tidyverse / glue Goto Github PK

First of all, thanks for this great package! I'm migrating some of my internal R pkg functions from sprintf, but found some scoping issues and decided to ask for some help here. I'm pretty sure it's not glue related, rather just PEBKAC on my end, but cannot figure out what's the problem.

Example code:

lapply(1:2, function(x) glue::glue('{x * 2}'))
#> [[1]]
#> 2
#> 
#> [[2]]
#> 4

Trying to use this with a helper function:

f <- function(msg) {
    glue::glue(msg)
}
lapply(1:2, function(x) f('{x * 2}'))
#> Error in eval(x, envir = envir): object 'x' not found

Then trying to specify the parent frame:

f <- function(msg) {
    glue::glue(msg, .envir = parent.frame())
}
lapply(1:2, function(x) f('{x * 2}'))
#> Error in eval(x, envir = envir): object 'msg' not found

So now it's evaluated in the parent frame, but looking for msg instead of x defined withing msg.

I tried a couple of other things eg

f <- function(msg) {
    eval.parent(glue::glue(msg))
}
lapply(1:2, function(x) f('{x * 2}'))
#> Error in eval(x, envir = envir): object 'x' not found

But did not figure this out. Any suggestions?

So you can do:

collapse(1:4, ",", last = ", and ")
#> 1, 2, 3, and 4

unglue() so you could see the generalized form of whatever specific string that was made with glue().

Simple form would be something like:

library(glue)

unglue <- function(x) {
    stopifnot(inherits(x, "glue"))
    attr(x, "glue_string")
}

glue_string <- "echo '{saywhat}'"
x <- glue(glue_string, saywhat = "hello world")
attr(x, "glue_string") <- glue_string
str(x)
#> Class 'glue'  atomic [1:1] echo 'hello world'
#>   ..- attr(*, "glue_string")= chr "echo '{saywhat}'"

x
#> echo 'hello world'
unglue(x)
#> [1] "echo '{saywhat}'"

Ambitious form would also extract out the value of the inputs

unglue_ambitious(x)
#> [1] "echo '{saywhat}'"
#>  ..- attr(*, "saywhat")= chr "hello world"

Sometimes you have the values that you want to interpolate in a list (example: the params object in an Rmarkdown file), and you want to interpolate using these values. Currently, you must manually convert this to an environment in order to do so using glue, but it would be nice if the function would do it for you.

Example:

# using named args
glue("Hello {firstname} {lastname}", firstname="John", lastname="Doe")


person <- c(firstname="John", lastname="Doe")
# Trying to pass a named vector as .envir doesn't workglue("Hello {firstname} {lastname}", .envir=person)
# Must do this instead
glue("Hello {firstname} {lastname}", .envir=as.environment(as.list(person)))

The .envir arg should probably satisfy rlang::is_dictionary if it isn't an environment.

E.g something like:

f('my name is {name}, but do not \\{messwiththis\\}')

or (better since less typing and also concordant with what Python does):

f('my name is {name}, but do not {{messwiththis}}')

wherein the {{ would get replaced by {.

Happy to help with PR if you want help.

Great feature! I was wondering if it should happen before the command substitution. Because right now I get:

l <- list(a = list(a1 = 1, a2 = 2, a3 = 3), b = "foo")
cat(yaml::as.yaml(l))
#> a:
#>   a1: 1.0
#>   a2: 2.0
#>   a3: 3.0
#> b: foo
glue("
  # Indented a bit in the code, but not in the output
  { yaml::as.yaml(l) }
")
#> # Indented a bit in the code, but not in the output
#> a:
#> a1: 1.0
#> a2: 2.0
#> a3: 3.0
#> b: foo

Would you be willing to bump past the CRAN version? I'm using things that were added since then and want to capture that in DESCRIPTION. I already have glue in Remotes but feel like I should also state that version must be greater than 1.0.0.

