New function (as.double2), plus documentation and tests.

[knapply]

This should be a drop-in replacement for as.double() for character vectors. Take a peek when you get a chance and see if there's some case or behavior that's still missing.

N <- 1e6
set.seed(8675309)
strings <- c(
  "NaN", "-NaN", "nan", "-nan", "Inf", "-Inf", "inf", "-inf", "infinity", "-infinity", NA_character_,
  paste0(letters, LETTERS, month.abb, month.name),
  as.character(seq.Date(as.Date("1970-01-01"), as.Date("1970-02-01"), by = 1L))
)
input <- sample(c(
  paste0(" \r\n\t\f\v", c(0.0, sqrt(seq(1, N))), " \r\n\t\f\v"),
  rep(strings, 3)
))

library(RcppFastFloat)
suppressWarnings(
microbenchmark::microbenchmark(
  as.double(input),
  as.double2(input),
  check = "equal"
)
)
##> Unit: milliseconds
##>               expr      min       lq     mean   median       uq      max neval
##>   as.double(input) 667.3757 684.0287 720.9842 696.3353 743.7095 968.3394   100
##>  as.double2(input) 130.9257 137.0082 158.7741 141.0736 150.5398 584.5855   100

[eddelbuettel]

I may let this sit and stew til we're through newbies at CRAN but will then get to it presto.

[eddelbuettel]

Don't change it now, but I have a slight preference for the old markdown 6.0.1 which I have in an opt-in directory. It does not do the stooopid dance of forcing a recompilation and rebuild only to deal with Rd creation. And with that I often say no to markdown too. (At least in stull-small projects such as this. Exceptions can be made and are made.)

[eddelbuettel]

I also realize that it is awkward with collaborators who won't have 6.0.1 around. Ah well. RStudio lacks flexibility in invoking alternate converters, sadly.

[knapply]

No worries, just help remind me if I forget/don't get to it before you're looking to ship to CRAN.

[eddelbuettel]

We can probably keep it. I am on the fence on md-in-Rd too. It's on in some projects, I am just not too consisten. No worries right now. This all looks svelte as usual.

[eddelbuettel]

We're on CRAN now so I'll merge this. Looks good, as usual :)