[ARROW-14434] R crashes when making an empty selection for Datasets with DateTime

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Duplicate
• Affects Version/s: 5.0.0
• Fix Version/s: 6.0.0
• Component/s: R
• Environment: OS = Ubuntu 20.04 ; I use Architect IDE (an ide base on eclipse). But the crash also happens with just R console. R = 3.6.3 ; ; See attached files for session info output and an R crash report.
• Flags: Important
• External issue URL: https://github.com/apache/arrow/issues/29998

Language

R

• Assignee: Unassigned
• Reporter: Lennart Tuijnder
• Created: 22/Oct/21 07:32
• Updated: 11/Jan/23 08:39
• Resolved: 22/Oct/21 12:58

Description

R (3.6.3) crashes when querying a dataset using the "?arrow:: Dataset" functionality when the following conditions are met:

• The dataset to query contains a data-time/time column
• An empty selection is made with dplyr::filter on the Dataset object
• the dplyr::collection method is called. -> (at this point the crash happens)

This crash happens both when the dataset is locally defined or situated on an S3 bucket.

Here is a minimal example to reproduce the bug:

library(dplyr)
library(lubridate)

# If you remove the dataTime column no crashing occurs.
df <- tibble(
	time = seq(5,10,length.out = 10000),
	dateTime = as_datetime(1511870400) + time # dataTime columns causes crash!
)
file <- tempdir()
arrow::write_dataset(df, file)

testdf <- arrow::open_dataset(file) %>%
	# filter(time > 5 & time <6) %>% # When selecting non-empty it does not crash
	filter(time < 5 ) %>% # select empty and it crashes!
	collect()# it crashes when you do collect()

R crashes with the following message:

• • • • caught segfault ****
        address 0x8, cause 'memory not mapped'

I have included in the attachment the full R console output when running the above code.

 

Issue Links

• duplicates: ARROW-13761 [R] arrow::filter() crashes (aborts R session)