[FLINK-11890] Replace Table API string-based expressions by a Java DSL

• Type: New Feature
• Status: Closed
• Priority: Major
• Resolution: Implemented
• Component/s: Table SQL / API
• Assignee: Dawid Wysakowicz
• Reporter: Timo Walther
• Created: 12/Mar/19 15:58
• Updated: 27/Oct/20 14:47
• Resolved: 27/Oct/20 14:47

Description

Currently, expressions in the Table API can be defined in two ways. Either via the implicit Scala DSL or via custom strings:

Table revenue = orders
  .filter("cCountry === 'FRANCE'")
  .groupBy("cID, cName")
  .select("cID, cName, revenue.sum AS revSum");

// vs.

val revenue = orders
  .filter('cCountry === "FRANCE")
  .groupBy('cID, 'cName)
  .select('cID, 'cName, 'revenue.sum AS 'revSum)

The Java version of the Table API was always treated as a third class citizen.

In the past the string-based expression parser was/is:

• multiple times out of sync with the Scala DSL
• buggy because it was not properly tested
• it blows up the API as every method must accept both representations like select(String) and select(Expression)
• confusing for users because it looks like SQL but is actually only SQL-like
• does not provide JavaDocs or further help in case of parse errors within the IDE, a user can only consult the online documentation

We should discuss alternatives for the string-based expressions and might come up with a DSL-like approach to make the Table API ready for the future.

This issue would be a big design shift for Table API users in Java. We need to discuss if the benefits are worth the change. A first proof-of-concept/design proposal will follow shortly.

Issue Links

• mentioned in: Page Loading...