Error due to type conflict in new version of tibble

[jaspercooper]

hi @jessecambon — thanks for this awesome package. It solves a huge problem in my workflow!

I've encountered a hiccup that appears to be due to tibble. The following code ran fine before I updated tibble, but broke afterward:

geo(
  street = "9500 Gilman Dr.,",
  city = "San Diego",
  state = "CA",
  postalcode = "92161",
  method = "cascade"
)

The error reads:

Error: Assigned data `retry_results` must be compatible with existing data.
ℹ Error occurred for column `lat`.
✖ Can't convert from <double> to <logical> due to loss of precision.
* Locations: 1.

The error seems to be coming from here:

tidygeocoder/R/geo_methods.R

Line 70 in d971ebc

combi[na_indices,] <- retry_results

Tibble is not happy that the variable types are being overwritten.

If instead you change line 70 to something more explicit like combi[,c("lat","long")] <- retry_results[,c("lat","long")] everything runs just fine.

Hope this helps, thanks again for the great package!

[jessecambon]

Thanks for reporting this, I was able to replicate the issue. It looks like this error is coming from vctrs:

https://vctrs.r-lib.org/reference/vec_cast.html

I haven't gotten a chance to dig into it yet, but the solution may be to force the lat/long coordinates to be formatted as a double in the extract_results or geo functions.

The Census geocoder is returning 'logical' format for the lat/long columns:

a <- geo(
    street = "9500 Gilman Dr.,",
    city = "San Diego",
    state = "CA",
    postalcode = "92161",
    method = "census"
)

summary(a)
#>    street              city              state            postalcode          lat            long        
#> Length:1           Length:1           Length:1           Length:1           Mode:logical   Mode:logical  
#> Class :character   Class :character   Class :character   Class :character   NA's:1         NA's:1        
#> Mode  :character   Mode  :character   Mode  :character   Mode  :character   

When using method = 'cascade', the numeric lat/long results from the 'osm' geocoder are being put into these 'logical' columns which is what causes the error.

[jaspercooper]

Yeah, that sounds right. It may be as simple as fixing the variable types in the initial_results tibble. I'd check out debugonce(geo_cascade) again though --- it also appears the subsetting by na_indices around line 70 isn't quite doing what it's supposed to (presumably singling out the columns with NAs), as it's defined as a scalar on line 56.

[jessecambon]

@jaspercooper I've got a bug fix for you to try out:

devtools::install_github("jessecambon/tidygeocoder", ref = 'cascade-na-fix')

I changed the the lat/long NA values from NA to as.numeric(NA) in the extract_results() and get_na_results() functions in utils.R. I haven't tested too extensively yet so let me know if you run into any issues. The na_indices variable in geo_method.R is for the row numbers for addresses that were not found by the first method tried.

[jaspercooper]

Thanks, I actually found a workaround in the meantime just by writing my own "cascade" script with geocode_osm and geocode_census. Worked great, those functions are awesome and saved me a lot of time. Thanks for this great package!