# ============================================================
# TurboOCR — CPU-only build (ONNX Runtime backend, no GPU required)
# Base image: ubuntu:24.04
#
# Produces: /app/build_cpu/paddle_cpu_server (HTTP + gRPC server)
#
# Image size: ~500 MB (vs ~10 GB for the GPU image).
# No TRT compilation on first start — ONNX Runtime is used directly.
# Startup is fast (~30 s) and requires no NVIDIA driver.
#
# Build:  docker build -f Dockerfile.cpu -t turboocr-cpu .
# ============================================================

ARG TURBOOCR_VERSION=v2.1.1
ARG ORT_VERSION=1.22.0
# Registry mirror prefix — leave empty for direct pull.
# China users: set to "docker.m.daocloud.io/" to proxy Docker Hub via DaoCloud.
ARG DOCKER_MIRROR=

FROM ${DOCKER_MIRROR}ubuntu:24.04

# Re-declare ARGs after FROM so they remain in scope
ARG TURBOOCR_VERSION
ARG ORT_VERSION

ENV DEBIAN_FRONTEND=noninteractive

# Install build dependencies
RUN apt-get update && apt-get install -y --no-install-recommends \
    cmake \
    g++ \
    make \
    pkg-config \
    libopencv-dev \
    nginx \
    gosu \
    libgrpc++-dev \
    libc-ares-dev \
    libprotobuf-dev \
    protobuf-compiler \
    protobuf-compiler-grpc \
    libjsoncpp-dev \
    uuid-dev \
    zlib1g-dev \
    libssl-dev \
    git \
    wget \
    curl \
    ca-certificates \
    && rm -rf /var/lib/apt/lists/*

# Install Drogon HTTP framework (async, epoll-based)
RUN cd /tmp && \
    git clone --depth 1 --branch v1.9.12 https://github.com/drogonframework/drogon.git && \
    cd drogon && git submodule update --init && \
    mkdir build && cd build && \
    cmake .. -DBUILD_EXAMPLES=OFF -DBUILD_CTL=OFF -DBUILD_ORM=OFF \
             -DBUILD_POSTGRESQL=OFF -DBUILD_MYSQL=OFF -DBUILD_SQLITE=OFF \
             -DBUILD_REDIS=OFF -DBUILD_TESTING=OFF && \
    make -j$(nproc) && make install && \
    rm -rf /tmp/drogon

# Install ONNX Runtime C++ SDK
RUN cd /tmp && \
    wget -q "https://github.com/microsoft/onnxruntime/releases/download/v${ORT_VERSION}/onnxruntime-linux-x64-${ORT_VERSION}.tgz" && \
    tar xzf "onnxruntime-linux-x64-${ORT_VERSION}.tgz" && \
    cp -r "onnxruntime-linux-x64-${ORT_VERSION}/include/"* /usr/local/include/ && \
    cp "onnxruntime-linux-x64-${ORT_VERSION}/lib/libonnxruntime.so"* /usr/local/lib/ && \
    ldconfig && rm -rf /tmp/onnxruntime*

# Clone TurboOCR at the pinned release tag
RUN git clone --depth 1 --branch "${TURBOOCR_VERSION}" \
    https://github.com/aiptimizer/TurboOCR.git /app

WORKDIR /app

# Install fastpdf2png (PDF renderer — PDFium vendored in third_party/).
# Copy vendored libpdfium first so the installer does not need network access.
RUN cp third_party/pdfium/lib/libpdfium.so /usr/lib/ && ldconfig && \
    bash scripts/install_fastpdf2png.sh && \
    { cp bin/libpdfium.so /usr/lib/ 2>/dev/null || true; } && \
    ldconfig

# Build CPU-only mode with ONNX Runtime backend
RUN mkdir -p build_cpu && cd build_cpu && \
    cmake .. -DUSE_CPU_ONLY=ON -DFETCH_MODELS=OFF && \
    make -j$(nproc)

# Create non-root user and redirect /app/models/rec into the named cache volume.
RUN useradd -m -s /bin/bash ocr \
    && chmod +x /app/scripts/entrypoint.sh \
    && mkdir -p /home/ocr/.cache/turbo-ocr/models/rec /app/models \
    && ln -s /home/ocr/.cache/turbo-ocr/models/rec /app/models/rec

# Fetch all PP-OCRv5 language bundles (SHA256-verified from pinned GitHub Release)
ARG OCR_INCLUDE_SERVER=1
ENV OCR_INCLUDE_SERVER=${OCR_INCLUDE_SERVER}
RUN bash scripts/fetch_release_models.sh \
    && chown -R ocr:ocr /app /home/ocr/.cache

EXPOSE 8000 50051

ENTRYPOINT ["/app/scripts/entrypoint.sh"]
CMD ["./build_cpu/paddle_cpu_server"]