install.packages("devtools")
trying URL 'https://mirrors.ebi.ac.uk/CRAN/bin/windows/contrib/3.4/devtools_1.13.2.zip'
Content type 'application/zip' length 442517 bytes (432 KB)
downloaded 432 KB

package 'devtools' successfully unpacked and MD5 sums checked

The downloaded binary packages are in
C:\Users\edguevara\AppData\Local\Temp\Rtmp4gSnoj\downloaded_packages

library(tidyverse)
Error: package or namespace load failed for 'tidyverse' in loadNamespace(j <- i[[1L]], c(lib.loc, .libPaths()), versionCh
eck = vI[[j]]):
there is no package called 'glue'
devtools::install_github("tidyverse/glue")
Downloading GitHub repo tidyverse/glue@master
from URL https://api.github.com/repos/tidyverse/glue/zipball/master
Error: running command '"C:/PROGRA1/R/R-341.0/bin/x64/R" --no-site-file --no-environ --no-save --no-restore --quiet CMD
config CC' had status 1

It would be simpler to only accept delimiters of length 1.

It feels like collapse() would be more consistent with glue() (and more convenient) if it returned character(0) if the input has length zero.

library(glue)
glue("{x}", x = "hi")
#> hi
glue("{x}", x = character(0))
#> character(0)
collapse(x = c("hi", "bye"), last = " and ")
#> [1] "hi and bye"
collapse(x = c(NULL, "bye"), last = " and ")
#> [1] "bye"
collapse(x = c("hi", NULL), last = " and ")
#> [1] "hi"
collapse(x = NULL, last = " and ")
#> [1] ""
collapse(x = character(), last = " and ")
#> [1] ""

Set of helpers

Hi, thanks for providing a great package.
Do you have any plan in mind to release this fstrings library on CRAN?
I'm thinking about importing your library in my library.

When any expression (but the last) is NULL then glue returns an empty character vector. In case the last expression is NULL it ignores the last expressions and returns the result of the other expressions.

library(glue)

empty <- c()
glue(empty, 1)
glue(1, empty)
#> 1
glue(1, 2, empty)
#> 12
glue(1, empty, 0)

glue("{empty}", 1)
glue(1, "{empty}")
glue(1, "{empty}", 0)

A version of glue() that automatically quotes results, useful for constructing SQL queries.

I'm trying to use glue to make bulleted lists for messages. I'm a bit puzzled. Maybe this is #47?

I don't understand why I lose the two spaces before the asterisk, as it's not leading whitepace of the first line.

glue::glue("The stuff before the bullet list
              * one bullet")
#> The stuff before the bullet list
#> * one bullet

Adding a new line at the start helps, but I'm not sure why, because now the bulleted line is the 3rd (and last) line.

glue::glue("
           The stuff before the bullet list
             * one bullet")
#> The stuff before the bullet list
#>   * one bullet

Adding a new line to the end doesn't help, but I though it might, because it makes my bulleted line NOT the last line.

glue::glue("The stuff before the bullet list
              * one bullet
           ")
#> The stuff before the bullet list
#> * one bullet

I would like to do

iris %>% mutate(glued_stuff = glue("This {Species} has a petal length of {Petal.Length}"))

instead of

iris %>% mutate(glued_stuff = glue_data(., "This {Species} has a petal length of {Petal.Length}"))

What do you think about this?

I apologize in advance if this is me not understanding how to use these tools appropriately. But the intention is to have a nicer replacement to paste0("sth_", quo_name(expr) because when it is more complicated than just a prefix or suffix paste/0 is hella ugly and to read.

setup

library(rlang)
library(glue)

glue + rlang = 😞

glue_quo <- function(expr) {
    expr <- enquo(expr)
    glue("expr name is {expr}", quo_name(expr))
}
glue_quo(expr = 3)
#> expr name is ~3
#> expr name is 33

