[ https://issues.apache.org/jira/browse/CRUNCH-650?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16083090#comment-16083090 ]
Josh Wills commented on CRUNCH-650: ----------------------------------- Hey [~davw] is super-nice, thanks. My concern with shapeless has always been the extra JAR dependency, esp. since folks use it so heavily downstream of the project, and I wasn't sure how good the backwards compatibility story was there (i.e., I would like to avoid another Guava nightmare.) What do you think? > Support case classes and tuples without reflection or codegen using shapeless > ----------------------------------------------------------------------------- > > Key: CRUNCH-650 > URL: https://issues.apache.org/jira/browse/CRUNCH-650 > Project: Crunch > Issue Type: Improvement > Components: Scrunch > Reporter: David Whiting > Assignee: David Whiting > Attachments: > 0001-Replace-generated-tuple-code-and-reflective-case-cla.patch > > > Scrunch support for case classes is pretty buggy, only supports a subset of > scala versions, and fails at runtime instead of compile time if something > goes wrong. Additionally, the support of tuple types via code generation is > awkward and unnecessary. > This patch should hopefully solve both of these with a shapeless > (https://github.com/milessabin/shapeless) based solution which should be much > simpler and more reliable, by defining a mechanism for deriving PTypeH > instances for HLists where PTypeH instances for each type in the list is > known. > We have used it in a few production cases at SC for a while, and the unit and > integration test suite passes, but I didn't write any new tests specifically > for this purpose. Hopefully they should be covered by the existing tests > using case classes and tuples. -- This message was sent by Atlassian JIRA (v6.4.14#64029)