A fellow rubyist, Zach Church, wrote a great article on coercion in
Ruby. It’s not something I’ve seen covered often so I thought I’d
share with the community at large:
http://mutuallyhuman.com/blog/2011/01/25/class-coercion-in-ruby
Zach D.
On Wed, Jan 26, 2011 at 8:58 AM, zdennis [email protected] wrote:
A fellow rubyist, Zach Church, wrote a great article on coercion in
Ruby. It’s not something I’ve seen covered often so I thought I’d
share with the community at large:http://mutuallyhuman.com/blog/2011/01/25/class-coercion-in-ruby
Zach D.
Thanks for sharing, very well done.
On Wed, Jan 26, 2011 at 3:58 PM, zdennis [email protected] wrote:
A fellow rubyist, Zach Church, wrote a great article on coercion in
Ruby. It’s not something I’ve seen covered often so I thought I’d
share with the community at large:http://mutuallyhuman.com/blog/2011/01/25/class-coercion-in-ruby
In case you weren’t aware: I covered this thoroughly a while ago when
I described what it needs to create a numeric class which nicely plays
along with other numeric classes.
http://blog.rubybestpractices.com/posts/rklemme/019-Complete_Numeric_Class.html
Cheers
robert
Just a small nitpick:
if other.is_a? TimeInterval
TimeInterval.new(value + other.value)
elsif other.is_a? Numeric
TimeInterval.new(value + other)
else
raise TypeError, “#{other.class} can’t be coerced into TimeInterval”
end
would probably be better written as:
case other
when TimeInterval
TimeInterval.new(value + other.value)
when Numeric
TimeInterval.new(value + other)
else raise TypeError, “#{other.class} can’t be coerced into
TimeInterval”
end
On Wednesday, January 26, 2011 02:52:12 pm Tony A. wrote:
would probably be better written as:
case other
when TimeInterval
TimeInterval.new(value + other.value)
when Numeric
TimeInterval.new(value + other)
else raise TypeError, “#{other.class} can’t be coerced into TimeInterval”
end
Or at least:
if other.kind_of? TimeInterval
…
elsif other.kind_of? Numeric
…
Remember duck typing? I don’t care if it is_a Numeric, not really. I
can’t
really think of many cases where is_a? makes more sense than kind_of?.
On Wed, Jan 26, 2011 at 9:20 PM, David M. [email protected]
wrote:
Or at least:
if other.kind_of? TimeInterval
…
elsif other.kind_of? Numeric
…Remember duck typing?
Good point, but that said, it can be done even better:
First, in TimeInterval, alias_method :to_int, :value
Then when implementing +:
if other.respond_to? :to_int
TimeInterval.new(value + other.to_int)
elsif other.respond_to? :coerce
a, b = other.coerce(self)
a + b
else raise TypeError, “#{other.class} can’t be coerced into
TimeInterval”
end
Thanks Robert, I hadn’t seen your post before. Will add to my reader
docket!
Zach
On Wed, Jan 26, 2011 at 10:29 AM, Robert K.
Disclaimer: The length of this post in particular has nothing whatsoever
to do
with how strong or well-informed I am about the topic. I actually didn’t
know
about coercion until today, so I may be missing something obvious.
Proceed
with caution…
On Wednesday, January 26, 2011 10:52:25 pm Tony A. wrote:
Remember duck typing?
Good point,
Well, to a point. I just reflexively change is_a? to kind_of? everywhere
I see
it, unless there’s a good reason not to. I don’t have a problem with
your
cases, just pointing out an alternative.
but that said, it can be done even better:
First, in TimeInterval, alias_method :to_int, :value
If this article is correct:
http://www.rubyfleebie.com/to_i-vs-to_int/
…then I’m not sure that’s the case. Do we always want to represent a
time
interval as an integer number of seconds? Maybe, but my instinct is that
maybe
there are other integers we could get out of this object.
if other.respond_to? :to_int
TimeInterval.new(value + other.to_int)
elsif other.respond_to? :coerce
a, b = other.coerce(self)
a + b
else raise TypeError, “#{other.class} can’t be coerced into TimeInterval”
end
While I agree that to_int is the correct way to check if something’s an
integer, the original check was for anything numeric, so it might not be
sufficient (though it’s probably fine for this class so far).
So maybe something like:
if other.respond_to?(:to_int) || other.kind_of?(Numeric)
TimeInterval.new(value + other.to_int)
elsif other.respond_to? :coerce
begin
a, b = other.coerce(self)
a + b
rescue TypeError
if other.respond_to?(:to_i)
self + other.to_i
else
raise
end
end
elsif other.respond_to?(:to_i)
self + other.to_i
else
raise TypeError, “#{other.class} can’t be coerced into TimeInterval”
end
That could probably be DRYed up a bit.
The idea is, if something has a to_int or is a Numeric of some sort,
it’s
claiming that it is just a number, so we’re probably free to treat it
as
such. Otherwise, let it try to coerce first before we fall back to
tricks like
to_i.
Why?
It’s a contrived example to begin with, so my answer will be just as
contrived, but… Suppose I create, say, an HourInterval class, which
represents time intervals of hours. Then:
HourInterval.new(5)
You’d expect that to be five hours, right? Then you’d also expect:
HourInterval.new(5).to_i
…that should return 5. However, if we just let it to_i (or to_int, if
it’s
pretending to be just a number), then TimeInterval will treat it as five
seconds instead of five hours, no matter what we do. That’s why I’d
suggest
giving it a chance to coerce first, and saving the fuzziest conversions
for
when that fails.
Now the only trick is to make sure that we never call ‘coerce’ from
inside a
‘coerce’ method, unless I’m missing something.
I’m told exception handling is expensive. I wonder if it might make
sense to
have a non-raising form of ‘coerce’, or a can_coerce? method, to avoid
that
overhead. But then, if you’re doing this sort of thing and you start
caring
about performance, you can always manually convert things ahead of time.
On Thursday, January 27, 2011 10:19:18 am Adam P. wrote:
Perhaps you were thinking of instance_of? here?
Apparently so. Sorry for the noise…
On Thu, Jan 27, 2011 at 4:20 AM, David M. [email protected]
wrote:
Am I missing something here? I don’t see how this is in keeping with the
idea of duck typing: is_a? is an alias for kind_of? and vice-versa.
http://www.ruby-doc.org/core/classes/Object.html#M001034
Perhaps you were thinking of instance_of? here?