session info

devtools::session_info()
#> Session info -------------------------------------------------------------
#>  setting  value                       
#>  version  R version 3.4.0 (2017-04-21)
#>  system   x86_64, darwin15.6.0        
#>  ui       X11                         
#>  language (EN)                        
#>  collate  en_CA.UTF-8                 
#>  tz       America/Vancouver           
#>  date     2017-05-20
#> Packages -----------------------------------------------------------------
#>  package   * version    date       source                          
#>  backports   1.0.5      2017-01-18 CRAN (R 3.4.0)                  
#>  base      * 3.4.0      2017-04-21 local                           
#>  compiler    3.4.0      2017-04-21 local                           
#>  datasets  * 3.4.0      2017-04-21 local                           
#>  devtools    1.13.1     2017-05-13 CRAN (R 3.4.0)                  
#>  digest      0.6.12     2017-01-27 CRAN (R 3.4.0)                  
#>  evaluate    0.10       2016-10-11 CRAN (R 3.4.0)                  
#>  glue      * 1.0.0      2017-05-21 Github (tidyverse/glue@41a5cff) 
#>  graphics  * 3.4.0      2017-04-21 local                           
#>  grDevices * 3.4.0      2017-04-21 local                           
#>  htmltools   0.3.6      2017-04-28 CRAN (R 3.4.0)                  
#>  knitr       1.15.20    2017-05-02 Github (yihui/knitr@f3a490b)    
#>  magrittr    1.5        2014-11-22 CRAN (R 3.4.0)                  
#>  memoise     1.1.0      2017-04-21 CRAN (R 3.4.0)                  
#>  methods   * 3.4.0      2017-04-21 local                           
#>  Rcpp        0.12.10    2017-03-19 CRAN (R 3.4.0)                  
#>  rlang     * 0.1.1.9000 2017-05-21 Github (tidyverse/rlang@7c2f7e8)
#>  rmarkdown   1.5        2017-04-26 CRAN (R 3.4.0)                  
#>  rprojroot   1.2        2017-01-16 CRAN (R 3.4.0)                  
#>  stats     * 3.4.0      2017-04-21 local                           
#>  stringi     1.1.5      2017-04-07 CRAN (R 3.4.0)                  
#>  stringr     1.2.0      2017-02-18 CRAN (R 3.4.0)                  
#>  tools       3.4.0      2017-04-21 local                           
#>  utils     * 3.4.0      2017-04-21 local                           
#>  withr       1.0.2      2016-06-20 CRAN (R 3.4.0)                  
#>  yaml        2.1.14     2016-11-12 CRAN (R 3.4.0)

glue::glue("{1:3}")
#> 1
#> 2
#> 3
glue::glue(1:3)
#> Error: All unnamed arguments must be length 1

Can the two expressions be made equivalent? This will simplify the transition from paste0().

{ } would make it hard to work with some languages such as TeX.

Can we have lst() or tibble() semantics?

glue::glue("{b}", a = 1, b = a)
#> Error in eval(expr, envir, enclos): object 'a' not found

Expected result: "1"

When the .x argument of glue_data() is an environment, the evaluation of ... takes place in .x and not in .envir, contrary to the documented role of .envir. In particular, if ... is not in .x, then glue_data() will fail.

Is this the intended behavior?

Example:

env <- new.env(parent = baseenv())
env$a <- "This is 'a'"

text <- "{a}"

glue_data(.x = env, text)
## Error in eval(x, envir = data, enclos = envir) : object 'text' not found

glue_data() has not evaluated text in .envir = parent.frame().

Naively, I would expect .x to be used solely for object look-up, i.e., as if .x were coerced to a list:

glue_data(.x = as.list(env), text)
## This is 'a'

I think it is good practice.

Glue is not working properly with UTF-8 encoded strings.

> x = "pão"
> x
[1] "pão"
> Encoding(x)
[1] "latin1"
> glue("{x}")
pão
> x = enc2utf8("pão")
> x
[1] "pão"
> Encoding(x)
[1] "UTF-8"
> glue("{x}")
pão

Ping me once you have the logo and I'll give you the details (which we should write up somewhere)

Not sure of the API, maybe an R6 class would be the easiest. This would allow users to add custom syntax for #39, or a syntax for automatically collapsing results, or #53.

Are there any plans of adding formatters?

To clarify, by “formatter” I mean (1) an interface and (2) a glue syntax that advanced formatting of glue tokens. Here’s an example:

glue('π = {pi}')
glue('π to two decimal places = {format(pi, digits = 3)}')

With formatters, the second line could be for instance written as

glue('π to two decimal places = {pi:0.2}')

Many libraries offer similar capabilities, for instance composite formatting in .NET.

Ideally this would be extensible, allowing users to register their own formatters. For instance the following is thinkable:

arrgh_glue = function (str, options) {
    gsub('r', 'R', str)
}

# optionally, though ideally not necessary:
#   register_glue_formatter(arrgh = arrgh_glue)

name = 'Konrad'
glue('Hello, {name:arrgh}!')
# Hello, KonRad!

