Fatal error: 'climits' file not found (macOS build failure)

[mariouuche]

Hello.

I'm trying to install scikit-learn in developer mode on macOs. I'm following the official development installation guide.
Everything works until this step: pip install --editable . --verbose --no-build-isolation --config-settings editable-verbose=true.
During compilation, I get:
../../sklearn/svm/src/libsvm/svm.cpp:71:10: fatal error: 'climits' file not found
71 | #include <climits>
| ^~~~~~~~~
1 error generated.

I'm using macOS 26.3 and I installed xcode command line tools with xcode-select --install. I tried resetting xcode, but it didn't fix the compilation issue.
Has anyone encountered this before on macOS? Is there something from the installation guide I might be missing?

Thanks in advance.

[DhimanTarafdar]

This is a known issue on macOS when using newer Xcode versions. The climits header was removed in recent macOS SDK versions.
Try these fixes in order:
Fix 1 — Reset Xcode path:
sudo xcode-select --switch /Applications/Xcode.app/Contents/Developer

Then retry your original install command.
Fix 2 — Reinstall Command Line Tools:
sudo rm -rf /Library/Developer/CommandLineTools
xcode-select --install

Fix 3 — Use conda instead of pip:
conda install -c conda-forge scikit-learn

Fix 4 — Set SDK path manually:
export SDKROOT=$(xcrun --sdk macosx --show-sdk-path)
pip install --editable . --verbose --no-build-isolation --config-settings editable-verbose=true

Note: Fix 4 is often the most reliable solution for climits not found errors on macOS 13+. The SDKROOT environment variable helps the compiler locate the correct SDK headers.

Hope this helps! Let me know if any of these solutions work for you.

[mariouuche]

Fix 3 worked for me, thank you very much!

[aniruddhaadak80]

Hey @mariouuche,

The 'climits' file not found error on macOS means the C++ compiler isn't finding the standard library headers. Running xcode-select --install or exporting your CXX flags to point to a Homebrew GCC/Clang installation usually fixes this linkage issue immediately.

Let me know if you make any progress or if you need another fresh set of eyes on it!

[Farshidbagheri]

Just a small clarification: <climits> itself has not been removed. It is a standard C++ header. This error usually means that the compiler used during the build cannot find the C++ standard library headers.

So the issue is probably not in sklearn/svm/src/libsvm/svm.cpp, but in the local compiler / SDK / environment configuration.

A quick way to confirm the problem is to test whether your compiler can build a minimal C++ file:

cat > /tmp/test.cpp <<'EOF'
#include <climits>
#include <iostream>

int main() {
    std::cout << INT_MAX << std::endl;
    return 0;
}
EOF

clang++ /tmp/test.cpp -o /tmp/test
/tmp/test

If this fails with the same climits error, then the problem is definitely your macOS compiler setup, not scikit-learn.

For a developer-mode scikit-learn build, I would try the following order:

xcode-select -p
xcrun --show-sdk-path
clang++ -v /tmp/test.cpp -o /tmp/test

If the SDK / command line tools look broken, reinstalling Command Line Tools can help:

sudo rm -rf /Library/Developer/CommandLineTools
xcode-select --install

Then retry the simple clang++ test before retrying the scikit-learn build.

If you are using a conda environment, another robust approach is to use the conda-forge compiler toolchain consistently inside the environment, rather than mixing system clang, Homebrew tools, and conda packages:

conda create -n sklearn-dev -c conda-forge python numpy scipy cython meson-python ninja compilers
conda activate sklearn-dev

pip install --no-build-isolation --editable . --verbose --config-settings editable-verbose=true

Also, conda install -c conda-forge scikit-learn works if the goal is simply to install scikit-learn, because it avoids the local source build. But it is not the same as fixing the editable developer build from a local clone.

So I would summarize it this way:

• If you only need scikit-learn installed: conda install -c conda-forge scikit-learn is fine.
• If you need developer mode from source: first make sure clang++ can compile a minimal file including <climits>.
• If the minimal test fails: fix Xcode Command Line Tools / SDK / compiler environment first.
• If the minimal test passes but scikit-learn still fails: then the issue is more likely related to the Python build environment or mixed compiler toolchains.