pmap strips Date

[AkhilGNair]

As title - purrr::pmap seems to strip date class.

print_date = function(b) b

b = as.Date("2012-03-22")

purrr::pmap(list(b), print_date)
#> [[1]]
#> [1] 15421

Not sure if this is intended? Caught me out though :)

[david-jankoski]

there were a couple of similar discussions around map and dates around here. this workaround does the job for me.

print_date = function(b) as.Date(b, origin = "1970-01-01")
b = as.Date("2012-03-22")
purrr::pmap(list(b), print_date)

[[1]]
[1] "2012-03-22"

hth.

[AkhilGNair]

Thanks @david-jankoski - That's what I'm currently doing. Just wanted to make sure it was known, I admitted didn't look around to see if anything similar has been asked.

[mungojam]

I hit the same problem with pwalk

[huftis]

The bug is not specific to dates; pmap() removes all classes. Note that the plain map() function does not have this bug. Simple reprex:

library(purrr)
x = "test"
class(x) = "foo"
pmap(list(x), identity)
#> [[1]]
#> [1] "test"
map(list(x), identity)
#> [[1]]
#> [1] "test"
#> attr(,"class")
#> [1] "foo"

And as @mungojam mentioned, pwalk() also has this bug:

> pwalk(list(x), ~ print(class(.)))
# [1] "character" # should have been "foo" ...

[lionel-]

This is more of a known limitation rather than a bug, but we have plans to fix this.

[mungojam]

I found today that flatten does this too and removed my Date class as well as doing the intended un-listing. Perhaps that is intentional?

[lionel-]

It is not intentional we've just been too busy with tidy eval and other projects :(

[mungojam]

No worries, I'm enjoying all your other work :)

[hadley]

The root cause is this behaviour:

x <- as.Date("2012-03-22")
x[[1]]
#> [1] "2012-03-22"
list(x)[[c(1,1)]]
#> [1] 15421

This seems like a base R problem?

[yutannihilation]

This seems like a base R problem?

It seems so to me; [[, or do_subset2(), tries to dispatch the method based only on the first argument: https://github.com/wch/r-source/blob/58aec3ee94f2a0769974520f6cb2d99c49fb829f/src/main/subset.c#L894-L914

If it fails, the default implementation of subset2 will be used for all subsets, but, IIUC, it doesn't care about the classes or attributes: https://github.com/wch/r-source/blob/58aec3ee94f2a0769974520f6cb2d99c49fb829f/src/main/subset.c#L1029-L1056

[czeildi]

Related: I'd love to see a map_date and map_datetime functions. My use case: there are some non-vectorized datetime related functions, e.g. lubridate::with_tz and I need map to create a new column with them in a data frame. I only found a workaround to arrive a plain datetime column.

library(purrr)
library(lubridate)
library(tibble)
library(dplyr)

df <- tibble(
  dttm_orig = ymd_hms(c("2018-04-01 08:00:00", "2017-03-02 03:00:00")),
  tz = c("UTC", "CET"),
  sales_amount = c(34, 56)
)

mutate(df, purchase_time = map2(dttm_orig, tz, ~with_tz(.x, .y)))
#> # A tibble: 2 x 4
#>   dttm_orig           tz    sales_amount purchase_time
#>   <dttm>              <chr>        <dbl> <list>       
#> 1 2018-04-01 08:00:00 UTC           34.0 <dttm [1]>   
#> 2 2017-03-02 03:00:00 CET           56.0 <dttm [1]>

mutate(df, purchase_time = map2_chr(dttm_orig, tz, ~as.character(with_tz(.x, .y))) %>% ymd_hms())
#> # A tibble: 2 x 4
#>   dttm_orig           tz    sales_amount purchase_time      
#>   <dttm>              <chr>        <dbl> <dttm>             
#> 1 2018-04-01 08:00:00 UTC           34.0 2018-04-01 08:00:00
#> 2 2017-03-02 03:00:00 CET           56.0 2017-03-02 04:00:00

Created on 2018-05-01 by the reprex package (v0.2.0).

I realize this might not be top priority however I would love to vote up an issue dedicated to this specifically.

[batpigandme]

however I would love to vote up an issue dedicated to this specifically.

@czeildi, if there isn't one, it sounds like you might want to open this as its own issue 👍

[hadley]

To resolve this issue, we need to make sure to file a bug in R's bugzilla.

[yutannihilation]

Just out of curiosity, why is this not just about replacing .l[[c(j, i)]] with .l[[j]][[i]] here?

purrr/src/map.c

Lines 189 to 192 in d0a8081

	// Construct call like .l[[c(j, i)]]
	SEXP j_ = PROTECT(Rf_ScalarInteger(j + 1));
	SEXP ji_ = PROTECT(Rf_lang3(Rf_install("c"), j_, nj == 1 ? one : i));
	SEXP l_ji = PROTECT(Rf_lang3(R_Bracket2Symbol, l, ji_));

(I'm afraid this is not a bug, since it's reasonable that the method dispatch happens only once for one method.)

[burchill]

It might already be known, but calling as_vector() on a list of dates also strips Date.

[yutannihilation]

@burchill Though it seems similar, I feel it's a separate issue. Maybe you should file a new issue. (BTW, I guess what you need is not as_vector(), but something like flatten_vec(). I expect _vec-family will be implemented in future. e.g. map_vec() on #499)

[mikmart]

Just out of curiosity, why is this not just about replacing .l[[c(j, i)]] with .l[[j]][[i]] here?

@yutannihilation I was curious, too, and tested making that change in this branch. As far as I can tell that should fix this issue; or at least the test suite passes and the original simple example is resolved.

[hadley]

@mikmart oh good idea. Want to do a PR?

[yutannihilation]

Thanks for the PR, @mikmart!