in reply to Re^3: chaining method calls
in thread chaining method calls

$my_object_variable->foo($x)->bar($y)->ni($z);

[download]
The trouble is, that sort of construct already has a meaning and this isn't it. Chaining methods on different objects is common. Like this, for example: 
$car->wheels()->turn('right');

[download]
That kind of thing is common on most OO languages, and it's what I expect when I see chained method calls. This "return yourself" business foils my expectations.

Replies are listed 'Best First'.
Re: Re: Re^3: chaining method calls
by Ovid (Cardinal) on Jun 12, 2003 at 21:06 UTC

    The way I see this idiom used is fairly consistent. $self is only returned for mutators (upon success). Thus, the expectations are clear: mutators always return $self upon success and accessors never return $self. If accessors return an object, it's known the the object is not the calling object. That, actually, is yet another reason why I prefer to have separate accessors and mutators ('name' or 'get_name' vs. 'set_name'). This clear distinction makes the expectations clear. And again, if you don't like the chained features, you don't have to use it. Chained method calls are usually an extra feature as opposed to the replacement of a feature that is already used.

    I also note that you object to this idiom because it's not common. If it were a bad idiom, I would consider this to be a benefit. If it's a good idiom, then that's a negative. Which way it falls is still kind of up in the air, though. I'm glad you raised the issue.

    Cheers,
    Ovid

    New address of my CGI Course.
    Silence is Evil (feel free to copy and distribute widely - note copyright text)

[reply]
      The way I see this idiom used is fairly consistent. $self is only returned for mutators (upon success). Thus, the expectations are clear: mutators always return $self upon success and accessors never return $self.

      Consistent or not, because there's no first-class support for cascading in Perl, the reader is forced to do a lot of digging to distinguish a simulated cascade from sloppy coupling. Basically, to be sure, you have have to read all of the methods. In a first-class cascade (like those supported by Smalltalk) a surface read is sufficient.

[reply]

In Section Meditations