Merge pull request #6 from flyconnectome/add-test-coverage

Add tests for previously-uncovered paths (34% -> 66% coverage)

Author: jefferis

.github/workflows/test-coverage.yaml

@@ -37,7 +37,8 @@ jobs:
         shell: Rscript {0}
         run: |
           reticulate::install_miniconda()
-          reticulate::py_install(c("numpy", "pandas"))
+          # pyarrow lets the use_arrow = TRUE conversion path be exercised
+          reticulate::py_install(c("numpy", "pandas", "pyarrow"))
           pak::local_install()
           cat(sprintf("RETICULATE_PYTHON=%s\n", reticulate::py_config()$python),
               file = Sys.getenv("GITHUB_ENV"), append = TRUE)

tests/testthat/test-convert.R

@@ -119,6 +119,63 @@ test_that("pandas2df recovers 64-bit ids and object columns", {
   expect_identical(out$label, c("a", "b"))
 })
 
+test_that("pandas2df flattens object columns and normalises datetimes", {
+  skip_if_no_module("pandas")
+  # An object column of python ints (flattened to an atomic vector), a genuine
+  # list-valued object column (a multi-select, left as a list), an all-None
+  # object column (-> NA), and a datetime column (-> UTC POSIXct).
+  df <- reticulate::py_eval(
+    paste0("__import__('pandas').DataFrame({",
+           "'n': __import__('pandas').Series([1, 2, 3], dtype=object), ",
+           "'tags': __import__('pandas').Series([['AB'], ['CD'], []], dtype=object), ",
+           "'blank': __import__('pandas').Series([None, None, None], dtype=object), ",
+           "'when': __import__('pandas').to_datetime(",
+           "['2020-01-01', '2020-01-02', '2020-01-03'])})"),
+    convert = FALSE)
+  out <- pandas2df(df)
+  # object ints flattened to a plain atomic vector, not a list of length-1s
+  expect_false(is.list(out$n))
+  expect_identical(out$n, c(1L, 2L, 3L))
+  # a genuine list-valued (multi-select) column is left intact as a list
+  expect_true(is.list(out$tags))
+  expect_identical(out$tags[[1]], "AB")
+  expect_length(out$tags[[3]], 0L)
+  # an all-None object column collapses to NA
+  expect_true(all(is.na(out$blank)))
+  # datetimes come back as UTC POSIXct
+  expect_s3_class(out$when, "POSIXct")
+  expect_identical(attr(out$when, "tzone"), "UTC")
+  expect_identical(as.character(as.Date(out$when)),
+                   c("2020-01-01", "2020-01-02", "2020-01-03"))
+})
+
+test_that("pandas2df use_arrow path round-trips via a feather file", {
+  skip_if_no_module("pandas")
+  skip_if_no_module("pyarrow")          # pandas.to_feather needs pyarrow
+  skip_if_not_installed("arrow")        # R arrow::read_feather
+  df <- reticulate::py_eval(
+    "__import__('pandas').DataFrame({'label': ['a', 'b', 'c'], 'n': [1, 2, 3]})",
+    convert = FALSE)
+  out <- pandas2df(df, use_arrow = TRUE)
+  # the arrow path always returns a tibble
+  expect_s3_class(out, "tbl_df")
+  expect_identical(out$label, c("a", "b", "c"))
+  expect_equal(as.integer(out$n), 1:3)
+})
+
+test_that("pandas2df use_arrow handles an empty frame", {
+  skip_if_no_module("pandas")
+  skip_if_not_installed("arrow")
+  skip_if_not_installed("tibble")
+  # the empty-frame branch returns via py_to_r + tibble, without touching arrow
+  df <- reticulate::py_eval(
+    "__import__('pandas').DataFrame({'label': [], 'n': []})", convert = FALSE)
+  out <- pandas2df(df, use_arrow = TRUE)
+  expect_s3_class(out, "tbl_df")
+  expect_identical(nrow(out), 0L)
+  expect_identical(names(out), c("label", "n"))
+})
+
 test_that("pandas2df fast-path preserves column order and nullable ids", {
   skip_if_no_module("pandas")
   ids <- c("720575940621039145", "720575940626877799")

