Update complex tests for R-devel (4.4.0) (#1883)

* Update complex tests

* NEWS bullet

Author: DavisVaughan

NEWS.md

@@ -1,5 +1,8 @@
 # vctrs (development version)
 
+* Fixed an issue with complex vector tests related to changes in R-devel
+  (#1883).
+
 * Added a class to the `vec_locate_matches()` error that is thrown when an
   overflow would otherwise occur (#1845).
 

tests/testthat/helper-vctrs.R

@@ -56,3 +56,15 @@ expect_equal <- function(object, expected, ...,
 raw2 <- function(...) {
   as.raw(list_unchop(list2(...), ptype = integer()))
 }
+cpl2 <- function(...) {
+  # R 4.4.0 changed `as.complex(NA_real/integer/logical)` so that it always uses
+  # a `0` in the imaginary slot. While this is reasonable, it is annoying for
+  # comparison purposes in tests, where we typically propagate the `NA`. As of
+  # rlang 1.1.1, `cpl()` inherits this behavior change so we have a custom version
+  # here that works the same on all R versions.
+  # https://github.com/wch/r-source/commit/1a2aea9ac3c216fea718f33f712764afc34f6ee8
+  out <- list2(...)
+  out <- as.complex(out)
+  out[is.na(out)] <- complex(real = NA_real_, imaginary = NA_real_)
+  out
+}

tests/testthat/test-slice-assign.R

@@ -681,7 +681,7 @@ test_that("can assign shaped base vectors with compact seqs", {
   expect_identical(vec_assign_seq(mat(lgl(1, 0, 1)), NA, start, size, increasing), mat(lgl(1, NA, NA)))
   expect_identical(vec_assign_seq(mat(int(1, 2, 3)), NA, start, size, increasing), mat(int(1, NA, NA)))
   expect_identical(vec_assign_seq(mat(dbl(1, 2, 3)), NA, start, size, increasing), mat(dbl(1, NA, NA)))
-  expect_identical(vec_assign_seq(mat(cpl(1, 2, 3)), NA, start, size, increasing), mat(cpl(1, NA, NA)))
+  expect_identical(vec_assign_seq(mat(cpl2(1, 2, 3)), NA, start, size, increasing), mat(cpl2(1, NA, NA)))
   expect_identical(vec_assign_seq(mat(chr("1", "2", "3")), NA, start, size, increasing), mat(chr("1", NA, NA)))
   expect_identical(vec_assign_seq(mat(raw2(1, 2, 3)), raw2(1), start, size, increasing), mat(raw2(1, 1, 1)))
   expect_identical(vec_assign_seq(mat(list(1, 2, 3)), NA, start, size, increasing), mat(list(1, NULL, NULL)))
@@ -695,7 +695,7 @@ test_that("can assign shaped base vectors with decreasing compact seqs", {
   expect_identical(vec_assign_seq(mat(lgl(1, 0, 1)), NA, start, size, increasing), mat(lgl(1, NA, NA)))
   expect_identical(vec_assign_seq(mat(int(1, 2, 3)), NA, start, size, increasing), mat(int(1, NA, NA)))
   expect_identical(vec_assign_seq(mat(dbl(1, 2, 3)), NA, start, size, increasing), mat(dbl(1, NA, NA)))
-  expect_identical(vec_assign_seq(mat(cpl(1, 2, 3)), NA, start, size, increasing), mat(cpl(1, NA, NA)))
+  expect_identical(vec_assign_seq(mat(cpl2(1, 2, 3)), NA, start, size, increasing), mat(cpl2(1, NA, NA)))
   expect_identical(vec_assign_seq(mat(chr("1", "2", "3")), NA, start, size, increasing), mat(chr("1", NA, NA)))
   expect_identical(vec_assign_seq(mat(raw2(1, 2, 3)), raw2(1), start, size, increasing), mat(raw2(1, 1, 1)))
   expect_identical(vec_assign_seq(mat(list(1, 2, 3)), NA, start, size, increasing), mat(list(1, NULL, NULL)))

tests/testthat/test-slice.R

@@ -38,7 +38,7 @@ test_that("can subset with missing indices", {
     expect_identical(vec_slice(lgl(1, 0, 1), i), lgl(0, NA))
     expect_identical(vec_slice(int(1, 2, 3), i), int(2, NA))
     expect_identical(vec_slice(dbl(1, 2, 3), i), dbl(2, NA))
-    expect_identical(vec_slice(cpl(1, 2, 3), i), cpl(2, NA))
+    expect_identical(vec_slice(cpl2(1, 2, 3), i), cpl2(2, NA))
     expect_identical(vec_slice(chr("1", "2", "3"), i), c("2", NA))
     expect_identical(vec_slice(raw2(1, 2, 3), i), raw2(2, 0))
     expect_identical(vec_slice(list(1, 2, 3), i), list(2, NULL))

