Refine select inline criteria to keep arranged computed columns

[ejneer]

Fixes #1437

This will stop arranged computed columns from being inlined away from subqueries. Of course the subquery ORDER BY warning will be thrown, but I think that is expected behavior.

Before:

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

db <- lazy_frame(x = 1, y = 2) 

db %>% 
  mutate(z = 1) %>% 
  arrange(x, z) %>%
  select(x)
#> <SQL>
#> SELECT `x`
#> FROM `df`
#> ORDER BY `x`, `z`

^{Created on 2024-02-07 with reprex v2.0.2}

After:

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

db <- lazy_frame(x = 1, y = 2) 

db %>% 
  mutate(z = 1) %>% 
  arrange(x, z) %>%
  select(x)
#> Warning: ORDER BY is ignored in subqueries without LIMIT
#> ℹ Do you need to move arrange() later in the pipeline or use window_order() instead?
#> <SQL>
#> SELECT `x`
#> FROM (
#>   SELECT `df`.*, 1.0 AS `z`
#>   FROM `df`
#> ) AS `q01`

^{Created on 2024-02-07 with reprex v2.0.2}

[hadley]

Thanks for taking a stab at this!

[hadley]

@mgirlich could you please take a look at this PR too?

[hadley]

I made a few tweaks here to hopefully make the logic a bit clearer.

[hadley]

I'm not sure this is testing the right thing because when I run it, I see:

SELECT `x`
FROM (
  SELECT `df`.*, 2.0 AS `z`
  FROM `df`
) AS `q01`
Warning message:
ORDER BY is ignored in subqueries without LIMIT
ℹ Do you need to move arrange() later in the pipeline or use window_order() instead? 

[hadley]

Or maybe that is the expected behaviour? But definitely worth a snapshot test, I think.

[ejneer]

Yeah that is expected. I've re-added the snapshot I removed in my prior commit.

[hadley]

Thanks for your work on this @ejneer — it's much appreciated!