During the week I solicited feedback on my assessment of "The received wisdom of TDD and private methods". I got a small amount of good, usefull feedback, but not as much as I was hoping for.
However Sean came to the fore with a very long response, and I figure it's worth posting here so other people spot it and read it.
'ere 'tis, unabridged:
There are broadly two schools of thought on the issue of "private" methods and TDD:
1. private methods are purely an implementation detail that arise as part of the "refactor" portion of the cycle - they're completely irrelevant to the tests and they never need testing (because they only happen as part of refactoring other methods when your tests are already passing).
2. encapsulation is not particularly helpful and it's fine to just make things public if you want to add tests for new behavior within previously private methods.
The former is the classical position: classes are a black box except for their public API, and it's that public API that you test-drive.
The latter is an increasingly popular position that has gained traction as people rethink OOP, start to use languages that don't have "private", or start working in a more FP style. Python doesn't really have private methods (sure, you can use a double underscore prefix to "hide" a function but it's still accessible via the munged name which is '_TheClass__theFunction' for '__theFunction' inside 'TheClass'). Groovy has a 'private' keyword (for compatibility with Java) but completely ignores it. Both languages operate on trust and assume developers aren't idiots and aren't malicious. In FP, there's a tendency toward making everything public because there are fewer side-effects and some helper function you've created to help implement an API function might be useful to users of your code - and it's safe when it has no side-effects!
When I started writing Clojure, coming from a background of C++, Java, and CFML, I was quite meticulous about private vs public... and in Clojure you can still easily access a "private" function by using its fully-qualified name, e.g., `#'some.namespace/private-function` rather than just `private-function` or `some.namespace/private-function`. Using `#'` bypasses the access check. And the idiom in Clojure is generally to just make everything public anyway, possibly dividing code into a "public API" namespace and one or more "implementation" namespaces. The latter contain public functions that are only intended to be used by the former - but, again, the culture of trust means that users _can_ call the implementation functions if they want, on the understanding that an implementation namespace might change (and is likely to be undocumented).
My current position tends to be that if I'm TDD-ing code and want to refactor a function, I'll usually create a new test for the specifics of the helper I want to introduce, and then refactor into the helper to make all the tests pass (the original tests for the existing public function and the new test(s) for the helper function). Only if there's a specific reason for a helper to be private would I go that route (for example, it isn't a "complete" function on its own, or it manages side-effects that I don't want messed with outside of the calling function, etc). And, to be honest, in those cases, I'd probably just make it a local function inside the original calling function if it was that critical to hide it.
Google for `encapsulation harmful` and you'll see there's quite a body of opinion that "private by default" - long held to be a worthy goal in OOP - is an impediment to good software design these days (getters and setters considered harmful is another opinion you'll find out there).
That was longer than I intended!
Yeah Sean but it was bloody good. It all makes sense, and also goes a way to make me think I'm not a lunatic (at least not in this specific context).
And I have a bunch of reading to do on this whole "encapsulation harmful" idea. I'd heard it mentioned, screwed my nose up a bit, but didn't follow up. Now I will. Well: when I have a moment.
Anyway, there you go. Cheers for the effort put in to writing this, Sean. I owe you a beer or two.