parse_number can't handle special space character as grouping_mark locale argument

[ghost]

Description

Using the parse_number function with a locale including special space character (e.g. : non breaking space) as an argument of locale(grouping_mark = '\U202F') don't lead to a properly parsed number

This kind of space is often used in french as a separator (eg: 1 234,00 for the UK version of 1,234.00). An example of csv files containing this kind of character can be found following this link.

Minimal reproducible example

library(readr)
number_space <- "1 234"
parse_number(number_space)
# Get '1', which is expected because no decimal_mark defined
parse_number(number_space, locale = locale(grouping_mark = ' '))
# Get as expected '1234'

# Define a non breaking space
special_space <-  "\U00A0"

number_special_space <- paste0(1, special_space, 234)

number_special_space
# "1 234"
parse_number(number_special_space, locale = locale(grouping_mark = ' '))
# Get '1', which is expected because no decimal_mark well defined
parse_number(number_special_space, locale = locale(grouping_mark = special_space))
# Get 1 although '1234' was expected

What I already tried

R can reconize this character and a possible workaround is to gsub these character before proceding.

special_space <-  "\U00A0"
number_special_space <- paste0(1, special_space, 234)
gsub(pattern = special_space, replacement = '', x = number_special_space)
# Get as expected '1234'
gsub(pattern = ' ', replacement = '', x = number_special_space)
# Get as expected '1 234'

Seems to come from internal method in readr. I tried to deep inside the code but my knowledge is too limited. readr:::parse_vector_ call a function readr_parse_vector_ i can't find in the source code. It's maybe in the CPP part of the code but I have little knowledge on cpp.

Session info

R version 3.4.3 (2017-11-30)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 7 x64 (build 7601) Service Pack 1

Matrix products: default

locale:
[1] LC_COLLATE=French_France.1252  LC_CTYPE=French_France.1252   
[3] LC_MONETARY=French_France.1252 LC_NUMERIC=C                  
[5] LC_TIME=French_France.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] Unicode_10.0.0-1 magrittr_1.5     lubridate_1.7.2  bindrcpp_0.2    
[5] dplyr_0.7.4      readr_1.1.1     

loaded via a namespace (and not attached):
 [1] Rcpp_0.12.15     utf8_1.1.3       crayon_1.3.4     assertthat_0.2.0
 [5] R6_2.2.2         pillar_1.1.0     stringi_1.1.6    cli_1.0.0       
 [9] rlang_0.1.6      rstudioapi_0.7   tools_3.4.3      stringr_1.2.0   
[13] glue_1.2.0       hms_0.4.1        yaml_2.1.16      compiler_3.4.3  
[17] pkgconfig_2.0.1  bindr_0.1        tibble_1.4.2  

[cderv]

@jomuller another possible workaround waiting for a fix is

• parsing your columns as character (setting col_character()for these columns in cols argument),
• use deprecated function tidyr::extract_numeric() in a mutate_at call for these columns as this function still works, even with \U00A0. parse_number, more powerful, is replacing tidyr::extract_numeric and a depreciation message says that. Know that tidyr::extract_numeric worked only with . as decimal

Minimal reprex:

(num <- paste(1, 234.5, sep = "\U00A0"))
#> [1] "1 234.5"
library(readr)
parse_number(num, locale = locale(grouping_mark = "\U00A0"))
#> [1] 1
tidyr::extract_numeric(num)
#> extract_numeric() is deprecated: please use readr::parse_number() instead
#> [1] 1234.5

Created on 2018-02-17 by the reprex package (v0.2.0).

[ghost]

@cderv Thank you for this workaround. Currently I'm using simply a regex to remove any non numeric character gsub(pattern = '[^0-9\\.]', '', x, perl = TRUE) which is very close to extract_numeric definition as.numeric(gsub("[^0-9.-]+", "", as.character(x))). Maybe not elegant but does the job. Using deprecated function could be dangerous for the futur of the code I'm writing.

[pachevalier]

I've got two wrapper function in the tricky package. parse_French_number() use parse_number() and unfrench_formatting() use the gsub approach.

[jimhester]

Closed by tidyverse/vroom@959b4b7, this should now be supported.