Make setup.py require dash

Really just a suggestion, but I believe it would be nice to have setup.py require dash by default: https://github.com/plotly/dash-bio/blob/bd5ce53c4529ce3e0c783cb61eb6eab7b298dd93/setup.py#L11-L20

That way, when the user installs dash-bio, they simply need to create a venv and run:

pip install dash-bio

with dash, dash-html-components, dash_renderer being installed as well.

It could be implemented along the line of:

 setup( 
     name=package_name, 
     version=package["version"], 
     author=package['author'], 
     packages=[package_name, '{}/utils'.format(package_name), '{}/component_factory'.format(package_name)], 
     include_package_data=True, 
     license=package['license'], 
     description=package['description'] if 'description' in package else package_name, 
     install_requires=[
         'dash', 'dash-html-components', 'dash_renderer'
     ] 
 ) 

[alexcjohnson]

An interesting discussion indeed. The app/library and abstract/specific distinctions seem like good starting points for thinking through what will actually happen when a random user installs and uses this software, but they’re not the end of the story.

An app, in this sense, means no further code is going to be connected to it, it’s just going to be run as is. In that case locking down the exact versions of all the pieces makes sense: that’s how the app was put together and tested, so stick with that.

dash-bio is fairly near the top of the stack, but it’s still a library in the sense that it integrates with the user’s code, which may independently import and use some of these same packages. We still test it with specific versions of all the dependents, so it would be nice if we could ensure that these same versions will always be used by everyone using it, but we can’t, at least not in Python, where a given package can only have a single version in a given environment (as opposed to JS where the recursive node_modules structure allows each package to use its own versions of dependents… but that has its own problems!). So to play nice with our users’ environments, we want to be flexible with versions of these dependencies.

That’s fine for things like pandas which is widely used and (presumably) stable enough that whatever version a user has already will very likely work for us too. But for dash it’s not fine: dash and dash-bio are being developed in tandem, we know there are breaking changes that preclude earlier versions, and it’s completely reasonable for us to require users to upgrade dash when they upgrade dash-bio. So pandas is abstract, dash is specific, and every other dependent should be evaluated on a similar basis.

[mkcor]

Hi @xhlulu,

Thanks for bringing this up. It’s probably okay to let dependencies be that ‘abstract’ in the setup.py file, although Dash Bio is more of an app than a library… For a discussion on this, please read: https://caremad.io/posts/2013/07/setup-vs-requirement/)

In any case, I would make sense to specify sklearn>=0.20.1 at least in the requirements.txt file. Cc’ing @shammamah here.