The size of libraries range all the way from single tiny functions like left-pad to massive frameworks like Django and Rails. How large do you think a library ought to be in order to be publishable?
I like to follow a rule of thumb I learned at a SciPy Conference a few years ago: whenever possible pull out functionality into separate modules. I think the GNU program mantra of do one thing and do it well applies. From a philosophical standpoint, I like the simplicity of the standalone left-pad library.
It is not always possible though, like in the case of Django: part of the appeal is the single batteries included framework for getting something up and the subsequent relative standardization of code bases that implement it. However, I tend to use 'flask' for its flexibility/modularity and simplicity especially in the context of simple data oriented services.
Scipy itself is a good example of a monolithic library that I think makes sense as a single module, since people using some of the functionality are much more likely to need other bits of scipy in the same code, and there are a lot (or at least some) of inter-dependencies within the scipy routines that make it largely more efficient for developers to keep it all together.
In the case of matplotlib, which lacks a native Geographic projection functionality. Multiple plugin modules have been developed to support that functionality, both with different approaches and features. If this had been an included feature to begin with, we wouldn't have the innovation and choice in different solutions that exists now.
It is not always possible though, like in the case of Django: part of the appeal is the single batteries included framework for getting something up and the subsequent relative standardization of code bases that implement it. However, I tend to use 'flask' for its flexibility/modularity and simplicity especially in the context of simple data oriented services.
Scipy itself is a good example of a monolithic library that I think makes sense as a single module, since people using some of the functionality are much more likely to need other bits of scipy in the same code, and there are a lot (or at least some) of inter-dependencies within the scipy routines that make it largely more efficient for developers to keep it all together.
In the case of matplotlib, which lacks a native Geographic projection functionality. Multiple plugin modules have been developed to support that functionality, both with different approaches and features. If this had been an included feature to begin with, we wouldn't have the innovation and choice in different solutions that exists now.