Support win arm64 in rust source #481

Ilia-Kosenkov · 2025-10-11T21:30:31Z

This is a draft of a PR to add native support for ARM64 on Windows.
It also drops support for RTOOLS40, non-ucrt R and 32 bit architecture (subject to debate).

R/use_rtools.R

Ilia-Kosenkov · 2025-10-12T16:16:51Z

I will try to add proper unit tests later. The only remaining question is should I retire 32bit and pre-ucrt support.

JosiahParry · 2025-10-24T17:39:00Z

Can you elaborate on how this PR is supposed to work? I'm quite confused.

The use_rtools() is only used in rust_source(). Do we need to add other places e.g. in the makevars to have actual arm64 support, or do we already have that? just not for rust_source() and the knitr engine?

JosiahParry · 2025-10-29T16:34:13Z

R/use_rtools.R

+get_r_version <- function() {
+  R.version
+}


Given that this just calls a constant value, can we remove this in favor of R.version so that we don't have to tease out the indirection?

This is a trade off for testability. Depending on static state (R.version) is suboptimal

There is no practical different between get_r_version() and R.version in this use case. The function's sole purpose is to grab R.version but with one level of ambiguity.

Yes, and so that you can unit test it

JosiahParry · 2025-10-29T16:35:19Z

R/use_rtools.R

+  throw_if_no_rtools()
+  throw_if_not_ucrt()


since these functions are only used once, would you mind either defining subroutines in the function or removing the function calls for the body calls instead?

No, this is the sole purpose of this: stop writing notebook style code with functions spanning hundreds of rows, with deep nesting.

I think that would make sense if the function was utilized in more than one place. Since it isn't it has the effect of obfuscating the function definition and code making it more challenging for others to contribute or understand the logic of the function.

R/source.R

JosiahParry · 2025-10-29T16:36:35Z

R/use_rtools.R

+
+is_windows_arm <- function() {
+  proc_arch <- Sys.getenv("PROCESSOR_ARCHITECTURE")
+  r_arch <- get_r_version()[["arch"]]


Suggested change

r_arch <- get_r_version()[["arch"]]

r_arch <- R.version[["arch"]]

Same reason as above

JosiahParry · 2025-10-29T16:37:30Z

R/use_rtools.R

+get_path_to_cargo_folder_arm <- function(rtools_root) {
+  path_to_cargo_folder <- file.path(rtools_root, "clangarm64", "bin")
+  path_to_cargo <- file.path(path_to_cargo_folder, "cargo.exe")
+  if (!file.exists(path_to_cargo)) {
+    cli::cli_abort(
+      c(
+        "{.code rextendr} on ARM Windows requires an ARM-compatible Rust toolchain.",
+        "i" = "Check this instructions to set up {.code cargo} using ARM version of RTools: {.url https://github.com/r-rust/faq?tab=readme-ov-file#does-rust-support-windows-on-arm64-aarch64}." # nolint: line_length_linter
+      ),
+      class = "rextendr_error"
+    )
+  }
+
+  normalizePath(path_to_cargo_folder, mustWork = TRUE)
+}


Can we add internal doc comments here? I don't necessarily follow what this function does 😅

How does this differ from rtools bin path funciton?

Path to rtools cargo installation, different subfolder

Ah, okay! So rtools now distributes cargo??!

No, but you can install cargo msys2 style on Arm machine because the target we need is not yet supported (Tier 2).

JosiahParry · 2025-10-29T16:39:32Z

R/use_rtools.R

+get_rtools_home <- function(rtools_version, is_arm) {
+  env_var <- if (is_arm) {
+    sprintf("RTOOLS%s_AARCH64_HOME", rtools_version)
+  } else {
+    sprintf("RTOOLS%s_HOME", rtools_version)
+  }
+
+  default_path <- if (is_arm) {
+    sprintf("C:\\rtools%s-aarch64", rtools_version)
+  } else {
+    sprintf("C:\\rtools%s", rtools_version)
+  }
+
+  normalizePath(
+    Sys.getenv(env_var, default_path),
+    mustWork = TRUE
+  )
+}


What is the difference between this and pkgbuild::rtools_path()?

