[
https://issues.apache.org/jira/browse/ARROW-10221?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17476930#comment-17476930
]
Dominik Moritz commented on ARROW-10221:
----------------------------------------
I think it's nice to have the fast {{toArray}}. We do have {{toJSON}}, which
always takes the slow path. However, I agree that it could be confusing to
accidentally miss nulls. And always using {{toJSON}} isn't good since it would
always take the slow pass even when we have no nulls.
Btw, I found that {{NaN}} is a valid value in typed arrays even though null is
not.
I think two three best solutions are
* Leave this as they are and ask people to use {{toJSON}} if they want to
guarantee to have nulls. We could add a way for users to get the null mask
(there already is {{isValid(index)}} but you need to ask for each value). Or we
could have a way for people to define null values (e.g. as -1 or NaN)
* Take the slow path when the data type for the vector is nullable.
Unfortunately, a user would have no easy option to take the fast pass anymore.
Maybe we could have an option to force the fast pass?
I'm not happy with either but
> [JS] toArray() method ignores nulls on some types.
> --------------------------------------------------
>
> Key: ARROW-10221
> URL: https://issues.apache.org/jira/browse/ARROW-10221
> Project: Apache Arrow
> Issue Type: Bug
> Components: JavaScript
> Affects Versions: 0.17.1
> Reporter: Ben Schmidt
> Priority: Minor
> Labels: pull-request-available
> Time Spent: 10m
> Remaining Estimate: 0h
>
> The .toArray() javascript method of vectors includes a shortcut to return the
> underlying typed array; but this doesn't respect null values, and so can
> return the wrong number.
>
> ```
> v = arrow.Vector.from(\{values: [1, 2, 3, 4, 5, null, 6],type: new
> arrow.Int32()})
> v.toArray()[5] // Incorrectly returns '0'
> v.get(5) // Correctly returns null
> ```
>
> Solution: Eliminate the fast method, always return Javascript arrays. It
> might be better to keep the old method in cases where there are guaranteed no
> nulls.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
