[rspec-users] unusual challenges speccing external software

Giles Bowkett gilesb at gmail.com
Thu Jan 17 02:20:22 EST 2008

This is something I've come up against twice. First, I have an
interactive editor which calls vi, emacs, or TextMate (etc.) via Unix,
and then loads the file handed to the editor back into IRB after it's
edited and saved. I guess that makes it a special case of a more
general question, which is how do you spec external processes, with a
slightly more complicated version of the standard answer, mocks.

The second instance is more complicated. I'm writing a MIDI code
generator and using it to drive Propellerhead Reason, which is
third-party closed-source commercial software which will run on my OS
X box but which is not actually transparent to Unix in any meaningful
way that I'm aware of. (The code generator can very probably also
drive any arbitrary MIDI consumer.) I'm using code from Topher Cyll's
book "Practical Ruby Projects" to create Ruby wrappers to C methods in
Apple's CoreMIDI library. The trap here is I'm calling those C methods
to generate MIDI elsewhere in the OS in some unknown way. That makes
speccing weird.

I've been able to refactor this code in a kind-of spec-first way. I
write scripts which use the API successfully and actually generate
music I can hear. Then I change the internals of the API and run the
scripts again to ensure they still generate the same music. In that
sense I'm basically doing the right thing, even though it's not
particularly systematic or automated. I'm comparing the output with my
ears rather than in a way that code can grok, but it still has the
feel of a spec-first workflow.

For developing the API itself, especially the parts which generate
MIDI, I set up a layer of abstraction. Instead of directly driving the
code which sends MIDI to external software, the code which generates
the music drives an output adapter, and the output adapter sends MIDI
to external software - or, in the case of the TextOutputAdapter, just
text-dumps the music. Since it all comes back as text, it's very easy
to write specs against.

But I've ended up with three artifacts that are *either* byproducts of
bad design *or* bad products of a compromised spec process. I created
three mock *classes* (as opposed to mock objects). For example, one is
the ArrayGenerator. In this system Generators return notes. The
ArrayGenerator "generates" an array of notes. It's a silly
implementation, but it's useful if you want to spec that some code can
take an arbitrary generator, which generates arbitrary notes, and then
process those notes correctly.

Sometimes I think my design has too many abstract parts. Sometimes I'm
happy with it. I want to improve my speccing approach so I can operate
with greater certainty. I think probably this approach is a good
direction, and that I should change the specs by making them more
specific. (For instance, there are places where I check that a
Generator subclass can do stuff, even though Generator itself isn't
fully specced.) I'm not quite sure though.

Giles Bowkett

Podcast: http://hollywoodgrit.blogspot.com
Blog: http://gilesbowkett.blogspot.com
Portfolio: http://www.gilesgoatboy.org
Tumblelog: http://giles.tumblr.com

More information about the rspec-users mailing list