Description of change

Building from source against recent llama.cpp releases fails with two distinct C++/CMake errors. This PR fixes both so that source download --release latest followed by source build works out of the box.

This is independent of #591 (which handles TypeScript-layer Gemma 4 support). Once #591 is merged, users who want to use Gemma 4 today still need these build fixes, because 3.18.1 bundles llama.cpp b8390, while Gemma 4's gemma4 architecture only shipped in a later llama.cpp release — so source download --release latest is the only path to it.

Context: I hit this while trying to run Gemma 4 via source download --release latest (which pulled b8831).

1. llama/addon/addon.cpp:239 — std::atomic_bool copy-initialization

AppleClang 17 (Xcode 17) rejects copy-initialization from bool because std::atomic has a deleted copy constructor:

error: copying variable of type 'std::atomic_bool' (aka 'atomic<bool>')
       invokes deleted constructor
    static std::atomic_bool loaded = false;
                            ^        ~~~~~

Fixed by switching to brace-initialization:

static std::atomic_bool loaded{false};

This is a minimal, standards-conforming change that is backward-compatible with all supported C++11+ compilers.

2. llama/CMakeLists.txt:132 — common target renamed to llama-common

Upstream llama.cpp renamed the common CMake target to llama-common (the produced library is now libllama-common.dylib). The current link declaration resolves to a bare -lcommon flag, which the linker cannot find:

ld: library 'common' not found
clang++: error: linker command failed with exit code 1

Fixed by linking against the renamed target:

target_link_libraries(${PROJECT_NAME} "llama-common")

How this was verified

• npm run test:typescript — passes
• npm run lint — passes
• npm run test:standalone — passes (28 files, 169 tests)
• npx --no node-llama-cpp source download --release latest followed by source build — succeeds on macOS 15 arm64 (Apple M-series), AppleClang 17, Node.js 22.15, Metal backend
• A minimal end-to-end reproduction (loading gemma-4-E4B-it-Q8_0.gguf via LlamaChatSession with getLlama("lastBuild")) loads the model and streams responses correctly

Test output screenshot

1. npm run test:typescript / npm run lint / npm run test:standalone

2. npx --no node-llama-cpp source download --release latest

➜ npx --no node-llama-cpp source download --release latest
Repo: ggml-org/llama.cpp
Release: latest
GPU: Metal

✔ Fetched llama.cpp info
✔ Removed existing llama.cpp directory
✔ Cloned ggml-org/llama.cpp (GitHub)
◷ Compiling llama.cpp

(... compile output truncated ...)

[ 94%] Built target llama-common
[ 94%] Building CXX object CMakeFiles/llama-addon.dir/addon/AddonGrammarEvaluationState.cpp.o
[ 95%] Building CXX object CMakeFiles/llama-addon.dir/addon/AddonGrammar.cpp.o
[ 95%] Building CXX object CMakeFiles/llama-addon.dir/addon/AddonModelData.cpp.o
[ 96%] Building CXX object CMakeFiles/llama-addon.dir/addon/AddonModel.cpp.o
[ 96%] Building CXX object CMakeFiles/llama-addon.dir/addon/AddonContext.cpp.o
[ 98%] Building CXX object CMakeFiles/llama-addon.dir/addon/AddonSampler.cpp.o
[ 98%] Building CXX object CMakeFiles/llama-addon.dir/addon/addonGlobals.cpp.o
[ 98%] Building CXX object CMakeFiles/llama-addon.dir/addon/AddonModelLora.cpp.o
[ 98%] Building CXX object CMakeFiles/llama-addon.dir/addon/addon.cpp.o
[ 98%] Building CXX object CMakeFiles/llama-addon.dir/addon/globals/addonLog.cpp.o
[ 99%] Building CXX object CMakeFiles/llama-addon.dir/addon/globals/addonProgress.cpp.o
[ 99%] Building CXX object CMakeFiles/llama-addon.dir/addon/globals/getGpuInfo.cpp.o
[100%] Building CXX object CMakeFiles/llama-addon.dir/addon/globals/getMemoryInfo.cpp.o
[100%] Building CXX object CMakeFiles/llama-addon.dir/addon/globals/getSwapInfo.cpp.o
[100%] Linking CXX shared library Release/llama-addon.node
[100%] Built target llama-addon
✔ Compiled llama.cpp

Pull-Request Checklist

• Code is up-to-date with the master branch
• npm run format to apply eslint formatting
• npm run test passes with this change
• This pull request links relevant issues as Fixes #0000 — N/A (no existing issue tracks these specific build errors; related context in feat: Gemma 4 support #591)
• There are new or updated unit tests validating the change — N/A (build-level fix; verified by existing CI compilation step and full test:standalone suite)
• Documentation has been updated to reflect this change — N/A (no public API or docs affected)
• The new commits and pull request title follow conventions explained in pull request guidelines