`sf` objects error on print if `sf` is not loaded

[mikemahoney218]

Apologies if I'm in the wrong spot for this issue!

If we create any sf object with classes "sf", "tbl_df", "tbl", "data.frame" and save it via save:

library(sf)
x <- read_sf(system.file("shape", "nc.shp", package = "sf", mustWork = TRUE))
save(x, file = "x.rda")

And then load in a new R session without loading sf, the geometry column often causes an error when printing:

# Load an sf object without loading sf:
load("x.rda")
# Printing without any processing works, but doesn't print nicely:
x
#> (output snipped)
# Printing after processing fails:
dplyr::mutate(x, y = 1)
#> Error in `vec_size()`:
#> ! `x` must be a vector, not a <sfc_MULTIPOLYGON/sfc> object.
#> Run `rlang::last_error()` to see where the error occurred.
# This is "sticky", as printing now _always_ fails:
x
#> Error in `vec_size()`:
#> ! `x` must be a vector, not a <sfc_MULTIPOLYGON/sfc> object.
#> Run `rlang::last_error()` to see where the error occurred.
# Run any sf function to load the package:
sf::sf_extSoftVersion()
# Printing now works
dplyr::mutate(x, y = 1)
#> Simple feature collection with 100 features and 15 fields
#> (output snipped)

The traceback looks like this:

<error/vctrs_error_scalar_type>
Error in `vec_size()`:
! `x` must be a vector, not a <sfc_MULTIPOLYGON/sfc> object.
---
Backtrace:
     ▆
  1. ├─base `<fn>`(x)
  2. ├─pillar:::print.tbl(x)
  3. │ └─pillar:::print_tbl(...)
  4. │   ├─base::writeLines(...)
  5. │   ├─base::format(...)
  6. │   └─pillar:::format.tbl(...)
  7. │     └─pillar:::format_tbl(...)
  8. │       └─pillar::tbl_format_setup(...)
  9. │         ├─pillar:::tbl_format_setup_dispatch(...)
 10. │         └─pillar:::tbl_format_setup.tbl(...)
 11. │           └─pillar:::df_head(x, n)
 12. │             ├─pillar:::vec_head(as.data.frame(x), n)
 13. │             │ └─vctrs::vec_size(x)
 14. │             ├─base::as.data.frame(x)
 15. │             └─tibble:::as.data.frame.tbl_df(x)
 16. │               ├─base::`[<-`(`*tmp*`, unname, value = `<named list>`)
 17. │               └─tibble:::`[<-.tbl_df`(`*tmp*`, unname, value = `<named list>`)
 18. │                 └─tibble:::tbl_subassign(x, i, j, value, i_arg, j_arg, substitute(value))
 19. │                   └─tibble:::vectbl_recycle_rhs_rows(...)
 20. │                     ├─base::withCallingHandlers(...)
 21. │                     └─vctrs::vec_recycle(value[[j]], nrow)
 22. └─vctrs:::stop_scalar_type(`<fn>`(`<s_MULTIP>`), "x", `<fn>`(vec_size()))
 23.   └─vctrs:::stop_vctrs(...)
 24.     └─rlang::abort(message, class = c(class, "vctrs_error"), ..., call = vctrs_error_call(call))

This is specific to the tibble print method; the data.frame print method doesn't error:

library(sf)
y <- read_sf(system.file("shape", "nc.shp", package = "sf", mustWork = TRUE))
y <- as.data.frame(y)
y <- st_as_sf(y)

save(y, file = "y.rda")

After restarting R:

# Load an sf object without loading sf:
load("y.rda")
# This doesn't error (though it doesn't print nicely either)
dplyr::mutate(y, z = 1)

[krlmlr]

Thanks. It looks like we fail on computing the vec_size() of the sf object if sf is not loaded:

sf <- structure(list(a = 3, g = structure(list(structure(1:2, class = c(
  "XY",
  "POINT", "sfg"
))), class = c("sfc_POINT", "sfc"), precision = 0, bbox = structure(c(
  xmin = 1,
  ymin = 2, xmax = 1, ymax = 2
), class = "bbox"), crs = structure(list(
  input = NA_character_, wkt = NA_character_
), class = "crs"), n_empty = 0L)), row.names = 1L, class = c(
  "sf",
  "data.frame"
), sf_column = "g", agr = structure(c(a = NA_integer_), class = "factor", .Label = c(
  "constant",
  "aggregate", "identity"
)))

vctrs::vec_size(sf)
#> [1] 1
vctrs::vec_size(sf$a)
#> [1] 1
vctrs::vec_size(sf$g)
#> Error in `vctrs::vec_size()`:
#> ! `x` must be a vector, not a <sfc_POINT/sfc> object.

library(sf)
#> Linking to GEOS 3.9.1, GDAL 3.2.3, PROJ 7.2.1; sf_use_s2() is TRUE
vctrs::vec_size(sf$g)
#> [1] 1

^{Created on 2022-06-10 by the reprex package (v2.0.1)}

A fix on the sf end could be to add an explicit "list" class to the sfc class:

sf <- structure(list(a = 3, g = structure(list(structure(1:2, class = c(
  "XY",
  "POINT", "sfg"
))), class = c("sfc_POINT", "sfc"), precision = 0, bbox = structure(c(
  xmin = 1,
  ymin = 2, xmax = 1, ymax = 2
), class = "bbox"), crs = structure(list(
  input = NA_character_, wkt = NA_character_
), class = "crs"), n_empty = 0L)), row.names = 1L, class = c(
  "sf",
  "data.frame"
), sf_column = "g", agr = structure(c(a = NA_integer_), class = "factor", .Label = c(
  "constant",
  "aggregate", "identity"
)))

class(sf$g) <- c(class(sf$g), "list")

vctrs::vec_size(sf$g)
#> [1] 1

^{Created on 2022-06-10 by the reprex package (v2.0.1)}

What is the original use case, or motivation, for printing without loading the package?

[mikemahoney218]

I ran into this originally when adding an sf object as data to a package (in tidymodels/spatialsample#33 ). Because the package didn't import sf but rather referred to functions via ::, this meant that anything that would print the data before processing it -- which we did in documentation and in tests -- would error. The easy fix was to just @import sf, but it took a second to figure that out given the error and traceback surfaced.

[krlmlr]

Thanks. I think this is the right thing to do. Closing, we can't really do much here.

[mikemahoney218]

I did just realize this means you can't print data from sf objects loaded via data without loading the package as well:

data(boston_canopy, package = "spatialsample")
boston_canopy
#> Error in `vec_size()`:
#> ! `x` must be a vector, not a <sfc_MULTIPOLYGON/sfc> object.

^{Created on 2022-06-13 by the reprex package (v2.0.1)}

Still not a hard thing to work around (just load the package), but probably the more common place this occurs.

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue and link to this old issue if necessary.