The .NET documentation has some more examples that show the scope of what could be possible.

e.g.

f(
  'My name is {name},',
  ' my age next year is {age + 1},',
  ' my anniversary is {format(anniversary, "%A, %B %d, %Y")}.',
  name = "Fred",
  age = 50,
  anniversary = as.Date("1991-10-12")
)

https://github.com/joshuaulrich/rchk-docker should be useful

Running the example for glue_sql in the README throws the error:

Error in glue_sql("\n SELECT {var}\n FROM {tbl}\n WHERE {tbl}.sepal_length > {num}\n AND {tbl}.species = {val}\n ", :
could not find function "glue_sql"

I need to run glue::glue_sql for it to work.

Aside from the change in class, glue() acts as the identity function on strings without expressions, except the empty string: glue("") gives character(0).

Is this exception intended?

A few years ago I asked this question on SO

http://stackoverflow.com/questions/16595621/error-safe-templating-with-brew-whisker

The problem is summarised as follows: given 1) a template string, 2) a named list of values, insert those values in the template (with evaluation of R code), but return an error if one or more named values are missing. The solutions with brew or whisker were not particularly elegant.

I think glue provides a good alternative, but it would be nice to have a helper function that automatically enables checking of absent values in the environment (and get their name in the error message).

library(glue)
data <- list( name = "Chris", value= 124)
do.call(glue, list('Hello {name} You have just won ${value}!', .envir=as.environment(data)))
# Hello Chris You have just won $124!
do.call(glue, list('Hello {name} You have just won ${value}!', .envir=as.environment(data[-1])))
# Error in eval(expr, envir, enclos) : object 'name' not found

Inputs are not vectorized

I'm puzzled by this statement in the main help file. The example mtcars %>% glue_data("{rownames(.)} has {hp} hp") seems to immediately go counter to this (not to mention lots of usage I've made). Can this be reworded so the meaning is more clear?

Seems glue corrupt the input string with space or tabs at begin.

I tried with CRAN and git version of the package.

devtools::install_github("tidyverse/glue")
cyr <- 2017
period <- 2
rows_limit <- 1000

x <- glue::glue('SELECT MIN_MEAN_QUANTITY_OUT,
                        MIN_MEAN_QUANTITY_IN,
                        MAX_MEAN_PRICE_OUT,
                        MAX_MEAN_PRICE_IN
                   FROM LOAD_BOUNDS_BY_YM_NROWS 
                  WHERE YEAR = {cyr}
                    AND MONTH = {period - 1}
                    AND MIN_NUM_OF_ROWS_PQ = {rows_limit}')
x

Output:

SELECT MIN_MEAN_QUANTITY_OUT,
MIN_MEAN_QUANTITY_IN,
MAX_MEAN_PRICE_OUT,
MAX_MEAN_PRICE_IN
LOAD_BOUNDS_BY_YM_NROWS 
YEAR = 2017
MONTH = 1
MIN_NUM_OF_ROWS_PQ = 1000

FROM WHERE, AND are missing in the result.

Session info:

R version 3.4.0 (2017-04-21)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Arch Linux

Matrix products: default
BLAS: /usr/lib/R/lib/libRblas.so
LAPACK: /usr/lib/R/lib/libRlapack.so

locale:
 [1] LC_CTYPE=ru_RU.UTF-8       LC_NUMERIC=C               LC_TIME=ru_RU.UTF-8        LC_COLLATE=C              
 [5] LC_MONETARY=ru_RU.UTF-8    LC_MESSAGES=ru_RU.UTF-8    LC_PAPER=ru_RU.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C             LC_MEASUREMENT=ru_RU.UTF-8 LC_IDENTIFICATION=C       

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

loaded via a namespace (and not attached):
[1] compiler_3.4.0  tools_3.4.0     parallel_3.4.0  glue_1.1.1.9000

paste0() and sprintf() treat NA_character_ as "NA":

sv = c(809, NA_character_, "help")
paste0(sv, ".io")
#> [1] "809.io"  "NA.io"   "help.io"
sprintf("%s.io", sv)
#> [1] "809.io"  "NA.io"   "help.io"

Not sure if it's intended, but glue::glue() also fails to suppress NA:

glue::glue("{sv}.io")
#> 809.io
#> NA.io
#> help.io

stringr::str_c performs better:

stringr::str_c(sv, ".io")
#> [1] "809.io"  NA        "help.io"

x <- character()
glue::glue("{x}")
#> Error: Variables must be length 1 or 1

I think this should return an empty character vector.

