Separated llvm docker

Author: leventeBajczi

backend/Dockerfile

@@ -1,45 +1,34 @@
+ARG LLVM_IMAGE=ghcr.io/ftsrg/cir-demo-llvm:latest
+
+# Allow copying LLVM from the prebuilt llvm image into this image. This way you
+# can build the LLVM image locally (tag it with the same name) and then build
+# the backend image on your machine without running the costly LLVM build in CI.
+FROM ${LLVM_IMAGE} AS llvm
+
 FROM node:25-trixie
 
 WORKDIR /app
 
+# Copy LLVM installation from the llvm image
+
 RUN wget 'https://github.com/ftsrg/theta/releases/latest/download/Theta.zip' -O /app/Theta.zip \
  && unzip /app/Theta.zip \
  && rm /app/Theta.zip \
  && sed -i 's/export JAVA_VERSION=17/export JAVA_VERSION=21/g' /app/Theta/theta-start.sh
 
 # copy package and source
 COPY backend /app/backend
+COPY --from=llvm /opt/cir /app/backend/bin
 
 WORKDIR /app/backend
 RUN npm install --production
 
 ENV DEBIAN_FRONTEND=noninteractive
 RUN apt-get update \
  && apt-get install -y --no-install-recommends ca-certificates libstdc++6 \
-	build-essential cmake ninja-build pkg-config python3 git \
+    build-essential cmake ninja-build pkg-config python3 git \
     openjdk-21-jdk-headless libgomp1 libmpfr-dev \
-    clang lld
-
-# Build LLVM/ClangIR if backend/bin doesn't exist
-# This checks if bin directory exists in the COPY context, and if not, builds LLVM
-RUN if [ ! -d /app/backend/bin ]; then \
-    cd /tmp && \
-    git clone --depth=1 https://github.com/llvm/clangir.git && \
-    cmake -S clangir/llvm -B build -GNinja \
-        -DCLANG_ENABLE_CIR=ON \
-        -DLLVM_ENABLE_PROJECTS="clang;mlir" \
-        -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-        -DLLVM_TARGETS_TO_BUILD=host \
-        -DCMAKE_CXX_COMPILER=clang++ \
-        -DCMAKE_C_COMPILER=clang \
-        -DLLVM_USE_LINKER=lld \
-        -DBUILD_SHARED_LIBS=ON \
-        -DCMAKE_INSTALL_PREFIX=/app/backend/bin \
-        -DLLVM_FORCE_VC_REVISION=$(head -c7 /tmp/clangir/.git/refs/heads/main)-zorn && \
-    cmake --build build -j$(nproc) && \
-    cmake --install build && \
-    cd / && rm -rf /tmp/clangir /tmp/build; \
-    fi
+ && rm -rf /var/lib/apt/lists/*
 
 # Build the xcfa-mapper C++ tool so the backend can call it. We copy the
 # source tree and invoke CMake to produce the binary at xcfa-mapper/build/xcfa-mapper

docker/README.md

@@ -26,3 +26,33 @@ Usage
 Notes
 - The generated cert is self-signed; browsers will warn. For production, use a CA-signed certificate.
 - The backend credentials file is stored as plaintext JSON for simplicity; for real deployments use docker secrets or a vault.
+
+Building the LLVM base image locally
+-----------------------------------
+
+If you want to build the heavy LLVM/ClangIR image locally (so the backend image can reuse it), you can build and tag it locally and optionally push it to GitHub Container Registry (ghcr.io).
+
+1) Build locally and tag as the expected registry name (so backend builds pick it up):
+
+```sh
+docker build -f docker/llvm.Dockerfile -t ghcr.io/ftsrg/cir-demo-llvm:latest .
+```
+
+2) (Optional) Push to GitHub Container Registry if you want others or CI to pull it:
+
+```sh
+# Create a PAT with "read:packages" and "write:packages" and (recommended) repo scope.
+echo "${CR_PAT}" | docker login ghcr.io -u YOUR_GITHUB_USERNAME --password-stdin
+docker push ghcr.io/ftsrg/cir-demo-llvm:latest
+```
+
+After the image is available locally (or in ghcr), building the backend image will copy `/opt/cir` from that llvm image into the backend image. The backend Dockerfile accepts a build-arg `LLVM_IMAGE` so you can override the reference if you used a different tag:
+
+```sh
+docker build --build-arg LLVM_IMAGE=ghcr.io/ftsrg/cir-demo-llvm:latest -f backend/Dockerfile -t ghcr.io/ftsrg/cir-demo-backend:latest .
+```
+
+Notes:
+- Building LLVM is resource- and time-intensive. Building it once locally and pushing it to ghcr is usually faster overall.
+- The `docker/llvm.Dockerfile` installs LLVM into `/opt/cir` (this is what the backend copies into its final image).
+

docker/llvm.Dockerfile

@@ -0,0 +1,36 @@
+FROM debian:trixie-slim
+
+ARG DEBIAN_FRONTEND=noninteractive
+ENV LC_ALL=C.UTF-8
+
+RUN apt-get update \
+ && apt-get install -y --no-install-recommends \
+    ca-certificates git cmake ninja-build build-essential pkg-config python3 \
+    clang lld wget curl ca-certificates libstdc++6 libgomp1 libmpfr-dev \
+ && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /tmp
+
+# Clone ClangIR (shallow) and build LLVM with CLANG_ENABLE_CIR
+RUN git clone --depth=1 https://github.com/llvm/clangir.git && \
+    cmake -S clangir/llvm -B build -GNinja \
+        -DCLANG_ENABLE_CIR=ON \
+        -DLLVM_ENABLE_PROJECTS="clang;mlir" \
+        -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+        -DLLVM_TARGETS_TO_BUILD=host \
+        -DCMAKE_CXX_COMPILER=clang++ \
+        -DCMAKE_C_COMPILER=clang \
+        -DLLVM_USE_LINKER=lld \
+        -DBUILD_SHARED_LIBS=ON \
+        -DCMAKE_INSTALL_PREFIX=/opt/cir \
+        -DLLVM_FORCE_VC_REVISION=$(head -c7 clangir/.git/refs/heads/main)-zorn && \
+    cmake --build build -j$(nproc) && \
+    cmake --install build && \
+    rm -rf /tmp/clangir /tmp/build
+
+ENV PATH="/opt/cir/bin:${PATH}"
+
+LABEL org.opencontainers.image.title="cir-demo-llvm" \
+      org.opencontainers.image.description="Base image with LLVM+Clang built with CLANG_ENABLE_CIR and installed to /opt/cir"
+
+CMD ["/bin/sh","-c","echo 'This image only provides LLVM in /opt/cir' && /bin/sh"]