Using positional arguments for more than two or so things creates connascence of position, which quickly becomes annoying.
Using block based DSLs so that you can break things down into small functions that each do their part is a good way around this, but these tend to be hard to extend dynamically.
Typically well designed Ruby systems provide both: Pseudo keyword arguments for dynamic needs, and some DSL-type syntax for pleasing the eyes.
Yes, it sucks that we don't have real keyword arguments. But no, it's not nearly as bad as you think.
positional arguments for more than two or so things creates connascence of position
Which is precisely the reason that you shouldn't pass more than two arguments to a method (unless you are creating a list or hash for their intended purpose). If you have more than two arguments to a method, it's a sign that you should reify them into an object.
Using block based DSLs so that you can break things down into small functions that each do their part is a good way around this
Why do people love making things complicated by misusing blocks. Just create an object and set slots on it. Blocks are best used for delaying computation, not setting up state.
I think that Zed Shaw's essay on the "Master" and the "Expert" is applicable here: http://www.zedshaw.com/essays/master_and_expert.html
Context is king, and what you've done is taken a single point from my book, ripped out its context, strawmanned it to death, and used it to bolster you're own "Top 100 rails" status.
I encourage readers to actually see what I had to say in the book by downloading the PDF before taking these arguments too seriously. If there are some constructive examples to be shown that seem promising and pass peer review, they'll definitely make it into the open source version.
It might even be worth it to write up a section describing the tradeoffs that richcollins has pointed out, but it's definitely not something as universal as he makes it sound.
I've mostly stopped using Rails and Ruby in favor of Io, so I have no interest in bolstering my Rails status. I was just providing evidence to counter your assertion that I was a Ruby newb.
The notion that you can follow singular design principles in all situations is absurd, if you ask me.
I agree that "it depends", but using Hashes as arguments has become idiomatic Ruby, which is unfortunate in most cases.
I still would really like to see examples of well designed open source Ruby projects that follow what you consider to be decent design principles. I think that'd make a much stronger case for your argument, and might open my eyes to something that's been in a blind spot.
Personally, I feel like a certain amount of API design is bound to what consumers will expect. We can certainly stretch and shift their tendencies, but if we fly in the face of them, our perfection in a vacuum will never be appreciated.
Until then, stop acting like you're some sort of authority (NOTE: This is not to claim I am -- because I'm not and no one is.) I just think it'd make sense to put your ideas into a bit of context so people can see that you're not just being critical for the sake of being critical.
#from your book
def distance(*points)
case(points.length)
when 2
x1,y1,x2,y2 = points.flatten
when 4
x1,y1,x2,y2 = points
else
raise ArgumentError, "Points may be specified as [x1,y1], [x2,y2] or x1,y1,x2,y2"
end
Math.hypot(x2 - x1, y2 - y1)
end
puts distance(1, 2, 5, 6)
puts distance([1, 2], [5, 6])
# why would you choose to make a complicated interface that requires a complicated implementation?
# simple, clear interfaces that allow for simple implementations
# are almost always preferrable to complicated ones that save a
# keystrokes through syntax
class Point
attr_accessor :x
attr_accessor :y
def initialize
self.x = 0
self.y = 0
end
def self.with(x, y)
new.set(x, y)
end
def set(x, y)
self.x = x
self.y = y
self
end
def distance_to(a_point)
Math.hypot(a_point.x - x, a_point.y - y)
end
end
puts Point.with(1, 2).distance_to(Point.with(5, 6))Point.with(1, 2).distance_to(Point.with(5, 6))
Over this?
distance [1,2], [5,6]
I guess it depends on how many interesting things you wanted to do with points. As things get more complicated, maybe something like this is worthwhile, but I don't feel that making everything an explicit object is categorically better.
But the section you're quoting is among the most contrived examples that RBP gives, it's mainly just part of a list of the different sorts of argument processing Ruby offers and the trade-offs betweeen them.
Maybe you skimmed and missed that, or maybe you read through and chose to ignore that detail.
0. It breaks the fundamental OO convention and doesn't encapsulate logic and data.
1. It's not clear what it means unless you look up the docs.
2. It's implementation is complex, making it more difficult to understand and maintain
3. It fails to separate concerns, making it less extensible/modifiable. e.g. What if you wanted to override the setting behavior? You couldn't do it in one place.
Remember you don't need to destructure the hash. You've got a local one initialized to sensible defaults, and you .merge! it with the param one, and then just use the hash values as your local vars.
I think the central issue is that there is too much friction to creating new classes in Ruby. It feels official and final. In Io, you just clone an existing object like you would send any other message.
I'd rather start with a hash and then see if it later demands more, than start inventing objects that, based on existing use cases, need only be hashes or structs.
I'd also rather go through a number of pre-1.0 iterations to get real-world usage to shake out the painful API parts. Up until then I'm happy to break compatibility if experience says there's some bad code in there.
Monkeybars, for example, had some methods that began life using positional args, but they were changed to use hash values. I hated it.
Aside from the extra typing required for every call, with positional args I got used to reading code and seeing things in a fixed order, so when I had to use the method myself I knew exactly what needed to go where. (These methods had 2 arguments, with maybe an optional third or so.)
With the hash args the order could vary so I had to keep looking up what the possible argument key values could be.
(Thing about using hashes for args: with positional arguments you need only know what role an argument is playing in order to know what sort of value to pass in, but with a hash you have to know the exact key name. That ends up being more stuff to look up in the docs, especially if the API designer has a quirky way of naming things.)
You might think that at least for a code reader the use of hash keys would better document the call, but the method name and the names of items being passed in were almost always sufficient.
I suspect this particular change was a result of over-thinking these methods and expected usage instead of actually paying attention to what the code was doing right now, and how people were using it right now.
There are times when my gut tells me, "You know what's going to happen with this code later", and I try to take that into consideration, but (usually) simple wins for me.
You can set slots on objects in any order that you want to:
Foo.new.set_x(x).set_z(z) ...It would seem you only see the surface beauty of things. [redacted -- good point]
Is that how you defend an API? Insult the user?
Holy shit that's stupid.
def to_hash
{:x => @x, :y => @y}
end
def to_ary
[@x,@y]
end
def [] i
case i
when 0,:x
@x
when 1,:y
@y
end
end
Should work if the receiving method interprets the arguments properly.To be clear, I think that it is great that you created this book. I just responded as I did because I'm often frustrated when I use Ruby APIs that sacrifice good design (imo) for syntactic terseness.
If RBP becomes "Gregory Brown's Best Practices", it'll be epic fail. I am encouraging folks to prove whatever they can wrong in this book.