> f('my name is {name}, but do not \\{messwiththis\\}')
Error in fstring_(x, function(x) paste(collapse = sep, fun(eval(parse(text = x),  : 
  <text>:1:13: unexpected input
1: messwiththis\
                ^

Seems to be an fstrings thing since the following parses fine:

> 'my name is {name}, but do not \\{messwiththis\\}'
[1] "my name is {name}, but do not \\{messwiththis\\}"

Also happy to help if you want help.

I have CI set up on gitlab for my package builds. When I started using glue, I get the following segfaults (I am posting the output of all the jobs):

Running with gitlab-ci-multi-runner 1.11.1 (a67a225)
  on dev (9d172ed6)
Using Shell executor...
Running on AnalyticsDevVM...
Fetching changes...
Removing TTdashboard.Rcheck/
Removing TTdashboard_0.2.39.tar.gz
HEAD is now at 538f5eb Merge branch '73-sac-tooltip-missing-total' into 'dev'
Checking out 538f5ebf as dev...
Skipping Git submodules setup
$ awk '{sub(/quick/, "all"); print}' tests/testthat.R > tests/tmp.R
$ mv tests/tmp.R tests/testthat.R
$ R -e "cov<-covr::package_coverage(type='all');print(cov);covr::report(cov,file='index.html',browse=FALSE)"

R version 3.3.2 (2016-10-31) -- "Sincere Pumpkin Patch"
Copyright (C) 2016 The R Foundation for Statistical Computing
Platform: x86_64-pc-linux-gnu (64-bit)

R is free software and comes with ABSOLUTELY NO WARRANTY.
You are welcome to redistribute it under certain conditions.
Type 'license()' or 'licence()' for distribution details.

  Natural language support but running in an English locale

R is a collaborative project with many contributors.
Type 'contributors()' for more information and
'citation()' on how to cite R or R packages in publications.

Type 'demo()' for some demos, 'help()' for on-line help, or
'help.start()' for an HTML browser interface to help.
Type 'q()' to quit R.

> cov<-covr::package_coverage(type='all');print(cov);covr::report(cov,file='index.html',browse=FALSE)
Error: Failure in `/tmp/RtmpJdZkqe/R_LIBS3d672089871e/TTdashboard/TTdashboard-tests/testthat.Rout.fail`
> library(testthat)
> library(TTdashboard)
> 
> test_check("TTdashboard", filter = "all")

 *** caught segfault ***
address 0x30, cause 'memory not mapped'

Traceback:
 1: .Call(to_impl, unnamed_args, function(expr) as.character(eval2(parse(text = expr),     envir = env, data = .x)))
 2: glue_data(NULL, ..., .sep = .sep, .envir = .envir)
 3: glue::glue("SELECT ScenarioId, {flexcol} CountryId, FactorId, YearId, DataValue\n        FROM {table}\n        USE INDEX ({xbindex})\n        WHERE {fixedcol} = {fixed} AND\n        {specific_col} = {specific_value} AND\n        ScenarioOriginId = {scenario_origin} AND\n        ScenarioId = {globals$scenarioId} AND\n        {flexcol} In ({flex_cntrs})",     flex_cntrs = collapse(flex, ","))
 4: eval(expr, envir, enclos)
 5: eval(lhs, parent, parent)
 6: glue::glue("SELECT ScenarioId, {flexcol} CountryId, FactorId, YearId, DataValue\n        FROM
Execution halted
ERROR: Job failed: exit status 1

This happens from time to time, usually the CI job completes without errors on retry. Here is my devtools:session_info() output:

> devtools::session_info()
Session info -------------------------------------------------------------------
 setting  value                       
 version  R version 3.3.2 (2016-10-31)
 system   x86_64, linux-gnu           
 ui       X11                         
 language (EN)                        
 collate  en_US.UTF-8                 
 tz       <NA>                        
 date     2017-04-25                  

Packages -----------------------------------------------------------------------
 package  * version    date       source                         
 devtools   1.12.0     2016-06-24 CRAN (R 3.3.0)                 
 digest     0.6.10     2016-08-02 cran (@0.6.10)                 
 glue     * 0.0.0.9000 2017-03-29 Github (tidyverse/glue@099b283)
 memoise    1.0.0      2016-01-29 CRAN (R 3.3.0)                 
 withr      1.0.2      2016-06-20 CRAN (R 3.3.0)        

Any ideas on how to debug further?