tests/testthat/test-type-bare.R

@@ -243,13 +243,27 @@ test_that("safe casts to complex works", {
 })
 
 test_that("NA casts work as expected", {
+  # This goes through a special path for <unspecified>
   expect_equal(vec_cast(lgl(NA), cpl()), NA_complex_)
-  expect_equal(vec_cast(int(NA), cpl()), NA_complex_)
 
   # TODO: Use our own cast routines here?
-  # `as.complex(NA_real_)` and `Rf_CoerceVector(NA_real_)` coerce to
-  # `complex(real = NA_real_, imaginary = 0)` for some reason, but this may
-  # change in the future https://stat.ethz.ch/pipermail/r-devel/2023-April/082545.html
+  # It isn't great that this logical `NA` cast returns a different `NA`
+  # than the one above with just `lgl(NA)` (which is seen as unspecified). i.e.
+  # check the `Im()` slot between the two in R >=4.4.0. We can fix this with our
+  # own cast routines rather than using `vec_coerce_bare()`.
+  expect_type(vec_cast(lgl(NA, TRUE), cpl()), "complex")
+  expect_identical(is.na(vec_cast(lgl(NA, TRUE), cpl())), c(TRUE, FALSE))
+
+  # TODO: Use our own cast routines here?
+  # `as.complex(NA/NA_real_/NA_integer_)` and `Rf_CoerceVector(NA/NA_real_/NA_integer_)`
+  # have gone back and forth about what they return in the `Im()` slot. In some
+  # R versions they return `0` and in others they return `NA_real_`.
+  # https://stat.ethz.ch/pipermail/r-devel/2023-April/082545.html
+  # https://stat.ethz.ch/pipermail/r-devel/2023-September/082864.html
+  # expect_equal(vec_cast(int(NA), cpl()), NA_complex_)
+  expect_type(vec_cast(int(NA), cpl()), "complex")
+  expect_identical(is.na(vec_cast(int(NA), cpl())), TRUE)
+
   # expect_equal(vec_cast(dbl(NA), cpl()), NA_complex_)
   expect_type(vec_cast(dbl(NA), cpl()), "complex")
   expect_identical(is.na(vec_cast(dbl(NA), cpl())), TRUE)
@@ -263,13 +277,21 @@ test_that("Shaped NA casts work as expected", {
   exp_mat <- mat(NA_complex_)
   to_mat <- matrix(cpl())
 
-  expect_equal(vec_cast(mat(lgl(NA)), to_mat), exp_mat)
-  expect_equal(vec_cast(mat(int(NA)), to_mat), exp_mat)
-
   # TODO: Use our own cast routines here?
-  # `as.complex(NA_real_)` and `Rf_CoerceVector(NA_real_)` coerce to
-  # `complex(real = NA_real_, imaginary = 0)` for some reason, but this may
-  # change in the future https://stat.ethz.ch/pipermail/r-devel/2023-April/082545.html
+  # `as.complex(NA/NA_real_/NA_integer_)` and `Rf_CoerceVector(NA/NA_real_/NA_integer_)`
+  # have gone back and forth about what they return in the `Im()` slot. In some
+  # R versions they return `0` and in others they return `NA_real_`.
+  # https://stat.ethz.ch/pipermail/r-devel/2023-April/082545.html
+  # https://stat.ethz.ch/pipermail/r-devel/2023-September/082864.html
+
+  # expect_equal(vec_cast(mat(lgl(NA)), to_mat), exp_mat)
+  expect_type(vec_cast(mat(lgl(NA)), to_mat), "complex")
+  expect_identical(is.na(vec_cast(mat(lgl(NA)), to_mat)), matrix(TRUE))
+
+  # expect_equal(vec_cast(mat(int(NA)), to_mat), exp_mat)
+  expect_type(vec_cast(mat(int(NA)), to_mat), "complex")
+  expect_identical(is.na(vec_cast(mat(int(NA)), to_mat)), matrix(TRUE))
+
   # expect_equal(vec_cast(mat(dbl(NA)), to_mat), exp_mat)
   expect_type(vec_cast(mat(dbl(NA)), to_mat), "complex")
   expect_identical(is.na(vec_cast(mat(dbl(NA)), to_mat)), matrix(TRUE))