Lessons learnt in setting up a build and test workflow on WSL on windows

[kumaakh]

It seemed attractive to be able to use a wsl instance on a windows machine to build and test Apra Pipes. It lead to multiple lessons... here is a quick wrap around all of them:

1. Vampire/setup-wsl@v1 is a great action to do most things related to wsl. It creates a wrapper script arround every command/script that you'd want to run on the WSL instance. it works almost like how docker container based builds work in GHA.
2. There are many side effects of checking out code on the windows host using action/checkout action:
   • all scripts (sh, bash, python) need to be passed thriugh dos2unix else they don't work on wsl
   • checkout on windows does not create symbolic links which may be needed in many situations... e.g. we typically link *.so.1 file to *.so file. the file in question was too small so i chose to make a copy instead. It appears that this can fixd by following this
3. The over speed of every thing is very very slow on WSL. e.g. I used the same intel i9 machine for windows build as well as wsl build and I make sure that they are triggered by the same "runner" resource hence they do not ever run concurrently. I steps with require too many file copy tasks e.g. vcpkg install take 5x to 7x longer on WSL as compared to windows host. Likewise the main cmake based it self is also very slow.
   • this forced me to start caching GST and libmp4 build even on self hosted builders to save multiple minutes of build time

[kumaakh]

1. Getting the CUDA installation right on WSL is very important... this documentation helps alot.
2. VVIMP it says make sure that you install cuda-toolkit and not cuda I followed local CUDA11.7 on WSL .
3. The objective is to NOTinstall the driver on WSL install steps and they worked fine.
4. In the end I had to add /usr/local/cuda/bin to PATH via .profile and source ~\.profle
5. install cudnn from offline deb installer
6. install tesorrt from offline deb installer