bad handling of Int64 column of dataframe when reading in R with read_feather

[rezaeir]

I'm new with using the arrow pacakge. my code is as follows:

# Python
import pandas as pd
df = pd.DataFrame(dict(a=[1,2], b=[3,4]))

df.to_feather('test.feather')

then

# R
library('arrow')
df = read_feather('test.feather')
data.matrix(df)

which instead of coercing to double gives me this unusual output

a	 b
4.940656e-324	1.482197e-323
9.881313e-324	1.976263e-323

Which part of my code is wrong or What should I do here to fix this?

[nealrichardson]

This appears to be a feature of how data.matrix() interacts with bit64::integer64 class objects. Here's a reprex without involving arrow:

> df <- data.frame(a=bit64::as.integer64(1:2), b=bit64::as.integer64(3:4))
> df
 a b
1 1 3
2 2 4
> data.matrix(df)
 a b
[1,] 4.940656e-324 1.482197e-323
[2,] 9.881313e-324 1.976263e-323

You could fix this in your example either by providing a schema in Python with int32 types, or by calling as.integer on the columns of your data.frame before calling data.matrix().

One could argue that we should downcast int64 to int32 if there are no out of bounds values since that's what R can natively handle. I made https://issues.apache.org/jira/browse/ARROW-9083 to consider that.