[
https://issues.apache.org/jira/browse/ARROW-60?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15236631#comment-15236631
]
Micah Kornfield commented on ARROW-60:
--------------------------------------
Hi Kai,
I had a chance to think this through and wanted to know what you thought about
the following: I think the implementation of the builder should actually be
pretty straightforward. You should be able to extend ArrayBuilder
(https://github.com/apache/arrow/blob/master/cpp/src/arrow/builder.h).
You can likely have two constructors:
1. One that takes a list of builders and generated the type for the
ArrayBuilder class on the fly
2. One that takes the type (a struct) and generates the builders on the fly
(https://github.com/apache/arrow/blob/master/cpp/src/arrow/types/construct.h
would be useful for this. I think it might pay to add a version of the method
that takes a unique_ptr as an output parameter to make it easier for this
use-case.)
For additional methods: Add an Append method modeled after
(https://github.com/apache/arrow/blob/master/cpp/src/arrow/types/list.h#L144).
It would be be slightly different since you don't need to keep track of
offsets. It would be nice if appending a null value also appended a null to
each of the child builders, but that would require a new virtual method
"AppendNull" on ArrayBuilder. I think its worth it as it would make client
code a little bit more concise, but I think there might be some debate due to
performance and API consistency.
I don't think a "vectorized" append is needed yet, unless you can think of a
use-case for one.
Have you made any progress? How does this align with your thinking?
> C++: Struct type builder API
> -----------------------------
>
> Key: ARROW-60
> URL: https://issues.apache.org/jira/browse/ARROW-60
> Project: Apache Arrow
> Issue Type: New Feature
> Components: C++
> Reporter: Wes McKinney
> Assignee: Kai Zheng
>
> At the moment, we are able to represent struct metadata but not actually
> build struct-type arrays.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
