[
https://issues.apache.org/jira/browse/DRILL-3516?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14633745#comment-14633745
]
Jacques Nadeau commented on DRILL-3516:
---------------------------------------
My notes to better provide the conceptual overview:
The key thing to understand is that a UDF is slightly incomplete Java. This is
because Drill actually rips apart the functionality of the UDF and recomposes
it directly within expression evaluation code. This gives a substantial
performance and memory benefit but also creates some challenges. As such,
there are some key rules one should follow:
- Don't use imports
- Both class file and source file have to be on the classpath (yes, Drill uses
the source of the function)
- Any JAR files holding UDF resources must include drill-module.conf marker
file so that Drill knows to include that JAR in consideration for UDF loading
- ValueHolders should be treated as structs. As such, don't call any methods
on ValueHolders
> UDF documentation doesn't get people where they need to be
> ----------------------------------------------------------
>
> Key: DRILL-3516
> URL: https://issues.apache.org/jira/browse/DRILL-3516
> Project: Apache Drill
> Issue Type: Bug
> Reporter: Ted Dunning
>
> The UDF documentation on the web side rooted at
> http://drill.apache.org/docs/develop-custom-functions/ does not describe the
> high level process for how UDF's are used, nor does it describe why simple
> things like toString can't work on *Holder data structures.
> This leads to huge confusion and frustration on the part of potentially
> contributors.
> Here are some pertinent threads:
> http://mail-archives.apache.org/mod_mbox/drill-user/201507.mbox/%3CCACAwhF%3DBxs-bXNdrm0pNJ4e8hZiaueqtZMhJ%3DRiBpf%3Dt%3DzEOWA%40mail.gmail.com%3E
> http://mail-archives.apache.org/mod_mbox/drill-user/201507.mbox/%3CCACAwhFmwWkP6udc05UEGFTzpEsaRvAxSRKW%2B2Mg-ijYX8QoQxQ%40mail.gmail.com%3E
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
