Poor substr() translation

[michal-przydacz]

Hi,
I was trying to create an SQL query which would grab the end of a string. In this example it was supposed to be the last five characters. However, the show_query function shows it returns null rather than calculated integers.
Is that a correct behaviour?

library(RSQLite)
library(DBI)
library(dplyr)
library(dbplyr)

# Create example SQL database and load with example table
conn <- DBI::dbConnect(RSQLite::SQLite(), "")
DBI::dbWriteTable(conn, 
                  "iris", 
                  iris %>% distinct(Species), 
                  overwrite = TRUE)

# Expected behaviour
DBI::dbGetQuery(conn,
                "SELECT `Species`
                        ,LENGTH(`Species`) AS `Length`
                        ,SUBSTR(`Species`, LENGTH(`Species`)-4, LENGTH(`Species`)) AS `Trimmed`
                FROM `iris`")
#>      Species Length Trimmed
#> 1     setosa      6   etosa
#> 2 versicolor     10   color
#> 3  virginica      9   inica

# Actual code example; start and stop substituted with NULL
tbl(conn, DBI::Id(Table = "iris")) %>%
    mutate(Length = length(Species),
           Trimmed = substr(Species, start = length(Species)-4, stop = length(Species))) %T>%
    show_query() %>%
    collect()
#> Warning in substr(Species, start = length(Species) - 4, stop = length(Species)):
#> NAs introduced by coercion
#> Warning in pmax(as.integer(stop) - start + 1L, 0L): NAs introduced by coercion
#> <SQL>
#> SELECT `Species`, length(`Species`) AS `Length`, SUBSTR(`Species`, NULL, NULL) AS `Trimmed`
#> FROM `iris`
#> Warning in substr(Species, start = length(Species) - 4, stop = length(Species)):
#> NAs introduced by coercion

#> Warning in substr(Species, start = length(Species) - 4, stop = length(Species)):
#> NAs introduced by coercion
#> # A tibble: 3 x 3
#>   Species    Length Trimmed
#>   <chr>       <int> <lgl>  
#> 1 setosa          6 NA     
#> 2 versicolor     10 NA     
#> 3 virginica       9 NA

^{Created on 2021-01-21 by the reprex package (v0.3.0)}

I've tried using nchar and len instead of R length but with the same effect (My actual database uses len rather than length).

Thanks

Edit: Provided a better example showing that using native SQL code gets desired result

[hadley]

Minimal reprex:

library(dbplyr)
library(dplyr, warn.conflicts = FALSE)

db <- memdb_frame(x = c("abc", "defsg", "highs"))

db %>% 
  mutate(y = substr(x, 2, nchar(x) - 1)) %>% 
  show_query()
#> Warning in pmax(as.integer(stop) - start + 1L, 0L): NAs introduced by coercion
#> <SQL>
#> SELECT `x`, SUBSTR(`x`, 2, NULL) AS `y`
#> FROM `dbplyr_001`

^{Created on 2021-01-21 by the reprex package (v0.3.0.9001)}

[hadley]

Hmmm, this is going to be hard to support, I think. The easiest work around is to use sub_str()

library(dbplyr)
library(dplyr, warn.conflicts = FALSE)

db <- memdb_frame(x = c("abc", "defsg", "highs"))

db %>% 
  mutate(y = str_sub(x, 2, -2))
#> # Source:   lazy query [?? x 2]
#> # Database: sqlite 3.34.0 [:memory:]
#>   x     y    
#>   <chr> <chr>
#> 1 abc   b    
#> 2 defsg efs  
#> 3 highs igh

^{Created on 2021-01-21 by the reprex package (v0.3.0.9001)}

[michal-przydacz]

Thanks for the reply.
Unfortunately, this work around won't work, as it still misses out the last character.

library(dbplyr)
library(stringr)
translate_sql(str_sub(Test, 1, -2))
#> <SQL> SUBSTR(`Test`, 1, LENGTH(`Test`) - 1)

translate_sql(str_sub(Test, 1, length(Test)))
#> Warning in str_sub(Test, 1, length(Test)): NAs introduced by coercion
#> Error in if (end == -1L) {: missing value where TRUE/FALSE needed

^{Created on 2021-01-22 by the reprex package (v0.3.0)}

Interestingly, str_sub() already uses the translation to length() but doesn't seem to be able to do it when specifically asked

[hadley]

Use -1 if you want to include the last character.

[michal-przydacz]

Yep, that'd do it, thanks.
I think in that case this issue is unnecessary, as str_sub() with negative start/stop options do exactly that.
I pretty much always use tidyverse function but somehow within dbplyr related calls I tend to use base. I will change that now.
Thanks.

[michal-przydacz]

The str_sub() works in a reprex example but not with the SQL database I work with.
The translated substring function in this case is SUBSTRING and str_sub(..., start = -10, end = -1) when passing on attributes skips the last one entirely, causing a missing attribute error.

This reprex illustrates skipping 3rd attribute

library(dbplyr)
library(stringr)
translate_sql(str_sub(Test, -10, -1),
                     con = simulate_mssql())
#> <SQL> SUBSTRING(`Test`, -10)

^{Created on 2021-01-25 by the reprex package (v0.3.0)}

[hadley]

Ah bummer, it seems like SQL server doesn't have an option third argument, unlike pretty much every other database (and, as far as I can tell, the SQL spec).

[hadley]

That said, there's clearly a bug because no database is going to accept a negative start value.

[hadley]

No, that's not right — SUBSTR() does seem accept negative values; but SUBSTRING() does not.

[hadley]

Looks like all the backends that use SUBSTRING currently have an incorrect translation (SQL server, redshift, HANA)