readr guesses unexpected column types for values containing a “D”

[klmr]

parse_double apparently interprets D as an alternative to E for scientific exponent notation. For people used to R (unless they also know Fortran), this is quite unexpected, and does not seem to be documented anywhere. Compare to the behaviour of core R:

$ readr::parse_double('12d3')
[1] 12000

$ as.numeric('12d3')
[1] NA
Warning message:
NAs introduced by coercion

$ str2lang('12d3')
Error in str2lang("12d3") : <text>:1:3: unexpected symbol
1: 12d3
      ^

So far, so good. Unfortunately this leads to surprises during automatic column type guessing. For instance:

$ readr::read_csv(I("test\n12d3"))
Rows: 1 Columns: 1
── Column specification ───────────────────────────────────────────────────────────────────
Delimiter: ","
dbl (1): test

ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
# A tibble: 1 × 1
   test
  <dbl>
1 12000

I’d wager that this isn’t the expected or desired behaviour for most uses of ‘readr’. — Is there maybe a way to disable this? Something to the effect of “guess column types, but use a conservative parser for number formats.” Or, alternatively, maybe “guess column types, but do not consider scientific exponent notation.”

[hadley]

Somewhat more minimal reprex:

readr::parse_double('12d3')
#> [1] 12000
as.numeric('12d3')
#> Warning: NAs introduced by coercion
#> [1] NA

^{Created on 2023-07-31 with reprex v2.0.2}

If we fix this, it's possible that it might break existing code, but that seems fairly unlikely.

[hadley]

Looks like there are a few letters where this is a problem:

library(readr)

x <- paste0("12", LETTERS, "3")

df <- data.frame(x, parsed = parse_double(x))
#> Warning: 21 parsing failures.
#> row col               expected actual
#>   1  -- no trailing characters   12A3
#>   2  -- no trailing characters   12B3
#>   3  -- no trailing characters   12C3
#>   7  -- no trailing characters   12G3
#>   8  -- no trailing characters   12H3
#> ... ... ...................... ......
#> See problems(...) for more details.
subset(df, !is.na(parsed))
#>       x parsed
#> 4  12D3  12000
#> 5  12E3  12000
#> 6  12F3  12000
#> 12 12L3  12000
#> 19 12S3  12000

^{Created on 2023-07-31 with reprex v2.0.2}