[
https://issues.apache.org/jira/browse/ARROW-4722?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16891335#comment-16891335
]
Wes McKinney commented on ARROW-4722:
-------------------------------------
I have run into this issue in ARROW-3772 where Parquet decoders currently pass
around bitmaps in pieces (see eg
https://github.com/apache/arrow/blob/master/cpp/src/parquet/encoding.h#L56).
Currently we allow passing null/not-null markers using a bytemap in the builder
classes, but not a bitmap. I propose that we have a simple bitmap object with
structure like:
{code}
class Bitmap {
const uint8_t* data_;
int64_t length_;
int64_t offset_;
};
{code}
I'm not sure how to best handle non-const bitmaps, though. We might have to have
{code}
class MutableBitmap : public Bitmap {
...
};
{code}
A first place where this would be useful to have is as an alternative to any
ArrayBuilder method that currently takes {{valid_bytes}}, like
https://github.com/apache/arrow/blob/master/cpp/src/arrow/array/builder_primitive.h#L119
{{Bitmap}} can have helper methods {{GetReader}} and {{GetWriter}} for
constructing {{BitmapReader}} and {{BitmapWriter}} respectively
> [C++] Implement Bitmap class to modularize handling of bitmaps
> --------------------------------------------------------------
>
> Key: ARROW-4722
> URL: https://issues.apache.org/jira/browse/ARROW-4722
> Project: Apache Arrow
> Issue Type: New Feature
> Components: C++
> Reporter: Benjamin Kietzman
> Priority: Minor
> Fix For: 1.0.0
>
>
> This could be a simple view or it could own a {{shared_ptr<Buffer>}}. In
> either case, it would greatly simplify situations where a {{pointer, offset,
> length}} are currently passed
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
