactly as the row-semantics
> \{{Table#fromChangelog}} / \{{Table#toChangelog}} already do.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
