Regression: geom_sf() is now too aggressive in removing rows

[clauswilke]

I just encountered this problem where geom_sf() removes polygons that are plotted without outline. This is a regression in current ggplot2 master. I assume it's due to #3491.

cc @yutannihilation

library(ggplot2)
library(sf)
#> Linking to GEOS 3.7.2, GDAL 2.4.2, PROJ 5.2.0

# make outline of the world in robinson coordinates
lats <- c(90:-90, -90:90, 90)
longs <- c(rep(c(180, -180), each = 181), 180)

robin_outline <- 
  list(cbind(longs, lats)) %>%
  st_polygon() %>%
  st_sfc(crs = "+proj=longlat +ellps=WGS84 +datum=WGS84 +no_defs") %>% 
  st_sf() %>%
  st_transform(crs = "+proj=robin +lat_0=0 +lon_0=0 +x0=0 +y0=0")

# plot with line
ggplot(robin_outline) +
  geom_sf(fill = "#56B4E950", color = "black")

# plot without line
ggplot(robin_outline) +
  geom_sf(fill = "#56B4E950", color = NA)
#> Warning: Removed 1 rows containing missing values (geom_sf).

^{Created on 2019-09-22 by the reprex package (v0.3.0)}

[yutannihilation]

Opps..., I think your analysis is correct. Sorry for the regression. Then, we probably need to define a handle_na() that handle NAs differently depending on the type of geometry.

[clauswilke]

It seems to me that the true, real, underlying issue is that grid is overly conservative with checking input and doesn't gracefully degrade with missing or NA input in many cases.

@thomasp85 Any chance that we can get grid to be more tolerant to missing input, so we don't have to constantly code circles around grid's limitations?

Just one of many such examples:

library(grid)

grid.points(numeric(0), numeric(0))
#> Error in unit(x, default.units): 'x' and 'units' must have length > 0

^{Created on 2019-09-22 by the reprex package (v0.3.0)}

[thomasp85]

I won’t rule that out but the call is ultimately Paul’s — I’ll talk with him about it

[yutannihilation]

#3537 is an attempt to fix this. But, I think I don't figure out what to do when the data is of mixed types of geometries. Any ideas...?

(edit: Sorry, I misunderstood about the process of when missing values are removed)

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/