Extend sql_vector or add possibility to treat over() differently

[tomasgreif]

Some SQL engines (namely Oracle) do not support currently generated SQL for window functions where more partitions or order by expressions are used, for example:

win_test <- data.frame(A = c('a', 'a', 'b', 'b', 'c', 'c'),
                       B = c('d', 'd', 'd', 'e', 'e', 'e'),
                       C = c(1:6)
                       )
copy_to(my_db, win_test, name = 'WIN_TEST')
win_test_ora <- tbl(my_db, from = 'WIN_TEST')

x <- mutate(group_by(win_test_ora, A, B), 
            leadf  = order_by(c(A, C), lead(A, 1, 'x'))

)

This generates SQL:

<SQL>
SELECT "A", "B", "C", "leadf"
FROM (SELECT "A", "B", "C", LEAD("A", 1.0, 'x') OVER (PARTITION BY ("A", "B")  ORDER BY ("A", "C") AS "leadf"
FROM "WIN_TEST") "_W17"

The only issue is that parentheses are used to enclose each set of expressions. This performed by sql_vector function.

[mwillumz]

Greenplum acts the same way. My current approach is to fork dplyr on github and modify the sql_vector() to parens=FALSE for the partition and order vectors. As you mention this seems to be the only issue. Not ideal but I've not been able to find a better temporary fix. Also, Greenplum is built off of Postgres so I'm a bit surprised that the syntax is different. Would postgres work without the parentheses?

mtcars2 <- copy_to(src_postgres(), mtcars)

Test <- group_by(mtcars2,cyl,gear) %>%
  arrange(mpg) %>%
  mutate(low=first(mpg))

Test$query

Test

But if I add parens=FALSE via fixInNamespace(over,'dplyr') the following runs as expected.

mtcars2 <- copy_to(src_postgres(), mtcars)

Test <- group_by(mtcars2,cyl,gear) %>%
  arrange(mpg) %>%
  mutate(low=first(mpg))

Test$query

[piccolbo]

Add HiveQL and SparkSQL to the list of engines who can not handle the extra parentheses.