Use drop=TRUE in rows check?

[MichaelChirico]

DBItest/R/spec-result-roundtrip.R

Line 333 in f19fb4d

eval(bquote(expect_identical(rows[1L, .(i)], .(value_or_testfun))))

We have a backend that always returns tbl_df/tbl/data.frame, which uses drop=FALSE by default, however, that runs into issues here (and perhaps other places?) where it's assumed that [ has drop=TRUE when doing rows[1,1], e.g.

── Failure: DBItest: Result: data_logical ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
rows[1L, 2L] (`actual`) not identical to FALSE (`expected`).

`actual` is an S3 object of class <tbl_df/tbl/data.frame>, a list
`expected` is a logical vector (FALSE)
Backtrace:
    ▆
 1. └─DBItest:::spec_result$data_logical(ctx = ctx, con = global_con)
 2.   └─DBItest:::test_select_with_null(...) at rbuild/R/spec-result-roundtrip.R:29:5
 3.     └─DBItest:::test_select(..., .add_null = "below") at rbuild/R/spec-result-roundtrip.R:264:3
 4.       ├─base::eval(bquote(expect_identical(rows[1L, .(i)], .(value_or_testfun)))) at rbuild/R/spec-result-roundtrip.R:333:7
 5.       │ └─base::eval(bquote(expect_identical(rows[1L, .(i)], .(value_or_testfun))))
 6.       └─testthat::expect_identical(rows[1L, 2L], FALSE)

Is it safe to always do drop=TRUE here, or should we just keep skipping this test under "only data.frame returns are supported"?

I lean towards the former, because check_df() has several checks, none of which requires exact [.data.frame syntax:

DBItest/R/spec-result-roundtrip.R

Line 317 in f19fb4d

rows <- check_df(dbGetQuery(con, query))

DBItest/R/utils.R

Lines 78 to 91 in f19fb4d

	check_df <- function(df) {
	expect_s3_class(df, "data.frame")
	if (length(df) >= 1L) {
	lengths <- unname(lengths(df))
	expect_equal(diff(lengths), rep(0L, length(lengths) - 1L))
	expect_equal(nrow(df), lengths[[1]])
	}
	df_names <- names(df)
	expect_true(all(df_names != ""))
	expect_false(anyNA(df_names))
	df
	}