Building fails on latest git master

[evertvorster]

Hi there!

Building from sources currently fails on my system. I can provide the full logs if requested, but for brevity's sake, here is the error:

--   Found lilv-0, version 0.24.26
CMake Error at /usr/lib/cmake/opencv4/OpenCVConfig.cmake:86 (find_package):
  By not providing "FindCUDA.cmake" in CMAKE_MODULE_PATH this project has
  asked CMake to find a package configuration file provided by "CUDA", but
  CMake did not find one.

  Could not find a package configuration file provided by "CUDA" (requested
  version 12.8) with any of the following names:

    CUDAConfig.cmake
    cuda-config.cmake

  Add the installation prefix of "CUDA" to CMAKE_PREFIX_PATH or set
  "CUDA_DIR" to a directory containing one of the above files.  If "CUDA"
  provides a separate development package or SDK, be sure it has been
  installed.
Call Stack (most recent call first):
  /usr/lib/cmake/opencv4/OpenCVConfig.cmake:108 (find_host_package)
  CMakeLists.txt:284 (find_package)

It looks like the compiler is looking for cuda 12.8 specifically.
On my system I have cuda 12.8.1 installed:

extra/cuda 12.8.1-1 (2.5 GiB 6.5 GiB) (Installed)

I also see references to opencv4. I have opencv4 installed, with cuda bindings:

extra/opencv-cuda 4.11.0-7 (79.9 MiB 1.7 GiB) (Installed)

Compilation from sources worked until fairly recently, but since it gets updated regularly and it is a rolling release something might have changed on my system that prevents building now.

Did some compiler switch related to this change recently? Which direction should I look into to solve this issue?
What are some things that I can try?

Here is some more information on my system:

Operating System: Arch Linux 
KDE Plasma Version: 6.3.4
KDE Frameworks Version: 6.12.0
Qt Version: 6.9.0
Kernel Version: 6.13.8-arch1-1.2-g14 (64-bit)
Graphics Platform: Wayland
Processors: 32 × AMD Ryzen 9 7945HX3D with Radeon Graphics
Memory: 62.0 GiB of RAM
Graphics Processor 1: AMD Radeon 610M
Graphics Processor 2: NVIDIA GeForce RTX 4090 Laptop GPU
Manufacturer: ASUSTeK COMPUTER INC.
Product Name: ROG Strix G733PYV_G733PYV
System Version: 1.0

[ddennedy]

Nothing changed in MLT. It is a OpenCV-CUDA CMake issue. You can disable the OpenCV module in MLT if you would like.

[evertvorster]

Thanks for your input.
However, I do use the CUDA functionality. Problems that pop up in Arch usually starts popping up in other distros if it is not dealt with.
I'll have a poke around the opencv-cuda install to see if I can detect what is different about its install.

[ddennedy]

Is your log snippet actually from running cmake for MLT or from OpenCV? I cannot see from it anything that MLT needs to “deal with.”

[evertvorster]

This is compiling mlt from git master with the following commands:

build(){        rm -rf build
                mkdir -p build
                cd build
                cmake "../mlt"\
                  -DCMAKE_BUILD_TYPE=RelWithDebInfo \
                  -DCMAKE_INSTALL_PREFIX=/usr \
                  -DMOD_OPENCV=ON \
                  -DMOD_GLAXNIMATE=ON \
                  -DMOD_GLAXNIMATE_QT6=ON \
                  -DMOD_QT6=ON \
                  -DMOD_MOVIT=ON \
                  -DSWIG_PYTHON=ON
                make
                }

This used to work reliably up until now.
Of course, something might indeed have changed in regards with how opencv4 has been installed, this is a rolling release distrobution after all.

Its rather early to lay the blame on any package at the moment. However, I would like to know how to troubleshoot this further to figure out the best way to get mlt from git master to compile on this system again. The reason I reported it here, is that it is mlt that is not building, and I figured that you might be interested in something that might be potentially lurking in the future of other distrobutions.

[ddennedy]

CMake Error at /usr/lib/cmake/opencv4/OpenCVConfig.cmake

You should check the Issues in the OpenCV project.

[evertvorster]

I have checked the github page of OpenCV, and there are indeed some reports of issues with building against Qt 6.9.

The bit that I am not understanding is that I have opencv installed on my system. Does mlt not build against the binary that is installed? Is it somehow building a copy of opencv in it's own tree?

[ddennedy]

this project has asked CMake to find a package configuration file provided by "CUDA",

MLT does not try to find CUDA. The "this package" refers to the "CMake Error at /usr/lib/cmake/opencv4/OpenCVConfig.cmake" not provided by MLT, which is why I said it can be an OpenCV problem.

CMake generally provides 2 ways to find dependencies. One is good-old pkg-config via the pkg_check_modules() function. The other is find_package(), which uses CMake modules installed in various directories defined in a CMAKE_MODULE_PATH. Some are provided by CMake, but there are numerous of these so-called modules maintained by various parties, usually not the projects themselves! It is like organized chaos; I prefer pkg-config. You can check your package manager to see which provided that file.

[evertvorster]

Hmm, I can see that opencv-cuda does, indeed provide /usr/lib/cmake/opencv4/OpenCVConfig-version.cmake.
However, this package then goes off and tries to find:

    CUDAConfig.cmake
    cuda-config.cmake

As per the error.
The cuda install indeed does not provide any cmake config files, rather it provides pkgconfig files.
This must be a new development, and I'll see if I can pester the packager of cuda for comments.

Thanks for your insight.

[ddennedy]

I found this documentation page about FindCUDA.cmake that includes a deprecation notice. CMake 4, which generally removes deprecated things, was released a few weeks ago and creating big ripples in the ecosystem. For example, I created a fix for OpenCV for this bug opencv/opencv#27122. The fix was merged minutes ago; however, it was a simple minimal fix. I am not surprised if there are less explicit lurking issues like this one. What is your cmake version?

[evertvorster]

My cmake is at 4.0.1
Thanks for the extra info.
I'll be bugging the packagers of both opencv-cuda and cuda itself.
Well, probably not cuda itself, but at least opencv-cuda, because it is their cmake file that references FindCUDA.cmake. Might even have a look at opencv-cuda repo again to see if there is any reports on this...

[evertvorster]

Update for you here...
I opened an issue with opencv, specifically for this issue:
opencv/opencv#27236

The recommendation from the OpenCV guys is to build opencv-cuda with a compiler flag that enables CUDA as a language. I have reported this to the maintainer of the opencv-cuda, and hopefully once that has been updated mlt will compile again from sources, on Arch at least.

[evertvorster]

The changes have made it into opencv-cuda in Arch, and mlt-git builds again.