Inline select() in *_join()

[mgirlich]

Closes #867.

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

lf <- lazy_frame(x1 = 10, a = 1, y = 3, .name = "lf1")
lf2 <- lazy_frame(x2 = 10, b = 2, z = 4, .name = "lf2")

left/right/full/inner_join()

left_join(
  lf %>% select(a2 = a, x = x1),
  lf2 %>% select(x = x2, b),
  by = "x"
)

Before

SELECT `a2`, `LHS`.`x` AS `x`, `b`
FROM (
  SELECT `a` AS `a2`, `x1` AS `x`
  FROM `lf1`
) `LHS`
LEFT JOIN (
  SELECT `x2` AS `x`, `b`
  FROM `lf2`
) `RHS`
  ON (`LHS`.`x` = `RHS`.`x`)

After

SELECT `a` AS `a2`, `x1` AS `x`, `b`
FROM `lf1` AS `LHS`
LEFT JOIN `lf2` AS `RHS`
  ON (`LHS`.`x1` = `RHS`.`x2`)

semi/anti_join()

semi_join(
  lf %>% select(a2 = a, x = x1),
  lf2 %>% select(x = x2, b),
  by = "x"
)

Before

SELECT * FROM (
  SELECT `a` AS `a2`, `x1` AS `x`
  FROM `lf1`
) `LHS`
WHERE EXISTS (
  SELECT 1 FROM (
  SELECT `x2` AS `x`, `b`
  FROM `lf2`
) `RHS`
  WHERE (`LHS`.`x` = `RHS`.`x`)
)

After

SELECT `a` AS `a2`, `x1` AS `x`
FROM `lf1` AS `LHS`
WHERE EXISTS (
  SELECT 1 FROM (
  SELECT `x2` AS `x`, `b`
  FROM `lf2`
) `RHS`
  WHERE (`LHS`.`x1` = `RHS`.`x`)
)

The semi/anti_join() could be improved further but I'm not sure it's worth it (mostly interesting when the join columns have different names and one uses rename() instead of using a named by).

[mgirlich]

add_group_by() adds the field group_vars to any lazy_query object. Similarly for order_vars and frame. So it makes sense to be able to explicitly set them when creating these objects.

[mgirlich]

@hadley Would be great if you could give feedback

• whether you see problems with inlining
• you see edge cases that need to be tested

[hadley]

Your logic makes sense to me, and I can see any obvious holes in it.

Is it worth including a test for some of the suffix cases? I'm not sure how much that overlaps with this code.

[hadley]

I definitely like the FROM on its own line.

[mgirlich]

Is it worth including a test for some of the suffix cases?

It should not but rather be safe than sorry. So I added tests for suffix cases and named by.