partial evaluation of geom predicates only goes one level deep

[ateucher]

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

crd_mun <- bcdc_query_geodata("municipalities-legally-defined-administrative-areas-of-bc") %>% 
  select(ADMIN_AREA_ABBREVIATION, ADMIN_AREA_GROUP_NAME) %>% 
  filter(ADMIN_AREA_GROUP_NAME == "Capital Regional District") %>% 
  collect()

# Doesn't work
crd_greenspaces <- bcdc_query_geodata("local-and-regional-greenspaces") %>% 
  select(PARK_NAME, PARK_TYPE, PARK_PRIMARY_USE) %>% 
  filter(BBOX(st_bbox(crd_mun), crs = "EPSG:3005")) %>% #<<
  collect()
#> Error: Cannot embed a data frame in a SQL query.
#> 
#> If you are seeing this error in code that used to work, the most likely
#> cause is a change dbplyr 1.4.0. Previously `df$x` or `df[[y]]` implied
#> that `df` was a local variable, but now you must make that explict with
#> `!!` or `local()`, e.g., `!!df$x` or `local(df[["y"]))

the_box <- st_bbox(crd_mun)
# works
crd_greenspaces <- bcdc_query_geodata("local-and-regional-greenspaces") %>% 
  select(PARK_NAME, PARK_TYPE, PARK_PRIMARY_USE) %>% 
  filter(BBOX(the_box, crs = "EPSG:3005")) %>% #<<
  collect()

^{Created on 2019-11-27 by the reprex package (v0.3.0)}

[ateucher]

The problem seems to be not the nesting, but that all functions are assumed to be remote. From the examples in?dbplyr::partial_eval:

# Functions are always assumed to be remote. Use local to force evaluation in R.
f <- function(x) x + 1
partial_eval(quote(year > f(1980)), vars = vars)
partial_eval(quote(year > local(f(1980))), vars = vars)

So this does work:

crd_greenspaces <- bcdc_query_geodata("local-and-regional-greenspaces") %>% 
  filter(BBOX(local(st_bbox(crd_mun)), crs = "EPSG:3005")) %>% #<<
  collect()

[ateucher]

I don't know if we can insert a local() for any function that's not one of the geom predicates? Or make an informative error message/help section telling users to use local()...

[ateucher]

Also works (using !! instead of local()):

crd_greenspaces <- bcdc_query_geodata("local-and-regional-greenspaces") %>% 
  filter(BBOX(!!st_bbox(crd_mun), crs = "EPSG:3005")) %>% #<<
  collect()

[boshek]

In that sense the error you receive is actually quite good. Other than this part:

Error: Cannot embed a data frame in a SQL query.

and this part:

the most likely cause is a change dbplyr 1.4.0

So tweaking the error message and including an example should be sufficient?

[ateucher]

Yeah, the tricky part is going to be catching those in a general way so that we know when to throw that error...

[boshek]

Can you just catch the dbplyr error and assume that dbplyr will catch all those cases?