I couldn’t find a link to the presentation, but I did find notes of the talk.

Monkeypatching seems like a serious illness of the Ruby community; unlike the stodgy Java types, I firmly believe it is essential to be able to do it, but I also firmly believe it’s a strategy of last resort. Last resort. And it should never ever be done by a library or framework; only by application code which is standalone and won’t be mixed with other code down the line, or as part of a metaprogramming abstraction. Otherwise, you get clashes and interactions like this one.

method_missing, in contrast, is perfectly fine. I use it quite frequently without any ill effects (or rather its equivalent in my preferred language). In fact, it’s a prime enabler of OO metaprogramming; things like proxies and mock objects are impossible without it.

You’re right, Net::SFTP works well everywhere else, it’s only Rake (that I know of) that breaks it.

But when something breaks you have to fix it. I’m trying to learn by example so I can create code that — when it breaks, and most likely it would — will be easy to fix.

Figuring out what Rake does was easy, I just looked at the docs (document that which you extend). Figuring out what Net::SFTP is doing took a long while, digging through its code in search of the elusive mkdir that doesn’t exist, and then having to understand how it works to figure out the two patches.

And thanks for pointing out the confusion about stubbing.

I use RSpec for the test suite, it’s awesome and I love writing tests with it. The tests I’m referring to here were done by running the code against a live server and seeing what happens, so no mocking or stubbing involved.

Using method_missing to dispatch to underlying implementations is fine, especially if you don’t know what methods those implementations will support. See for example the EVDB Ruby API client. My choice would be to use explicit delegation whenever I know the methods I’m expecting to use, but Net:SFTP isn’t changing anytime soon.

And I think the other Chris meant something else by stubbing, than the sense in which you used it. Maybe the term mocking is more appropriate for his usage. The basic idea being that if the SFTP implementation expects :mkdir, then your tests won’t satisfy the expectation. Checkout RSpec and Mocha, both sweet mocking frameworks.