build error in ubuntu with libpng

[maminus]

🐛 Bug

I tried to build torchvision package.
I found build error at linking image.so.

To Reproduce

Steps to reproduce the behavior:

$ git clone https://github.com/pytorch/vision.git
$ cd vision
$ python setup.py install

And error message.

g++ -pthread -shared -B /opt/conda/compiler_compat -L/opt/conda/lib -Wl,-rpath=/opt/conda/lib -Wl,--no-as-needed -Wl,--sysroot=/ build/temp.linux-x86_64-3.6/data/python3.6/vision/torchvision/csrc/cpu/image/image.o build/temp.linux-x86_64-3.6/data/python3.6/vision/torchvision/csrc/cpu/image/readpng_cpu.o -Llibpng-config: --libdir option is disabled in Debian/Ubuntu -L/var/lib/jenkins/.local/lib/python3.6/site-packages/torch/lib -L/opt/rocm/lib -lpng -lc10 -ltorch -ltorch_cpu -ltorch_python -lc10_hip -ltorch_hip -o build/lib.linux-x86_64-3.6/torchvision/image.so
g++: error: option: No such file or directory
g++: error: is: No such file or directory
g++: error: disabled: No such file or directory
g++: error: in: No such file or directory
g++: error: Debian/Ubuntu: No such file or directory
g++: error: unrecognized command line option ‘--libdir’; did you mean ‘--dumpdir’?
error: command 'g++' failed with exit status 1

g++ option -Llibpng-config: --libdir option is disabled in Debian/Ubuntu is wrong.

Expected behavior

-L option is correct libpng library path.

Root Cause & Environment

Now, setup.py is using libpng-config command with --libdir option, but it is disabled in Debian/Ubuntu.

$ libpng-config --libdir
libpng-config: --libdir option is disabled in Debian/Ubuntu

I tried rocm/pytorch:rocm3.5_ubuntu18.04_py3.6 docker image, but I think it reproduce in normal Ubuntu18.04.

$ python -m torch.utils.collect_env
Collecting environment information...
PyTorch version: 1.7.0a0+f083cea
Is debug build: No
CUDA used to build PyTorch: Could not collect

OS: Ubuntu 18.04.4 LTS
GCC version: (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0
CMake version: version 3.14.0

Python version: 3.6
Is CUDA available: Yes
CUDA runtime version: Could not collect
GPU models and configuration: Could not collect
Nvidia driver version: Could not collect
cuDNN version: Could not collect

Versions of relevant libraries:
[pip] numpy==1.18.1
[pip] torch==1.7.0a0+f083cea
[pip] torchvision==0.8.0a0+75f5b57
[conda] blas                      1.0                         mkl  
[conda] mkl                       2020.1                      217  
[conda] mkl-include               2020.1                      217  
[conda] mkl-service               2.3.0            py36he904b0f_0  
[conda] mkl_fft                   1.0.15           py36ha843d7b_0  
[conda] mkl_random                1.1.1            py36h0573a6f_0  
[conda] numpy                     1.18.1           py36h4f9e942_0  
[conda] numpy-base                1.18.1           py36hde5b4d6_1  
[conda] torchvision               0.8.0a0+75f5b57          pypi_0    pypi

• torch : master branch
• torchvision : master branch

Additional context

--L_opts option is enabled instead, but it's empty path in my environment.

So I tried to following patch, then build is succeeded.

diff --git a/setup.py b/setup.py
index 01276f1..e95137f 100644
--- a/setup.py
+++ b/setup.py
@@ -262,14 +262,21 @@ def get_extensions():
             png_version = parse_version(png_version)
             if png_version >= parse_version("1.6.0"):
                 print('Building torchvision with PNG image support')
-                png_lib = subprocess.run([libpng, '--libdir'],
+                help_result = subprocess.run([libpng, '--help'],
+                                         stdout=subprocess.PIPE)
+                lib_opt = '--libdir'
+                if '--L_opts' in help_result.stdout.decode('utf-8'):
+                    lib_opt = '--L_opts'
+                png_lib = subprocess.run([libpng, lib_opt],
                                          stdout=subprocess.PIPE)
                 png_include = subprocess.run([libpng, '--I_opts'],
                                              stdout=subprocess.PIPE)
                 png_include = png_include.stdout.strip().decode('utf-8')
                 _, png_include = png_include.split('-I')
                 print('libpng include path: {0}'.format(png_include))
-                image_library += [png_lib.stdout.strip().decode('utf-8')]
+                png_lib = png_lib.stdout.strip().decode('utf-8')
+                if png_lib:
+                    image_library += [png_lib]
                 image_include += [png_include]
                 image_link_flags.append('png')
             else:

[fmassa]

Thanks for opening this issue!

cc @andfoy can you have a look?

[andfoy]

@maminus, could you please check if #2398 fixes this issue?