tests/testthat/test-env.R

@@ -0,0 +1,45 @@
+# Pure-R helpers in env.R. The provisioning machinery itself (install_miniconda,
+# conda_install, ...) has real side effects and is not unit-tested here; these
+# cover the option/env plumbing and the non-destructive "print the command"
+# branches of simple_python_base().
+
+test_that("np_condaenv honours the option and defaults to r-reticulate", {
+  withr::local_options(nat.python.condaenv = NULL)
+  expect_identical(np_condaenv(), "r-reticulate")
+  withr::local_options(nat.python.condaenv = "my-env")
+  expect_identical(np_condaenv(), "my-env")
+})
+
+test_that("ownpythonrequested reflects RETICULATE_PYTHON", {
+  withr::local_envvar(RETICULATE_PYTHON = "")
+  expect_false(ownpythonrequested())
+  withr::local_envvar(RETICULATE_PYTHON = "/opt/python/bin/python")
+  expect_true(ownpythonrequested())
+})
+
+test_that("checkownpython aborts for a non-standard Python", {
+  withr::local_envvar(RETICULATE_PYTHON = "")
+  # miniconda = FALSE means the user asked for their own Python
+  expect_error(checkownpython(miniconda = FALSE), "on your own")
+  withr::local_envvar(RETICULATE_PYTHON = "/opt/python/bin/python")
+  expect_error(checkownpython(miniconda = TRUE), "on your own")
+  # standard managed setup: no abort
+  withr::local_envvar(RETICULATE_PYTHON = "")
+  expect_silent(checkownpython(miniconda = TRUE))
+})
+
+test_that("check_reticulate is a no-op when check_python = FALSE", {
+  expect_true(check_reticulate(check_python = FALSE))
+  expect_invisible(check_reticulate(check_python = FALSE))
+})
+
+test_that("simple_python_base blast only prints and deletes nothing", {
+  withr::local_envvar(RETICULATE_PYTHON = "")
+  # the blast branch just prints the (destructive) unlink command; it must
+  # never touch the filesystem itself, and returns invisibly
+  existed <- dir.exists(reticulate::miniconda_path())
+  expect_invisible(res <- simple_python_base("blast", miniconda = TRUE))
+  expect_null(res)
+  # the miniconda directory is left exactly as it was
+  expect_identical(dir.exists(reticulate::miniconda_path()), existed)
+})