.na = "NA" by default. If NULL, then 'NA's are propagated (as they currently are in devel).

Would it make sense for

glue::glue("foo bar \
            baz")

to be equivalent to

glue::glue("foo bar baz")

so long messages can be broken down in several lines in the code?

Thanks for this great package. glue made my life easier with template strings.

But, one thing I feel a bit frustrated is that some S3 generic methods cannot process the glue object because they don't know the fact glue is almost character.

generate_msg <- function(x) {
  l <- list(msg = glue::glue('Praise the {x}!'))
  jsonlite::toJSON(l)
}

generate_msg("V8")
#> Error: No method asJSON S3 class: glue

class(glue::glue('Praise the {x}!', x = 'V8'))
#> [1] "glue"

I know just using unclass() or as.character will save me, but is it possible to let glue class inherit from character class? Or, are there some reason that we have to restrict its class glue only?

generate_msg <- function(x) {
  msg <- `class<-`(glue::glue('Praise the {x}!'), c('glue', 'character'))
  l <- list(msg = msg)
  jsonlite::toJSON(l)
}

generate_msg("V8")
#> {"msg":["Praise the V8!"]} 

With option to truncate (and insert ...)

f$"my name is {name}"

could work. Just class f and add a $ method like gsubfn does. Could also provide a different fun to avoid issues (e.g. F).

Saves a keystroke and also looks a tiny bit more like the python syntax.

This one is obviously low priority, but kind of cute, and different to the one I tweeted about.

Perhaps an installation command is looking for a previous version of R?

> devtools::install_github('tidyverse/glue')
Downloading GitHub repo tidyverse/glue@master
from URL https://api.github.com/repos/tidyverse/glue/zipball/master
Error: running command '"C:/PROGRA~1/R/R-33~1.3/bin/x64/R" --no-site-file --no-environ --no-save --no-restore --quiet CMD config CC' had status 127
> sessionInfo()
R version 3.3.3 (2017-03-06)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 14393)

locale:
[1] LC_COLLATE=English_United States.1252
[2] LC_CTYPE=English_United States.1252
[3] LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C
[5] LC_TIME=English_United States.1252

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

loaded via a namespace (and not attached):
[1] httr_1.2.1 R6_2.2.0 tools_3.3.3 withr_1.0.2
[5] curl_2.4 memoise_1.0.0 git2r_0.18.0 digest_0.6.12
[9] devtools_1.12.0

It should trim including the last line.

t <- glue:::trim
  t("
      Hello,
      World.
    ")
# should be the following (based on the indention of the last line)
# "  Hello,\n  World."
# Actually is
#> "Hello,\nWorld."

It appears that NA_character_ values are (erroneously?) special when using the new .na=NULL option in v1.2.0.

## FYI: dbconn is a RPostgres DBI connection
> dbconn <- DBI::dbConnect(RPostgres::Postgres(), host = myhost, dbname = mydb, user = me)

## a bunch of NAs of different type:
> l <- list(
    lgl = NA
   ,chr = NA_character_
   ,dbl = NA_real_
   ,int = NA_integer_
)

## first without the default .na = "NA" option:
> map_chr(l, function(x) glue_sql("foo = {x}", .con = dbconn))

          lgl          chr          dbl          int
   "foo = NA" "foo = 'NA'"   "foo = NA"   "foo = NA"

## now with .na = NULL, which should return _just_ <NA> for all four inputs:
> map_chr(l, function(x) glue_sql("foo = {x}", .con = dbconn, .na = NULL))

          lgl          chr          dbl          int
           NA "foo = 'NA'"           NA           NA

This appears to be specific to glue_sql, glue itself works as expected for all NA types.
Is this related to glue_sql or perhaps the underlying DBI calls? (FWIW I've tried with both RPostgres and RPostgreSQL and found identical results.)