Arch Linux User Repository

@ework thanks for pointing this. I'll remove in the next update, as I think it does not hurt let it there for the time being.

@dbermond, we fixed the exec stack issue in TensorRT 10.9 by linking with "-Wl,-z,noexecstack", so you can drop that patch if you'd like.

I'm trying to build this tensorrt package but running into an issue where the Python.h isn't found. I've these packages installed:

• cuda 12.8.0-1
• cudnn 9.7.0.66-1
• python 3.13.2-1

Contents of /usr/include/:

$ ls /usr/include | grep -i python
pybind11
python2.7
python3.13

$ fd "Python.h"
python3.13/Python.h
python2.7/Python.h

The output of the makepkg command:

~/packages/tensorrt/src/TensorRT/python/build ~/packages/tensorrt/src/TensorRT/python
CMake Deprecation Warning at CMakeLists.txt:18 (cmake_minimum_required):
  Compatibility with CMake < 3.10 will be removed from a future version of
  CMake.

  Update the VERSION argument <min> value.  Or, use the <min>...<max> syntax
  to tell CMake that the project requires at least <min> but has been updated
  to work with policies introduced by <max> or earlier.


-- Configuring done (0.0s)
-- Generating done (0.2s)
-- Build files have been written to: /home/sam/packages/tensorrt/src/TensorRT/python/build
[ 10%] Building CXX object CMakeFiles/tensorrt.dir/src/infer/pyAlgorithmSelector.cpp.o
In file included from /home/sam/packages/tensorrt/src/pybind11/include/pybind11/detail/../cast.h:13,
                 from /home/sam/packages/tensorrt/src/pybind11/include/pybind11/detail/../attr.h:13,
                 from /home/sam/packages/tensorrt/src/pybind11/include/pybind11/detail/class.h:12,
                 from /home/sam/packages/tensorrt/src/pybind11/include/pybind11/pybind11.h:13,
                 from /home/sam/packages/tensorrt/src/TensorRT/python/include/ForwardDeclarations.h:30,
                 from /home/sam/packages/tensorrt/src/TensorRT/python/src/infer/pyAlgorithmSelector.cpp:19:
/home/sam/packages/tensorrt/src/pybind11/include/pybind11/detail/../detail/common.h:213:10: fatal error: Python.h: No such file or directory
  213 | #include <Python.h>

@hsyysy fixed. Thanks for reporting.

There are two errors which can be solved.

1. It is mentioned here

https://forums.developer.nvidia.com/t/tensorrt-10-8-0-43-and-older-incompatible-with-new-glibc-2-41-r2-g0a7c7a3e283a/322769, this problem (caused by glibc 2.41) can be solved by adding

patchelf --clear-execstack ${pkgdir}/usr/lib/libnvinfer_builder_resource.so.${pkgver%.*}

to the package_tensorrt() function.

2. Error info of libnvonnxparser.so

this can be solved by removing 030-tensorrt-onnx-tensorrt-disable-missing-source-file.patch.

I even have that issue before Python 3.13 with the the pre-built ones. The issue seems to come from libnvonnxparser.so binary and python bindings are required for the package python-pytorch-tensorrt.

@thesword when you modify the PKGBUILD, you are for your own, and I cannot help you with a modified PKGBUILD. I've disabled the building of the main python bindings (and use the pre-built ones from upstream) for a reason: it fails to build, and I could not find a way to make them to build. But now now you enable it for yourself and found what I've just said: it fails to build. Do not enable it.

I modified PKGBUILD to include python bindings. I added local -x TENSORRT_MODULE="tensorrt"

diff --git a/PKGBUILD b/PKGBUILD
index 9f34781..c9454cd 100644
--- a/PKGBUILD
+++ b/PKGBUILD
@@ -90,20 +90,21 @@ build() {
         -Wno-dev
     cmake --build build

