如何通过Yocto安装编译好的TensorRT Python Wheel包？

我为特定版本Python编译了无额外依赖的TensorRT Python绑定，现在要将生成的.whl文件安装到Yocto镜像中，使用了如下配方：

SUMMARY = "NVIDIA® TensorRT™, an SDK for high-performance deep learning inference, includes a deep learning inference optimizer and runtime that delivers low latency and high throughput for inference applications."
HOMEPAGE = "https://github.com/NVIDIA/TensorRT"

SRC_URI = "file://tensorrt-8.2.1.9-cp38-none-linux_aarch64.whl"

LICENSE = "Proprietary"
LIC_FILES_CHKSUM = "file://${COMMON_LICENSE_DIR}/Proprietary;md5=0557f9d92cf58f2ccdd50f62f8ac0b28"
                                                      
DEPENDS += "python3 python3-pip"

FILES_${PN} += "\
    ${libdir}/${PYTHON_DIR}/site-packages/* \
"

do_install() {                                                                                      
    pip install ${S}/tensorrt-8.2.1.9-cp38-none-linux_aarch64.whl
}

但配方在do_install步骤执行失败，错误信息：

ERROR: Execution of '/home/user/Desktop/tegra-demo-distro/build/tmp/work/aarch64-oe4t-linux/tensorrt/8.2.1-r0/temp/run.do_install.56472' failed with exit code 127

问题原因 #

Exit code 127表示找不到指定命令，这里是因为Yocto的构建环境是隔离的，直接调用pip无法定位到正确的工具；同时默认的pip install会把包安装到构建主机环境，而非Yocto的目标根文件系统。

修正后的配方 #

SUMMARY = "NVIDIA® TensorRT™, an SDK for high-performance deep learning inference, includes a deep learning inference optimizer and runtime that delivers low latency and high throughput for inference applications."
HOMEPAGE = "https://github.com/NVIDIA/TensorRT"

SRC_URI = "file://tensorrt-8.2.1.9-cp38-none-linux_aarch64.whl"

LICENSE = "Proprietary"
LIC_FILES_CHKSUM = "file://${COMMON_LICENSE_DIR}/Proprietary;md5=0557f9d92cf58f2ccdd50f62f8ac0b28"
                                                      
DEPENDS += "python3 python3-pip"
# 声明运行时依赖，确保镜像包含Python核心环境
RDEPENDS_${PN} += "python3-core"

FILES_${PN} += "\
    ${libdir}/${PYTHON_DIR}/site-packages/* \
"

do_install() {                                                                                      
    # 使用Yocto提供的pip变量，指定安装到目标根目录，禁用依赖检查
    ${PYTHON_PIP} install --root=${D} --no-deps ${S}/tensorrt-8.2.1.9-cp38-none-linux_aarch64.whl
}

关键修改点 #

• 用Yocto官方提供的${PYTHON_PIP}变量替代直接调用pip，确保在构建隔离环境中能找到正确的工具
• 添加--root=${D}参数，强制将包安装到Yocto的目标根文件系统（${D}是Yocto内置的安装目录变量）
• 添加--no-deps参数，避免pip尝试自动下载依赖（适配你的whl无额外依赖的特性）
• 添加RDEPENDS_${PN}声明运行时依赖，确保最终镜像包含Python核心运行环境

内容的提问来源于stack exchange，提问作者Damien