A bit late to the party, I took a programming challenge from Mike Taylor :
Are you one of the 10% of programmers who can write a binary search?
The 10% is a reference to an essay by Jon Bentley, who found that only
10% of the programmers asked could write a binary search correctly given a
description of the algorithm. The challenge from Mike was to write it without
any TESTING whatsoever.
Here is my answer in Ruby:
Pretty standard. This was about 1/2 an hour from start to finish; mostly due to
not being able to run the code1 and obsessing over whether I had handled all
the edge cases, which is related to my motivation for taking the challenge.
Aside from being a fun exercise, the comments Mike received led him to write
two more blog posts!
The third in the series, Testing is not a substitute for thinking
has some interesting and keen observations. I agree with most of them.
However, the conclusion is absolutely, categorically wrong.
Bearing in mind that binary search is in fact a pretty
simple algorithm, it’s likely true that you could have
bashed your way through to a working implementation after
a few iterations of coding and testing. But –
Your code would be less clear
You wouldn’t understand clearly why it works
You would find it harder to extend or modify in future
Inflammatory stuff. And totally wrong.
A few paragraphs earlier, Mike lays out why he thinks this:
By using skills that are not often thought about
in these days of test-driven kool-ade, reflexive
application of design patterns and automatic refactoring
tools. Once more, let me be clear that all these things
are good so far as they go — really, I agree with you,
they are! — but they are no substitute for actually
thinking, and thinking is what’s needed for hard problems.
Indeed thinking is what is needed for hard problems. What Mike has failed to
acknowledge is that TDD is not a tool for writing code, TDD is a tool for
As an example of this, in the spirit of Mike’s challenge, I expressly wrote the
binary search code without test-driving it. However, instead of writing the
code directly from the description, I turned the description into a set of
constraints. As each of those constraints became clear, I wrote code. And in
the end I had a working binary search2.
So, am I saying Mike is an idiot? Absolutely not! After hearing more and more
critiques (both thoughtful and thoughtless) of TDD, I am coming around to the
conclusion that there is a problem. The problem is most people learn TDD
from a book or a blog or even from a vague description and some positive can-do
attitude (Commendable! Really!)
What has become apparent is that this will not work for most developers. What
does work is deliberate study and mentoring, ideally by pairing. A mentor who
is showing you more than just the syntactic mechanics of writing test code and
having it invoke the code under test. Someone who is explaining how tests are a
tool for understanding and solving the problem.
Folks interested in Software Craftsmanship have realized this truth about all
aspects of software development: you learn better from a mentor and from
TDD is no exception.
1 Sort-of, according to Mike’s rules I could have shaken out syntax errors
by running the code; in the case of Ruby, that felt like cheating, so I just
stared at it.
2 After the fact, I threw in a test scaffold to verify that it does work,
and that it is log(2)N.