It clones old behavior. Pkgbuild is not 100% reliable

JosiahParry · 2025-10-29T16:40:02Z

R/use_rtools.R

+get_rtools_bin_path <- function(rtools_home, is_arm) {
+  # c.f. https://github.com/wch/r-source/blob/f09d3d7fa4af446ad59a375d914a0daf3ffc4372/src/library/profile/Rprofile.windows#L70-L71 # nolint: line_length_linter
+  subdir <- if (is_arm) {
+    c("aarch64-w64-mingw32.static.posix", "usr")
+  } else {
+    c("x86_64-w64-mingw32.static.posix", "usr")
+  }
+
+  normalizePath(file.path(rtools_home, subdir, "bin"), mustWork = TRUE)
+}


What is the difference between this and pkgbuild::rtools_path()? Should we prefer pkgbuild here?

CGMossa · 2025-10-29T17:27:37Z

At its current state, this PR is not worth the feature it offers. There are 400 lines of code, plus the addition of two dependencies, to facilitate a compilation flag. That's just not worth it.

We have to have a discussion about the purpose of rextendr, and how it's written. Rextendr have shockingly few contributors, a code base with more functions than features, and a general stinch of no attention given to the dx nor ux.

Ilia-Kosenkov · 2025-10-29T17:30:34Z

Disagree on the first point, agree on the second. Feel free to switch the maintainer to another willing soul to improve maintenance.

CGMossa · 2025-10-29T17:32:34Z

Thanks for the hard work Ilia. I'll search for another caretaker of your work. And any time you feel like jumping on again, the door is always open.

codecov · 2025-10-29T18:02:24Z

Codecov Report

❌ Patch coverage is 95.12195% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.48%. Comparing base (841b343) to head (a1c372b).

Files with missing lines	Patch %	Lines
R/source.R	0.00%	3 Missing ⚠️
R/use_rtools.R	98.73%	1 Missing ⚠️

Additional details and impacted files

Files with missing lines	Coverage Δ
R/use_rtools.R	`98.73% <98.73%> (ø)`
R/source.R	`81.49% <0.00%> (+6.54%)`	⬆️

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

Ilia-Kosenkov added 3 commits October 11, 2025 23:57

Prototype use_rtools

76e1642

Update RTools usage and target

5b8eaa2

Lint

67891d1

JosiahParry reviewed Oct 12, 2025

View reviewed changes

R/use_rtools.R Outdated Show resolved Hide resolved

Ilia-Kosenkov added 8 commits October 18, 2025 15:53

Initial draft for tests

a88afb4

Checkpoint

7bedc2e

Refactor more

563ac03

Finish tests

9a8eb3a

Lint

3f20e3c

Parameterize test

3f1460d

Rename tests

458e17e

Fix typo

25fcb64

Ilia-Kosenkov added 4 commits October 28, 2025 20:27

Fix use_rtools bug

a651278

Update test

69af463

Merge branch 'main' into feature/Support-win-arm64-in-rust-source

c597089

Update NEWS

a1c372b

JosiahParry requested changes Oct 29, 2025

View reviewed changes

CGMossa closed this Oct 29, 2025

JosiahParry mentioned this pull request Oct 29, 2025

Support ARM on Windows #476

Open

	r_arch <- get_r_version()[["arch"]]
	r_arch <- R.version[["arch"]]

Support win arm64 in rust source #481

Support win arm64 in rust source #481

Conversation

Ilia-Kosenkov commented Oct 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Ilia-Kosenkov commented Oct 12, 2025

Uh oh!

JosiahParry commented Oct 24, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

CGMossa commented Oct 29, 2025

Uh oh!

Ilia-Kosenkov commented Oct 29, 2025

Uh oh!

CGMossa commented Oct 29, 2025

Uh oh!

codecov bot commented Oct 29, 2025

Codecov Report

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Ilia-Kosenkov commented Oct 11, 2025 •

edited

Loading