pretty_rel_path() fails on Windows with R-devel (R 4.3)

[yutannihilation]

This is reproducible on my local.

https://github.com/extendr/extendr/actions/runs/4210717913/jobs/7308589655#step:18:123

   ══ Failed tests ════════════════════════════════════════════════════════════════
  ── Failure ('test-helper-functions.R:32'): `pretty_rel_path()` works ───────────
  pretty_rel_path(...) (`actual`) not equal to normalizePath(...) (`expected`).

  actual vs expected
  - "R/extendr-wrappers.R"
  + "C:/Users/runneradmin/AppData/Local/Temp/RtmpQ9SQyF/working_dir/RtmpcB6a2L/file6f469214d6a/testpkg/R/extendr-wrappers.R"

  [ FAIL 1 | WARN 0 | SKIP 1 | PASS 95 ]

[Ilia-Kosenkov]

I am looking into this as well right now. I wonder what has changed.

[yutannihilation]

It seems the difference is that this line doesn't fail on R-devel

https://github.com/extendr/rextendr/blob/d22584825802c928d8dc1294ea9ec85e2e353a64/R/track_rust_source.R#L39

in case of this

https://github.com/extendr/rextendr/blob/d22584825802c928d8dc1294ea9ec85e2e353a64/tests/testthat/test-helper-functions.R#L33-L36

[Ilia-Kosenkov]

Strange, I cannot reproduce it locally with devtools::test(), nor with rcmdcheck::rcmdcheck()

[Ilia-Kosenkov]

Ah wait, found one https://github.com/extendr/rextendr/blob/d22584825802c928d8dc1294ea9ec85e2e353a64/tests/testthat/test-helper-functions.R#L22-L28

>  expect_equal(
+     pretty_rel_path(
+       file.path(pkg_root, "R", "extendr-wrappers.R"),
+       file.path(pkg_root, "src", "rust", "src", "lib.rs")
+     ),
+     "R/extendr-wrappers.R"
+   )
Error: pretty_rel_path(...) (`actual`) not equal to "R/extendr-wrappers.R" (`expected`).

`actual`:   "E:/.../rextendr/R/extendr-wrappers.R"
`expected`: "R/extendr-wrappers.R"
>

[yutannihilation]

Hmm, it's different one, but probably the same cause?

[Ilia-Kosenkov]

I am baffled. So this fragment https://github.com/extendr/rextendr/blob/d22584825802c928d8dc1294ea9ec85e2e353a64/tests/testthat/test-helper-functions.R#L32-L41 tests that if we search for a file starting outside of the package directory, we fail and return the full path. It works as expected everywhere (including locally in R-devel), but on CI it for some reason... succeeds and returns relative to the project root path.

[yutannihilation]

including locally

No, it fails on my local.

It seems R-devel recently made several changes to remove the limitation on the path length on Windows, so it might be that it no longer uses 8.3 short name in some cases and it mattered. Just a wild guess.

[Ilia-Kosenkov]

I am not sure, since we use normalizePath() which should expand that. It seems to be Windows-specific. If you cannot pinpoint the issue (and it is not an issue per se, since we expect pretty_rel_path() to fail yet it succeeds), and since I am unable to reproduce it, should we skip this test for now and revisit it later?

[yutannihilation]

The minimal reprex.

R 4.2.2

f <- function() {
  pkg_root <- withr::local_tempdir()
  usethis::create_package(pkg_root)
  setwd(pkg_root)

  path <- file.path(pkg_root, "..")
  rprojroot::find_package_root_file(path = path)
}

f()
#> ✔ Setting active project to 'C:/Users/Yutani/AppData/Local/Temp/RtmpEljiyX/file330c1c726ce4'
#> ✔ Creating 'R/'
#> ✔ Writing 'DESCRIPTION'
#> Package: file330c1c726ce4
#> Title: What the Package Does (One Line, Title Case)
#> Version: 0.0.0.9000
#> Authors@R (parsed):
#>     * First Last <first.last@example.com> [aut, cre] (YOUR-ORCID-ID)
#> Description: What the package does (one paragraph).
#> License: `use_mit_license()`, `use_gpl3_license()` or friends to pick a
#>     license
#> Encoding: UTF-8
#> Roxygen: list(markdown = TRUE)
#> RoxygenNote: 7.2.3
#> ✔ Writing 'NAMESPACE'
#> ✔ Setting active project to '<no active project>'
#> Error: No root directory found in C:/Users/Yutani/AppData/Local/Temp/RtmpEljiyX or its parent directories. Root criterion: contains a file "DESCRIPTION" with contents matching "^Package: "

^{Created on 2023-02-21 with reprex v2.0.2}

R 4.3.0

f <- function() {
  pkg_root <- withr::local_tempdir()
  usethis::create_package(pkg_root)
  setwd(pkg_root)

  path <- file.path(pkg_root, "..")
  rprojroot::find_package_root_file(path = path)
}

f()
#> ✔ Setting active project to 'C:/Users/Yutani/AppData/Local/Temp/Rtmp2DYIJ0/file54246b1212a7'
#> ✔ Creating 'R/'
#> ✔ Writing 'DESCRIPTION'
#> Package: file54246b1212a7
#> Title: What the Package Does (One Line, Title Case)
#> Version: 0.0.0.9000
#> Authors@R (parsed):
#>     * First Last <first.last@example.com> [aut, cre] (YOUR-ORCID-ID)
#> Description: What the package does (one paragraph).
#> License: `use_mit_license()`, `use_gpl3_license()` or friends to pick a
#>     license
#> Encoding: UTF-8
#> Roxygen: list(markdown = TRUE)
#> RoxygenNote: 7.2.3
#> ✔ Writing 'NAMESPACE'
#> ✔ Setting active project to '<no active project>'
#> [1] "."

^{Created on 2023-02-21 with reprex v2.0.2}

[yutannihilation]

should we skip this test for now and revisit it later?

I'm fine with either, but I hope we can help upstream (R itself or rprojroot) to figure out what the problem is.

[Ilia-Kosenkov]

Soo, now I can reproduce it locally. The issue was that I had a very old R-devel version, still 4.3 but from last year. I pulled the most recent one, and now observe the same behaviour as on CI. Will try to debug locally.

[Ilia-Kosenkov]

Here is a reprex for you, this reliably fails on release but works (evaluating path to ".") on devel.

pkg_path <- tempfile()
tryCatch(
    {
        pkg <- usethis::create_package(pkg_path)
        setwd(pkg)
        path_outside <- file.path(pkg_path, "..")
        projroot <- rprojroot::find_package_root_file(path = path_outside)
        print(projroot)
    },
    finally = unlink(pkg_path)
)
sessionInfo()

[yutannihilation]

Fixed by https://github.com/wch/r-source/commit/d444c07b7fe765ce65224faa2e85f0819a189e1a