str_replace() isn't able to replace strings with NAs, since 1.1.0

[filipsch]

In stringr version 1.0.0, the following worked fine:

> str_replace(c(NA, "abc", ""), pattern = "^$", NA)
[1] NA    "abc" NA   

In stringr version 1.1.0, an error is thrown:

> str_replace(c(NA, "abc", ""), pattern = "^$", NA)
Error in if (char == "$") { : missing value where TRUE/FALSE needed

Is this updated behavior intentional?

[hadley]

I don't think that should have ever worked — but it does need a better error message.

[denrou]

Is there a reason for this behavior?

Here's an example of some data I'm working on:

d> glimpse(raw_data$survey)
Observations: 3,497
Variables: 21
$ site_surface              <chr> "No Value", "No Value", "No Value", "No Value", "No Value", "No Value", "No Value", "No Val...
$ location_name             <chr> "Bâtiment AP1", "Site Romans", "No Value", "No Value", "No Value", "No Value", "No Value", ...
$ location_surface          <chr> "No Value", "No Value", "No Value", "No Value", "No Value", "No Value", "No Value", "No Val...
$ stats                     <chr> "Solved", "Solved", "Solved", "Solved", "Solved", "Solved", "Solved", "Solved", "Completed"...
$ value                     <chr> "No Value", "No Value", "No Value", "No Value", "No Value", "No Value", "No Value", "No", "...
...

Something like mutate_all(raw_data$survey, str_replace, pattern = "No Value", replacement = NA) would definitely be helpful in this case.

For now, I'm using this workaround:

  tidy_gsub <- function(x, pattern, replacement, ...) {
    gsub(pattern, replacement, x, ...)
  }

 raw_data$survey %>%
   mutate_all(tidy_gsub, pattern = "No Value", replacement = NA)

[hadley]

You'd be better off with if_else plus str_detect. str_replace() has to replace parts of a string and replacing part of a string with NA doesn't make sense.

[denrou]

I see your point. What do you think of an approach à la tidyr::replace_na:

replace_no_value <- function(data, no_value = "") {
  for (var in names(data)) {
    data[[var]][data[[var]] == no_value] <- NA
  }
  data
}

[hadley]

stringi seems to support it, so I implemented it anyway (you'll need to use NA_character_ though)

[denrou]

Perfect! Thank you.