-    # python bindings (fails to build with python 3.11)
-    #local _pyver
-    #_pyver="$(python -c 'import sys; print("%s.%s" %sys.version_info[0:2])')"
-    #local -x PYTHON_MAJOR_VERSION="${_pyver%%.*}"
-    #local -x PYTHON_MINOR_VERSION="${_pyver#*.}"
-    #local -x TARGET_ARCHITECTURE="$CARCH"
-    #local -x TRT_OSSPATH="${srcdir}/TensorRT"
-    #local -x CUDA_ROOT='/opt/cuda'
-    #local -x ROOT_PATH="${srcdir}/TensorRT"
-    #local -x EXT_PATH="$srcdir"
-    #local -x TRT_NONOSS_ROOT="${srcdir}/TensorRT-${pkgver}"
-    #git -C pybind11 checkout "v${_pybind11_ver}"
-    #cd TensorRT/python
-    #./build.sh
+    # python bindings
+    local _pyver
+    _pyver="$(python -c 'import sys; print("%s.%s" %sys.version_info[0:2])')"
+    local -x PYTHON_MAJOR_VERSION="${_pyver%%.*}"
+    local -x PYTHON_MINOR_VERSION="${_pyver#*.}"
+    local -x TARGET_ARCHITECTURE="$CARCH"
+    local -x TRT_OSSPATH="${srcdir}/TensorRT"
+    local -x CUDA_ROOT='/opt/cuda'
+    local -x ROOT_PATH="${srcdir}/TensorRT"
+    local -x EXT_PATH="$srcdir"
+    local -x TRT_NONOSS_ROOT="${srcdir}/TensorRT-${pkgver}"
+    local -x TENSORRT_MODULE="tensorrt"
+    git -C pybind11 checkout "v${_pybind11_ver}"
+    cd TensorRT/python
+    ./build.sh

     # python tools (pytorch-quantization currently fails to build)
     local _dir
@@ -132,9 +133,9 @@ package_tensorrt() {
 }

 package_python-tensorrt() {
-    #pkgdesc+=' (python bindings and tools)'
+    pkgdesc+=' (python bindings and tools)'
     pkgdesc+=' (python tools)'
-    #license=('LicenseRef-Custom' 'Apache-2.0')
+    license=('LicenseRef-Custom' 'Apache-2.0')
     license=('Apache-2.0')
     depends=('python' 'python-numpy' 'tensorrt')
     optdepends=('python-onnx: for onnx_graphsurgeon python module'
@@ -149,6 +150,7 @@ package_python-tensorrt() {
     #local _pyver
     #_pyver="$(python -c 'import sys; print("%s.%s" %sys.version_info[0:2])')"
     #python -m installer --destdir="$pkgdir" "TensorRT-${pkgver}/python/tensorrt-${pkgver%.*}-cp${_pyver/./}-none-linux_${CARCH}.whl"
+    python -m installer --destdir="$pkgdir" "TensorRT/python/build/bindings_wheel/dist/tensorrt-${pkgver%.*}-py3-none-any.whl"

     local _dir
     for _dir in TensorRT/tools/{onnx-graphsurgeon,Polygraphy,tensorflow-quantization}/dist

But now I got:

Python 3.13.1 (main, Dec  4 2024, 18:05:56) [GCC 14.2.1 20240910] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorrt
Traceback (most recent call last):
  File "<python-input-0>", line 1, in <module>
    import tensorrt
  File "/usr/lib/python3.13/site-packages/tensorrt/__init__.py", line 75, in <module>
    from .tensorrt import *
ImportError: /usr/lib/libnvonnxparser.so.10: undefined symbol: _ZTIN8onnx2trt16OnnxTrtExceptionE

@thesword Package updated. Disabled installation of the main tensorrt python bindings, as upstream currently does not support python 3.13 yet. I will enable it again when/if upstream supports python 3.13. Thanks for reporting.

It doesn't build anymore sice Python 3.13 update

==> Starting package_python-tensorrt()...
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/usr/lib/python3.13/site-packages/installer/__main__.py", line 98, in <module>
    _main(sys.argv[1:], "python -m installer")
    ~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.13/site-packages/installer/__main__.py", line 86, in _main
    with WheelFile.open(args.wheel) as source:
         ~~~~~~~~~~~~~~^^^^^^^^^^^^
  File "/usr/lib/python3.13/contextlib.py", line 141, in __enter__
    return next(self.gen)
  File "/usr/lib/python3.13/site-packages/installer/sources.py", line 162, in open
    with zipfile.ZipFile(path) as f:
         ~~~~~~~~~~~~~~~^^^^^^
  File "/usr/lib/python3.13/zipfile/__init__.py", line 1362, in __init__
    self.fp = io.open(file, filemode)
              ~~~~~~~^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: 'TensorRT-10.7.0.23/python/tensorrt-10.7.0-cp313-none-linux_x86_64.whl'
==> ERROR: A failure occurred in package_python-tensorrt().
    Aborting...