translate_sql() fails when the target for translation starts with a dot

[MichaelChirico]

I want to add a translation for .POSIXct() --> TIMESTAMP_FROM_UNIX_SECONDS():

https://cloud.google.com/dataflow/docs/reference/sql/timestamp_functions#timestamp_from_unix_seconds

However, the leading . appears to be throwing everything off. I added a simple placeholder translation in this branch:

https://github.com/MichaelChirico/dbplyr/tree/leading-dot

.POSIXct = sql_prefix("foo", 1)

With that, I get a failed translation:

library(dbplyr)
con=simulate_dbi()
translate_sql(.POSIXct(0), con=con)
# <SQL> .POSIXct(0.0)

(should be foo())

I looked into this a bit and I got lost in the details of the implementation. sql_variant() seems to be doing just fine, as base_scalar does contain the correct object still:

evalq(exists(".POSIXct", base_scalar), asNamespace("dbplyr"))
# [1] TRUE

However, this step apparently fails to find the correct translation:

dbplyr/R/translate-sql.R

Lines 148 to 149 in 860bd6a

	mask <- sql_data_mask(x, variant, con = con, window = window)
	escape(eval_tidy(x, mask), con = con)

I am not familiar enough with the details of tidy eval to know if there's an easy fix.

[hadley]

With zero recollection of the code my first guess this is because we use ls() somewhere.

[MichaelChirico]

I had the same instinct. indeed I came across these, but there must be more (possibly in another package?)

dbplyr/R/translate-sql-helpers.R

Line 67 in 860bd6a

missing <- setdiff(ls(aggregate), ls(window))

[MichaelChirico]

trace() (and devcontainer!) is our friend of course... filed #1530.