Feature Request: Refactoring Install Process

[Jordan-Pierce]

FYI, installing on MacOS throws an error at

        elif osused == 'Darwin':
            print('Trying to install gdal...')
            try:
                check_call(['brew', 'install', 'gdal'],
                        stdout=open(os.devnull, 'wb'), stderr=STDOUT)

and also

        out = subprocess.getstatusoutput(['cmake', '--version'])
        if out[0] != 0:
            if osused == 'Darwin':
                print('Trying to install cmake...')
                try:
                    check_call(['brew', 'install', 'cmake'],
                               stdout=open(os.devnull, 'wb'), stderr=STDOUT)
                except:
                    raise Exception('Impossible to install cmake through homebrew. Please install manually cmake before running '
                                    'this script.\nInstallation aborted.')

However, running the actual command from command line:

brew install gdal
brew install cmake

performs fine. Would help debugging, but I don't myself have a Mac.

[Jordan-Pierce]

Following up on this: we have a few folks that are installing on MacOS and the following steps have made it install smoothly:

(Thank Clint)

• install brew (if not already)
• brew install gdal
• brew install cmake
• brew install libmagic
• install miniconda, create conda environment (conda create --name taglab11 python=3.11 -y)
• pip install torch torchvision (No CUDA, no problems, yay)
• python TagLab/install.py

This has worked consistently three times on different laptops. I believe the calls to install brew within the install.py script is not doing as expected, and is perceived as an error.

What are your thoughts on having OS-specific installs (GDAL, brew, CUDA, sudo apt-get install, etc) outside of the install.py script, and instead just include those instructions in the readme / wiki? That way it's on the user instead of the install.py script. This would also remove the need for TagLab to specify versions of pytorch / CUDA, which is currently hard-coded and I'm assuming annoying to maintain.

That way, all OS-specific installs can be done by the user first, and then all OS-agnostic installs can be put into a pyproject.toml / setup.py script, so TagLab can be a pip install?

More than happy to make a PR for this.

@gaiapavoni @maxcorsini