Why does readr::read_csv need timezone data and Olson database?

[ttimbers]

This issue is closed, but I see no resolution: #320

I just ran into a similar situation today when teaching on a server. We were trying to use readr::read_csv to load this file: https://raw.githubusercontent.com/UBC-DSCI/dsci-100/master/materials/worksheet_01/marathon_small.csv

On one server which has tzdata installed, it works fine, on another that does not have this installed the following error message is given:

library(tidyverse)
marathon_small <- read_csv("marathon_small.csv")

Warning message in OlsonNames():
“no Olson database found”
Error: Unknown TZ UTC
Traceback:

1. read_csv("marathon_small.csv")
2. read_delimited(file, tokenizer, col_names = col_names, col_types = col_types, 
 .     locale = locale, skip = skip, comment = comment, n_max = n_max, 
 .     guess_max = guess_max, progress = progress)
3. col_spec_standardise(data, skip = skip, comment = comment, guess_max = guess_max, 
 .     col_names = col_names, col_types = col_types, tokenizer = tokenizer, 
 .     locale = locale)
4. guess_header(ds_header, tokenizer, locale)
5. guess_header_(datasource, tokenizer, locale)
6. default_locale()
7. locale()
8. check_tz(tz)
9. stop("Unknown TZ ", x, call. = FALSE)

This csv file is very simple (see snippet below). Why does readr::read_csv need timezone data and Olson database? This GitHub issue, appears to tracking down readr::read_csv depending on tzdata but why? Naively, it doesn't seem necessary, and read.csv works fine on either server.

.csv snippet:

age,bmi,km5_time_seconds,km10_time_seconds,sex
25.0,21.6221160888672,NA,2798,female
41.0,23.905969619751,1210.0,NA,male
25.0,21.6407279968262,994.0,NA,male
35.0,23.5923233032227,1075.0,2135,male
34.0,22.7064037322998,1186.0,NA,male
45.0,42.0875434875488,3240.0,NA,female
33.0,22.5182952880859,1292.0,NA,male
58.0,25.2340793609619,NA,3420,male

The smallest reproducible example I can get down to is this:

library(tidyverse)
default_locale()

which on the problematic server gives:

Warning message in OlsonNames():
“no Olson database found”
Error: Unknown TZ UTC
Traceback:

1. default_locale()
2. locale()
3. check_tz(tz)
4. stop("Unknown TZ ", x, call. = FALSE)

So the problem likely stems from the read_* functions depending on default_locale() for the locale argument by default. Is this really necessary? Naively, I think not, but I could be very wrong. Is there a way to say locale = None/NA/etc?

This is not just a problem for me, as I have come across others experiencing the same problem...

• Unknown TZ UTC swirldev/swirl#603
• R missing timezone data and Olson database jupyter/docker-stacks#502
• Problem with read_csv - Error #320

[jimhester]

Because readr reads date times using it.

Why doesn't your system have an Olson database?

[ttimbers]

In our case, I think because it wasn't always part of the docker-stacks for jupyter? But that begs the question, is every machine readr is used on expected to have that?

We can fix it on our machines, but I opened this issue because I am not the only one who seems to have come across this problem. Wonder if it would be nice/possible for read_* to not care about time zones if no date times are in the data being loaded? Or tell read_* not to read things as date/times and just as numbers/characters?

[jimhester]

Yes, every machine readr is used on is expected to have a olsen database, the docker containers you are using are too stripped down.

[ttimbers]

Thanks for the prompt response and explanation @jimhester!

[wangshun1121]

In docker container, here is a solution:

# Ubuntu
apt install tzdata
# CentOS
yum install tzdata

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/