tests/testthat/test-int64.R

@@ -24,3 +24,53 @@ test_that("rids2raw honours endianness", {
   expect_false(identical(rids2raw(1:3, endian = "little"),
                          rids2raw(1:3, endian = "big")))
 })
+
+test_that("pyids2bit64 round-trips a numpy int64 array", {
+  skip_if_no_module("numpy")
+  np <- reticulate::import("numpy", convert = FALSE)
+  ids <- c("123", "720575940621039145", "9223372036854775807")
+  arr <- np$array(ids, dtype = "int64")
+  # default: character out
+  expect_identical(pyids2bit64(arr), ids)
+  # as_character = FALSE: an exact integer64 vector
+  i64 <- pyids2bit64(arr, as_character = FALSE)
+  expect_s3_class(i64, "integer64")
+  expect_identical(as.character(i64), ids)
+})
+
+test_that("pyids2bit64 accepts uint64 within signed range and rejects overflow", {
+  skip_if_no_module("numpy")
+  np <- reticulate::import("numpy", convert = FALSE)
+  ok <- np$array(c("123", "9223372036854775807"), dtype = "uint64")
+  expect_identical(pyids2bit64(ok), c("123", "9223372036854775807"))
+  # a uint64 value beyond the signed range cannot be represented as int64
+  ov <- np$array(c("18446744073709551615"), dtype = "uint64")
+  expect_error(pyids2bit64(ov), "int64 overflow")
+  # unsupported dtype is rejected
+  f <- np$array(c("1.5"), dtype = "float64")
+  expect_error(pyids2bit64(f), "dtype=int64 or uint64")
+})
+
+test_that("pyids2bit64 handles an empty array", {
+  skip_if_no_module("numpy")
+  np <- reticulate::import("numpy", convert = FALSE)
+  empty <- np$array(list(), dtype = "int64")
+  expect_identical(pyids2bit64(empty), character())
+  expect_identical(pyids2bit64(empty, as_character = FALSE), bit64::integer64())
+})
+
+test_that("rids2pyint round-trips R ids back through pyids2bit64", {
+  skip_if_no_module("numpy")
+  ids <- c("123", "720575940621039145", "9223372036854775807")
+  # in-memory string path (default for short vectors)
+  arr <- rids2pyint(ids, numpyarray = TRUE)
+  expect_identical(pyids2bit64(arr), ids)
+  # a Python list of ints is the non-numpyarray default
+  lst <- rids2pyint(ids)
+  expect_s3_class(lst, "python.builtin.list")
+  # the file-marshalling path (usefile = TRUE) yields the same ids
+  arr2 <- rids2pyint(ids, numpyarray = TRUE, usefile = TRUE)
+  expect_identical(pyids2bit64(arr2), ids)
+  # a numpy array passed in is returned as-is
+  expect_identical(pyids2bit64(rids2pyint(arr, numpyarray = TRUE)), ids)
+})

tests/testthat/test-time.R

@@ -0,0 +1,20 @@
+# ts2pydatetime: the early-return for an already-converted datetime is pure R;
+# the conversion path needs the Python datetime module.
+
+test_that("ts2pydatetime returns an existing python datetime unchanged", {
+  # a value already carrying the datetime.datetime class is passed straight
+  # back, no Python needed
+  fake <- structure(list(), class = "datetime.datetime")
+  expect_identical(ts2pydatetime(fake), fake)
+})
+
+test_that("ts2pydatetime converts an R time to a UTC python datetime", {
+  skip_if_no_module("datetime")
+  t <- as.POSIXct("2020-01-02 03:04:05", tz = "UTC")
+  py <- ts2pydatetime(t)
+  expect_s3_class(py, "datetime.datetime")
+  # tzinfo is made explicit (UTC), and the instant round-trips exactly
+  expect_true(reticulate::py_to_r(py$tzinfo == reticulate::import("datetime")$timezone$utc))
+  back <- as.numeric(reticulate::py_to_r(py$timestamp()))
+  expect_equal(back, as.numeric(t))
+})

tests/testthat/test-utils.R

@@ -0,0 +1,31 @@
+# Internal utility helpers in utils.R. check_suggested is pure R; py_np needs
+# numpy and is skipped when Python is unavailable.
+
+test_that("check_suggested passes for an installed package", {
+  # reticulate is a hard dependency, so it is always present
+  expect_true(check_suggested("reticulate"))
+  expect_invisible(check_suggested("reticulate"))
+})
+
+test_that("check_suggested errors for a missing package with instructions", {
+  nope <- "a.package.that.does.not.exist"
+  expect_error(check_suggested(nope), "is required")
+  expect_error(check_suggested(nope), "install.packages")
+  # the purpose string is woven into the message when supplied
+  expect_error(check_suggested(nope, "for the widget path"),
+               "for the widget path")
+})
+
+test_that("py_np imports and caches the numpy module", {
+  skip_if_no_module("numpy")
+  # clear any cached handle so we exercise the import branch
+  rm(list = ls(.nat_python_cache), envir = .nat_python_cache)
+  np1 <- py_np()
+  expect_s3_class(np1, "python.builtin.module")
+  # second call is served from the session cache -- same object handle
+  np2 <- py_np()
+  expect_identical(np1, np2)
+  # convert = TRUE is cached under a separate key
+  np3 <- py_np(convert = TRUE)
+  expect_s3_class(np3, "python.builtin.module")
+})