I've always thought the chaining methods thing that jQuery popularized was a pristine example of an anti-pattern. It might save a few keystrokes, but I question whether anyone really writes code like this the first time through. I suspect people actually take perfectly fine working code and refactoring it into a chain just to make the code appear more clever, all the while making the code harder to read and debug.
I like chaining because it emphasizes that a set of method calls is part of one discrete action involving one object.
For example, take adding a header to some text: it's one action, but it involves multiple steps. First I create the actual element, then I give it a class name, add some text and finally insert it in the DOM. I think chaining not only makes the code easier to read (and type, although that isn't particularly important), but also makes its semantics clearer.
I applaud people for trying to find clean ways to add chaining to langauges like JS, but feel we are working at the wrong level of abstraction. Chaining is a syntax and grammer problem, not a semantic problem. Smalltalk solved this by making chaining part of the syntax. if you write:
foo
:bar;
:bash something;
:thenBlitz.
The semi-colons tell the interpreter to chain the method calls to the original receiver. That way you can read the sntax and know chaining is involved without needing to know whether those methods were written to be chained. The return value of a method is its semantics, and should not be chosen to try to implement syntax.
Summary: This problem should be solved by the parser, not by the function author.
p.s. paging @jashkenas!
p.p.s. Did you ever take that test yourself? Yes, my comments apply doubly to libraries like #andand.
Personally, I'm a bit iffy on the subject, because although the demonstrated usefulness of the idiom is evident in libraries like jQuery, chaining-as-a-language-feature is inherently focused on stringing together a sequence of imperative side effects, where it can be nicer to use expressions for their values.
For example, where the returned value is a part of the next step of the computation, chaining comes naturally:
list
.concat(other)
.map((x) -> x * x)
.filter((x) -> x % 2 is 0)
.reverse()
And where you need built-in language chaining to do the job for you, it's almost always because you're ignoring the return value, and using each method for its side effects:
So the enforced imperativeness of chaining syntax makes it feel a bit off. Still, if you were designing such a syntax from scratch, and not following Smalltalk's lead, what would you think it should look like?
The first “addClass” and “find” are both sent to “added,” “when” is sent to the result of “find”, “removeClass” and the second “addClass” are both sent to the result of “when.” It feels to me very much like how we indent scope and syntactic blocks like “if” or “unless."
Looks very nice! Though the discrimination between intending or not is done based on whether the previous action mutates the object (no new intend) or creates or returns different object (new intend), right? This is something that can't be known on the language level unless there's a specific syntax for it, which doesn't seem to be possible for Javascript based languages (e.g. https://github.com/jashkenas/coffee-script/issues/1738).
My point is that if I owned the parser, I would ignore what the function returns and send the message to the previous receiver when multiple message inocations are indented in parallel. Here's an example using functions that both return a value and have a side-effect:
['foo', ['bar']]
.pop()
.pop()
Would return 'foo' if I owned the parser, while:
['foo', ['bar']]
.pop()
.pop()
...would return 'bar'. You can infer from his what happens to the niial array in terms of side effects. My suggestion is that "chaining" method calls is a syntax issue and not a function issue, and that writing functions tonreturn a certain thing just to cater to how you like to write programs is hacking around a missing language feature.
pop() is a great example. Why shouldn't pop return the receiver? The first example makes sense, you ought to be able to chain calls to pop(). Then again, why shouldn't pop() return what it pops? You ought to be able to pop something and use it, just like the second example.
Baking chaining into what functions return forces the function author to choose On behalf of the function user. I do not blame people for doing this in a language lacking thenproper syntax, but given how much people do this, it seems worth considering for people writing new langauges or new syntaxes for existing languages.
"chaining-as-a-language-feature is inherently focused on stringing together a sequence of imperative side effects, where it can be nicer to use expressions for their values."
Out of interest, what do you mean by "it can be nicer to use expressions for their values"? Any example of that?
This website (much like all of Lea's work) has loads of really nice details – worth reading through the source code for anyone who works with CSS. Particularly the use of border-image, though there are a number of other nice little things going on here!
15 comments
[ 2.3 ms ] story [ 41.9 ms ] threadAlso, my own node-ContextChain https://github.com/duncanbeevers/node-ContextChain
Both of these libraries take the approach of wrapping the underlying object rather than modifying it.
For example, take adding a header to some text: it's one action, but it involves multiple steps. First I create the actual element, then I give it a class name, add some text and finally insert it in the DOM. I think chaining not only makes the code easier to read (and type, although that isn't particularly important), but also makes its semantics clearer.
Summary: This problem should be solved by the parser, not by the function author.
p.s. paging @jashkenas!
p.p.s. Did you ever take that test yourself? Yes, my comments apply doubly to libraries like #andand.
A dedicated syntax for chaining has been discussed several times for CoffeeScript, with different flavors:
https://github.com/jashkenas/coffee-script/issues/1495
https://github.com/jashkenas/coffee-script/issues/1291
https://github.com/jashkenas/coffee-script/issues/281
https://github.com/jashkenas/coffee-script/issues/1431
Personally, I'm a bit iffy on the subject, because although the demonstrated usefulness of the idiom is evident in libraries like jQuery, chaining-as-a-language-feature is inherently focused on stringing together a sequence of imperative side effects, where it can be nicer to use expressions for their values.
For example, where the returned value is a part of the next step of the computation, chaining comes naturally:
And where you need built-in language chaining to do the job for you, it's almost always because you're ignoring the return value, and using each method for its side effects: So the enforced imperativeness of chaining syntax makes it feel a bit off. Still, if you were designing such a syntax from scratch, and not following Smalltalk's lead, what would you think it should look like?pop() is a great example. Why shouldn't pop return the receiver? The first example makes sense, you ought to be able to chain calls to pop(). Then again, why shouldn't pop() return what it pops? You ought to be able to pop something and use it, just like the second example.
Baking chaining into what functions return forces the function author to choose On behalf of the function user. I do not blame people for doing this in a language lacking thenproper syntax, but given how much people do this, it seems worth considering for people writing new langauges or new syntaxes for existing languages.
Out of interest, what do you mean by "it can be nicer to use expressions for their values"? Any example of that?
Cheers!