This post originated from an RSS feed registered with Ruby Buzz
by Chad Fowler.
Original Post: Type Watching
Feed Title: ChadFowler.com
Feed URL: http://feeds2.feedburner.com/Chadfowlercom
Feed Description: Best practices, worst practices, and some generally obvious stuff about programming.
Dave has posted a clever piece of code,
showing how you could do runtime class checking in Ruby (or a Ruby-like
language) by simply wrapping methods and recording the first class that is
used for each method parameter. In his example, he uses it to throw an
error if any subsequent invocation has a parameter of a class different
from the first usage of that parameter. So, the following would fail:
bob.age = 32
bob.age = Duck.new
Though he warns that this is just a "coffee-break
implementation", the actual code is remarkably simple as well. I
consider this to be a hallmark of Dave’s.
As anyone in the dynamic programming languages world knows, "type
che
1000
cking"(1) is a touchy and usually unproductive subject (RubyTypesAndClasses).
The common complaint is that, without "type" information for
method parameters, code is hard to understand and maintain. I don’t
generally find this to be true for myself, but I can see and understand
that sometimes it would be helpful to have some hints as to what kinds of
objects a library author passes into certain method calls.
The simplest solution to this—which wouldn’t require a
compromise of the design philosophy of the language—would be for all
library authors to simply document their code, giving their users a hint as
to what each method expects. We’ve all heard that one before. Even
those of us with the best intentions do, on average, a poor job documenting
our own code. Additionally, what’s obvious to me may not be obvious
to you. So as a library author, I will probably either overdocument (waste
my own time and clutter the code) or underdocument.
Taking a cue from Dave, I did my own coffee-break implementation(2) of a
hack that would auto-document the class of parameters for a given set of
methods. It’s implemented as a modification to rdoc. The idea is that you can
"instrument" your code temporarily, run through a typical usage
of it, and have Ruby keep track of—and save—class information
for the method parameters of specified classes.
An example
I’ve written a servlet (which I’ll talk about in more detail in
a later post) that I would like to document via rdoc. Assuming the servlet
is called "ContinuationServlet", I can auto-generate parameter
class hints for this servlet by adding the following to my code:
require 'class_hints'
wrap_methods(ContinuationServlet)
...
(invoke webrick normally here)
...
File.open("seasidehints.out", "w") do |file|
file.write(Marshal::dump(Kernel::METHOD_PARAM_HINTS))
end
This will install a wrapper around every method of ContinuationServlet,
which will save class information for each parameter in the nested hash,
Kernel::METHOD_PARAM_HINTS. After having clicked around through the
application with a web browser, we just dump this to disk as a marshalled
Ruby object.
Like Dave, I’m just throwing this up here for comment. I’m not
sure anyone will find it useful, and the implementation is definitely a
hack. But, it seems like something +like this+ might help to quiet some of
the complaints that half the population seems to have about dynamically
typed languages. And, for example, could FreeRIDE use a trick like this to
provide intellisense-like hints in the IDE?
(1) The word "type" is frequ
577
ently misused in the Ruby world to
mean "class". Matz actually deprecated the #type message on
Object to clarify that a Ruby object’s "type" is more
focused on its capabilities than its class. Class is little more than a
convenience mechanism.