alternate openmp plugin changes by kevinushey · Pull Request #1414 · RcppCore/Rcpp

kevinushey · 2025-12-02T00:10:02Z

Pull Request Template for Rcpp

A somewhat tidier (IMHO) alternative to the approach in #1409. It also avoids changing the definition of the plugin on non-macOS systems.

Checklist

Code compiles correctly
R CMD check still passes all tests
Preferably, new tests were added which fail without the change
Document the changes by file in ChangeLog

kevinushey · 2025-12-02T00:12:00Z

R/Attributes.R

+    # get the C++ compiler from R
+    r <- file.path(R.home("bin"), "R")
+    cxx <- tryCatch(
+        system2(r, c("CMD", "config", "CXX"), stdout = TRUE),


Strictly speaking, this might not be correct if an alternate plugin causes a different variant of the C++ compiler to be used (e.g. cpp11 will use whatever compiler is registered for CXX11, which in theory might not match CXX). Is this worth worrying about?

Oh dear. The baker's dozen of CXX11 to CXX26. Good question. On the other hand if we cannot trust R to tell it about its CXX, who can we trust?

FWIW for me R CMD config CXX and R CMD config CXX17 return the same. [ Checks. ] But that is due to my personal .R/Makevars.

Yeah, it'd be odd to have one compiler for CXX11, and another for CXX14 -- doubly so on macOS. But lots of people will configure CXX without thinking to configure all of the other versioned CXX variants...

But I now remembered we can get closer using R CMD SHLIB, which lets R choose the appropriate C++ compiler based on whatever environment variables / user Makevars configuration is set, so maybe that's more appropriate. I just made a change to that effect.

Oof. This PR was supposed to be simple!

Nothing in life ever is provided you look closely enough....

kevinushey · 2025-12-02T00:13:39Z

R/Attributes.R

+        prefix <- if (machine == "arm64") "/opt/homebrew" else "/usr/local"
+
+        # build flags
+        cxxflags <- sprintf("-I%s/opt/libomp/include -fopenmp", prefix)


I confirmed that the version of LLVM clang provided in Homebrew accepts -fopenmp as-is, but I'm adding the library paths for libomp just in case.

eddelbuettel

Looks good to me!

eddelbuettel · 2025-12-02T01:19:27Z

@coatless could you chime in during the next day or two?

coatless

This hits all the main points. The only note I have is with the R CMD SHLIB seeming to be more complex and slightly more brittle; but, I'm sure @kevinushey has a good reason for it.

coatless · 2025-12-03T06:09:34Z

R/Attributes.R

-.plugins[["openmp"]] <- function() {
-    list(env = list(PKG_CXXFLAGS="-fopenmp",
-                    PKG_LIBS="-fopenmp"))
+.plugins[["openmp"]] <- if (Sys.info()[["sysname"]] == "Darwin") {


I really like how the detection simplifies the function definition on load.

coatless · 2025-12-03T06:19:37Z

R/Attributes.R

+    # extract the compiler invocation from the shlib output
+    # use some heuristics here...
+    index <- grep("make would use", output)
+    compile <- output[[index + 1L]]
+
+    # use everything up to the first include flag, which is normally
+    # the R headers from CPPFLAGS
+    idx <- regexpr(" -I", compile, fixed = TRUE)
+    cxx <- substring(compile, 1L, idx - 1L)


So, this would trigger:

R CMD SHLIB --dry-run $TMPDIR/openmp-detect-.cpp

Giving:

make cmd is make -f '/Library/Frameworks/R.framework/Resources/etc/Makeconf' -f '/Library/Frameworks/R.framework/Resources/share/make/shlib.mk' -f '/Users/ronin/.R/Makevars' SHLIB_LDFLAGS='$(SHLIB_CXXLDFLAGS)' SHLIB_LD='$(SHLIB_CXXLD)' SHLIB='/var/folders/b0/vt_1hj2d6yd8myx9lwh81pww0000gn/T//Rtmpcxzs34/openmp-detect-137df4ba71a0a.so' OBJECTS='/var/folders/b0/vt_1hj2d6yd8myx9lwh81pww0000gn/T//Rtmpcxzs34/openmp-detect-137df4ba71a0a.o' make would use clang++ -arch arm64 -std=gnu++17 -I"/Library/Frameworks/R.framework/Resources/include" -DNDEBUG -I/opt/R/arm64/include -I/opt/R/arm64/include -fPIC -falign-functions=64 -Wall -g -O2 -c /var/folders/b0/vt_1hj2d6yd8myx9lwh81pww0000gn/T//Rtmpcxzs34/openmp-detect-137df4ba71a0a.cpp -o /var/folders/b0/vt_1hj2d6yd8myx9lwh81pww0000gn/T//Rtmpcxzs34/openmp-detect-137df4ba71a0a.o if test "z/var/folders/b0/vt_1hj2d6yd8myx9lwh81pww0000gn/T//Rtmpcxzs34/openmp-detect-137df4ba71a0a.o" != "z"; then \ echo clang++ -arch arm64 -std=gnu++17 -dynamiclib -Wl,-headerpad_max_install_names -undefined dynamic_lookup -L"/Library/Frameworks/R.framework/Resources/lib" -L/opt/R/arm64/lib -L/opt/R/arm64/lib -o /var/folders/b0/vt_1hj2d6yd8myx9lwh81pww0000gn/T//Rtmpcxzs34/openmp-detect-137df4ba71a0a.so /var/folders/b0/vt_1hj2d6yd8myx9lwh81pww0000gn/T//Rtmpcxzs34/openmp-detect-137df4ba71a0a.o -F/Library/Frameworks/R.framework/.. -framework R ; \ clang++ -arch arm64 -std=gnu++17 -dynamiclib -Wl,-headerpad_max_install_names -undefined dynamic_lookup -L"/Library/Frameworks/R.framework/Resources/lib" -L/opt/R/arm64/lib -L/opt/R/arm64/lib -o /var/folders/b0/vt_1hj2d6yd8myx9lwh81pww0000gn/T//Rtmpcxzs34/openmp-detect-137df4ba71a0a.so /var/folders/b0/vt_1hj2d6yd8myx9lwh81pww0000gn/T//Rtmpcxzs34/openmp-detect-137df4ba71a0a.o -F/Library/Frameworks/R.framework/.. -framework R ; \ fi

After the regex, we're left with:

clang++ -arch arm64 -std=gnu++17

This is the same as:

R CMD config CXX

clang++ -arch arm64 -std=gnu++17

I'm not immediately seeing a huge benefit for a disk write and a regex to end up back at CXX config. Is there something regarding a custom toolchain that this would protect against?

See earlier discussion between @kevinushey and myself. It is possible that, say, CXX != CXX17 so that asking for the former does not give us the answer.

I agree with the sentiment of this being more brittle but on balance that may be what we have to go with.

Yeah, that's exactly it. For example, if the USE_CXX17 environment variable is set, then CXX17 will be used during compilation. Some users might be doing things like this in their .Renviron to enforce the use of their configured CXX17 compiler for different packages.

(And, as far as I can see, there isn't a way to ask R "what compiler do you want to use right now?" beyond something like this.)

eddelbuettel · 2025-12-03T19:38:41Z

So I think I am going to merge this one now, with a big thank you to both of you. Working on macOS remains both very rewarding (say those who use it 😉 ) and challenging (say those supporting it) but this ought to be a net step ahead.

alternate openmp plugin changes

da6266a

kevinushey commented Dec 2, 2025

View reviewed changes

alternative using R CMD SHLIB

9ad19e6

eddelbuettel approved these changes Dec 2, 2025

View reviewed changes

coatless reviewed Dec 3, 2025

View reviewed changes

Merge branch 'master' into feature/macos-openmp-plugin-tweaks

349a3dc

eddelbuettel merged commit 551c466 into master Dec 3, 2025
26 checks passed

eddelbuettel deleted the feature/macos-openmp-plugin-tweaks branch December 3, 2025 19:39

Uh oh!

Conversation

kevinushey commented Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Pull Request Template for Rcpp

Checklist

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!

eddelbuettel left a comment

Choose a reason for hiding this comment

Uh oh!

eddelbuettel commented Dec 2, 2025

Uh oh!

coatless left a comment

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!

kevinushey Dec 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

eddelbuettel commented Dec 3, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

kevinushey commented Dec 2, 2025 •

edited

Loading

kevinushey Dec 3, 2025 •

edited

Loading