You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@arrow.apache.org by "Neal Richardson (Jira)" <ji...@apache.org> on 2022/10/12 16:17:00 UTC

[jira] [Commented] (ARROW-18006) [R] integer coercion on parition variables

    [ https://issues.apache.org/jira/browse/ARROW-18006?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17616557#comment-17616557 ] 

Neal Richardson commented on ARROW-18006:
-----------------------------------------

In this case, you may want to specify the schema for the partition columns so that it doesn't try to infer them:

{code}
> open_dataset(f)
FileSystemDataset with 3 Parquet files
x: int32

See $metadata for additional Schema metadata
> open_dataset(f, partitioning = hive_partition(x = utf8()))
FileSystemDataset with 3 Parquet files
x: string

See $metadata for additional Schema metadata
{code}

> [R] integer coercion on parition variables
> ------------------------------------------
>
>                 Key: ARROW-18006
>                 URL: https://issues.apache.org/jira/browse/ARROW-18006
>             Project: Apache Arrow
>          Issue Type: Bug
>          Components: R
>    Affects Versions: 9.0.0
>            Reporter: Carl Boettiger
>            Priority: Major
>
> If we partition on a character-string column that is "integer-like", arrow assumes the resulting partition variable is an integer.  This is particularly surprising because in most other cases, making a column a partition variable coerces it to a character (makes sense, path names are strings), but in this case we have the opposite coercion where a string becomes an integer.  Consider this reprex:
> {code:java}
> library(arrow)f <- tempfile()
> df <- data.frame(x = c("00", "06", "18"))
> df |> write_dataset(f, partitioning = "x")
> open_dataset(f) # x is now an integer{code}
> {code:java}
> ## fails
> open_dataset(f) |> filter(x == "00")
>  {code}
> changing this behavior could break existing code I think, so maybe this shouldn't be considered a bug, but I think it merits discussion. 
> (For instance, we often see large datasets, e.g. from NOAA csv etc, that use zero-padded integers in path names to indicate file-specific metadata.  One side effect of the type-coercion is that it also strips  the zero-padding)



--
This message was sent by Atlassian Jira
(v8.20.10#820010)