Feature #10394: An instance method on Enumerator that evaluates the block under with self being the block variable.

• Status: Feedback
• Priority: Normal
• Assignee: -
• Target version: -

[sawa (Tsuyoshi Sawada)]

Background

There has been desire to omit the | | and the explicit receiver in a block used with an enumerator or an enumerable. Currently, when the content of the block is a single method that takes no argument, symbol-to-proc is used with the & syntax so that:

["foo", "bar"].map{|s| s.upcase}

can be written as:

["foo", "bar"].map(&:upcase)

There has repeated been proposals (#8987, #9076, #10318) that express this desire to do this even when the block involves a method chain or a method with arguments like the following:

["foo", "bar"].map{|s| s.concat("ber")}
["  foo ", "\tbar\n"].map{|s| s.strip.upcase}

Focus has been on modifying how a block is passed to the enumerable/enumerator, and there has not been consensus on how the syntax should be.

Proposal

Unlike the earlier proposals, I suggest that there should be an instance method on Enumerator, let's say Enumerator#as_self, that evaluates the block each time with self being the block variable that would be passed otherwise. With such method, the cases above would be written like this:

["foo", "bar"].map.as_self{concat("ber")}
["  foo ", "\tbar\n"].map.as_self{strip.upcase}

This adds no modification to the syntax, it just requires a new method Enumerator#as_self to be implemented. I consider this method being along the lines of Enumerator#with_index, Enumerator#with_object; it intervenes between an enumerator (related to a block-taking method) and a block, and let the block-taking method work in a modified way.

It resembles instance_eval, but is different in that it assigns to self what would be a block variable (which changes for each iteration), instead of assigning the receiver.

[matz (Yukihiro Matsumoto)]

• Status changed from Open to Feedback

I like the idea itself. But I don't think as_self is a good name.
Any other name proposal? Anyone?

Matz.

[gogotanaka (Kazuki Tanaka)]

Sounds good for me.
(Actually I face the difficulties to implement #10318 ... and it's painful)

I just come up with...

#as
#as_arg(s)
#through
#member
.... ; (

And I'd be happy to call method which needs more than 2 args, if it could be.

just idea.

[sawa (Tsuyoshi Sawada)]

If the name is less than five characters, then it would need less typing than the original form, (i.e., five-letter name .xxxxx{foo} would require as much typing as {|x| x.foo}), so there would be more motivation for using this method. It may be good to use some preposition:

["foo", "bar"].map.by{upcase}
["foo", "bar"].map.by{concat("ber")}
["  foo ", "\tbar\n"].map.by{strip.upcase}

If one wants to go with a non-letter method name, | may be a candidate:

["foo", "bar"].map.|{upcase}
["foo", "bar"].map.|{concat("ber")}
["  foo ", "\tbar\n"].map.|{strip.upcase}

The | character here is reminiscent of the unix pipe, which may be interpreted here as passing the values to the block, and it also resembles the || notation in the block. But I don't know if people might think that ugly. & would keep resemblance to the notation using & with symbol_to_proc, but I don't feel the necessity to do so:

["foo", "bar"].map.&{upcase}
["foo", "bar"].map.&{concat("ber")}
["  foo ", "\tbar\n"].map.&{strip.upcase}

Even an asterisk or caret may work.

["foo", "bar"].map.*{upcase}
["foo", "bar"].map.*{concat("ber")}
["  foo ", "\tbar\n"].map.*{strip.upcase}

["foo", "bar"].map.^{upcase}
["foo", "bar"].map.^{concat("ber")}
["  foo ", "\tbar\n"].map.^{strip.upcase}

[avit (Andrew Vit)]

It might be confusing if such a thing only exists for Enumerator blocks and nothing else.

["foo", "bar"].map.as_self { clear }
["foo", "bar"].tap.as_self { clear } # (not an Enumerator)

What would be the correct receiver in your proposal for the following example? What happens when the method is not defined in the block's "self" object?

@members = ["foo", "bar"]
def transform_all
  @members.each.as_self { transform }
end
def transform
  raise "this one?"
end

A little bit related: #10095 (for the proposed syntax "as" vs. "as_self")

[nobu (Nobuyoshi Nakada)]

Isn't it instance_eval?