R dplyr programatically identify column

For some objects an attribute identifies a special column, for example the geometry column in an sf object. For conducting some calculations in dplyr it would be good to easily identify these columns. I'm searching for a way to create a function that helps identifying this column. In the example below I can make a function that identifies this column but I still need to use the rlang splice operator (!!!).

require(sf)
require(dplyr)
n<-4
df = st_as_sf(data.frame(x = 1:n, y = 1:n, cat=gl(2,2)), coords = 1:2, crs = 3857) %>% group_by(cat)
# this is the example I start from however the geometry column is not guaranteed to have that name
df %>% mutate(d=st_distance(geometry, geometry[row_number()==1]))
#> Simple feature collection with 4 features and 2 fields
#> Geometry type: POINT
#> Dimension:     XY
#> Bounding box:  xmin: 1 ymin: 1 xmax: 4 ymax: 4
#> Projected CRS: WGS 84 / Pseudo-Mercator
#> # A tibble: 4 × 3
#> # Groups:   cat [2]
#>   cat      geometry d[,1]
#> * <fct> <POINT [m]>   [m]
#> 1 1           (1 1)  0   
#> 2 1           (2 2)  1.41
#> 3 2           (3 3)  0   
#> 4 2           (4 4)  1.41
# this works, however the code does not get easier to read
df %>% mutate(d=st_distance(!!!syms(attr(., "sf_column")), (!!!syms(attr(., "sf_column")))[row_number()==1]))
#> Simple feature collection with 4 features and 2 fields
#> ...
#> 4 2           (4 4)  1.41
# this works and is already better:
geometry_name<-function(x) syms(attr(x, 'sf_column'))
df %>% mutate(d=st_distance(!!!geometry_name(.), (!!!geometry_name(.))[row_number()==1]))
#> Simple feature collection with 4 features and 2 fields
#> ...  
#> 4 2           (4 4)  1.41

Ideally I would like to find a function that makes the following code work as this would be easiest for users:

df %>% mutate(d=st_distance(geometry_name(), geometry_name()[row_number()==1]))

Solution

Calling this kind of function without arguments requires that you assume symbols are present in the calling frame (in this case the . placeholder and the .data pronoun), so it won't work well outside of dplyr verbs, but if that suits your workflow, then you can do:

geometry_name <- function() {
  .data <- eval(quote(.data), parent.frame())
  nms <- names(eval(quote(.), parent.frame()))
  geo <- which(sapply(nms, function(x) inherits(.data[[x]], 'sfc')))
  if(length(geo) == 0) {
    stop('No geometry column detected')
  }
  if(length(geo) > 1) {
    warning('More than one geometry column. Only the first will be used.')
    geo <- geo[1]
  }
  .data[[nms[geo]]]
}

Using your example, this allows you to use your specified syntax:

df %>% 
  mutate(d = st_distance(geometry_name(), geometry_name()[row_number()==1]))
#> Simple feature collection with 4 features and 2 fields
#> Geometry type: POINT
#> Dimension:     XY
#> Bounding box:  xmin: 1 ymin: 1 xmax: 4 ymax: 4
#> Projected CRS: WGS 84 / Pseudo-Mercator
#> # A tibble: 4 x 3
#> # Groups:   cat [2]
#>   cat      geometry d[,1]
#> * <fct> <POINT [m]>   [m]
#> 1 1           (1 1)  0   
#> 2 1           (2 2)  1.41
#> 3 2           (3 3)  0   
#> 4 2           (4 4)  1.41

You could potentially make the function a bit more useful by allowing it to take a data argument, which if missing runs the above code (after checking for the presence of . and .data), but otherwise just finding and returning the sf column from data. This would allow use outside of dplyr verbs, but preserve the desired behaviour inside dplyr.

For example:

geometry_name <- function(data) {
  if(missing(data)) {
    .data <- tryCatch( { 
      eval(quote(.data), parent.frame())
    }, error = function(e){ 
      stop("Argument 'data' missing, with no default")
    })
    plchlder <- tryCatch({
      eval(quote(.), parent.frame())
    }, error = function(e) {
      stop("geometry_name can only be used without a 'data' argument ",
           "inside dplyr verbs")
    })
    nms <- names(plchlder)
    geo <- which(sapply(nms, function(x) inherits(.data[[x]], 'sfc')))
    if(length(geo) == 0) {
      stop('No geometry column detected')
    }
    if(length(geo) > 1) {
      warning('More than one geometry column. Only the first will be used.')
      geo <- geo[1]
    }
    return(.data[[nms[geo]]])
  }
  
  geo <- which(sapply(data, function(x) inherits(x, 'sfc')))
  if(length(geo) == 0) stop('No geometry column detected')
  if(length(geo) > 1) {
    warning('More than one geometry column. Only the first will be used.')
    geo <- geo[1]
  }
  return(data[[geo]])
}

Which gives the following behaviour

geometry_name(df)
#> [1] "geometry"

geometry_name()
#> Error in value[[3L]](cond) : 
#>   geometry_name can only be used without a 'data' argument inside 
#>   dplyr verbs

df %>% 
  mutate(d = st_distance(geometry_name(), geometry_name()[row_number()==1]))
#> Simple feature collection with 4 features and 2 fields
#> Geometry type: POINT
#> Dimension:     XY
#> Bounding box:  xmin: 1 ymin: 1 xmax: 4 ymax: 4
#> Projected CRS: WGS 84 / Pseudo-Mercator
#> # A tibble: 4 x 3
#> # Groups:   cat [2]
#>   cat      geometry d[,1]
#> * <fct> <POINT [m]>   [m]
#> 1 1           (1 1)  0   
#> 2 1           (2 2)  1.41
#> 3 2           (3 3)  0   
#> 4 2           (4 4)  1.41