Inline select() after *_join()

[mgirlich]

Closes #866.

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

lf1 <- lazy_frame(x = 1, a = 1, .name = "lf1")
lf2 <- lazy_frame(x = 1, b = 2, .name = "lf2")

left/right/full/inner_join()

left_join(lf1, lf2, by = "x") %>%
  select(b, x)

before

SELECT `b`, `x`
FROM (
  SELECT `LHS`.`x` AS `x`, `a`, `b`
  FROM `lf1` AS `LHS`
  LEFT JOIN `lf2` AS `RHS`
    ON (`LHS`.`x` = `RHS`.`x`)
) `q01`

after

SELECT `b`, `LHS`.`x` AS `x`
FROM `lf1` AS `LHS`
LEFT JOIN `lf2` AS `RHS`
  ON (`LHS`.`x` = `RHS`.`x`)

semi/anti_join()

semi_join(lf1, lf2, by = "x") %>%
  select(x)

Before

SELECT `x`
FROM (
  SELECT * FROM `lf1` AS `LHS`
  WHERE EXISTS (
    SELECT 1 FROM `lf2` AS `RHS`
    WHERE (`LHS`.`x` = `RHS`.`x`)
  )
) `q01`

After

SELECT `x`
FROM `lf1` AS `LHS`
WHERE EXISTS (
  SELECT 1 FROM `lf2` AS `RHS`
  WHERE (`LHS`.`x` = `RHS`.`x`)
)

^{Created on 2022-05-18 by the reprex package (v2.0.1)}

For semi/anti_join() this only works with projections (i.e. select() and rename()). In the long run one could merge lazy_semi_join_query() and lazy_select_query() as semi_join() and anti_join() are simply translated to WHERE (NOT) EXISTS. But I'd keep them separate for now and explore that at a different time.

[hadley]

LGTM!

I tried a couple of more exotic join edge cases and the results look great 😄

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

lf1 <- lazy_frame(x = 1, y = 1, .name = "lf1")
lf2 <- lazy_frame(x = 1, y = 2, .name = "lf2")

lf1 |> 
  left_join(lf2, by = "x") |> 
  select(x, y = y.y)
#> <SQL>
#> SELECT `LHS`.`x` AS `x`, `RHS`.`y` AS `y`
#> FROM `lf1` AS `LHS`
#> LEFT JOIN `lf2` AS `RHS`
#>   ON (`LHS`.`x` = `RHS`.`x`)

lf1 |> 
  left_join(lf2 |> rename(z = y), by = "x") |> 
  select(z)
#> <SQL>
#> SELECT `RHS`.`y` AS `z`
#> FROM `lf1` AS `LHS`
#> LEFT JOIN `lf2` AS `RHS`
#>   ON (`LHS`.`x` = `RHS`.`x`)

^{Created on 2022-08-03 by the reprex package (v2.0.1)}