Feature #2013: [PATCH] a = *b calls b.*@

• Status: Rejected
• Assignee: matz (Yukihiro Matsumoto)
• Target version: 3.0

=begin
This makes the * operator operate more similarly to the + and - operators. The binary versions of + and - call methods named + and -, while the unary versions call methods named +@ and -@. The binary * operator calls the method named *, but the unary * operator calls the method to_a currently, and used to call to_ary in 1.8.6 (and at some point may have called to_splat).

I think this makes for more consistent behavior, and hopefully it isn't just a foolish consistency. I brought up this idea as a question in Shugo Maeda's presentation at RubyKaigi 2009, and discussed it with Matz at Lone Star Ruby Conf 2009, and he thought the idea had merit.

Here's a basic example how this would look:

class MultiplePersonality
def *@
[self, [self, self]]
end
end

mp = MultiplePersonality.new
p1, mp2 = *mp

This comes with a patch that appears to work from my simple testing, but this is my first time working with the internal ruby code, so I apologize in advance if it doesn't do things correctly.

The patch modifies parse.y so the above is no longer a syntax error. It adds a USTAR token to the parser to represent that *@ token. It modifies the splatting to call *@ instead of to_a. Also, for backwards compatibility, it adds *@ to BasicObject, and has it call to_a if it responds to to_a. This allows code that defines to_a and expects that the unary * operator will call to_a to still work.

This patch is mostly for consistency, but it also allows the programmer to make to_a return one thing, and the unary * operator return something else. I can think of the following use case:

Represents an abstract set of rows in the database

class Dataset
def to_a
retrieve_database_rows
end
def *@
[self]
end
end
dataset = Dataset.new

Explicitly asking for an array means I want

an array of database rows represented by the

dataset.

rows = dataset.to_a

If the dataset had any rows, I want to debug print

each row separately. However, if it did not have

any rows, I want to debug print the dataset itself.

p(*(rows.empty? ? dataset : rows))

I'd like to thank Eleanor McHugh for helping me find the key part of ruby that needed to be modified to support this (in vm_insnhelper.c).
=end

[marcandre (Marc-Andre Lafortune)]

• Category set to core
• Assignee set to matz (Yukihiro Matsumoto)

=begin

=end

[mame (Yusuke Endoh)]

• Target version set to 3.0

=begin
Hi,

This makes the * operator operate more similarly to the + and - operators.

Interesting. It may be useful to simplify the semantics of multiple
assignment which is considered as one of most complex part of Ruby.

But it is very primitive part of Ruby semantics.
I think that such a spec change must not be included in 1.9.x.

So I set the target of this ticket to 2.0.

--
Yusuke Endoh [e-mail]
=end

[yugui (Yuki Sonoda)]

=begin
On Sun, Aug 30, 2009 at 2:53 PM, Jeremy Evans [e-mail] wrote:

I think this makes for more consistent behavior, and hopefully it isn't just a foolish consistency.  I brought up this idea as a question in Shugo Maeda's presentation at RubyKaigi 2009, and discussed it with Matz at Lone Star Ruby Conf 2009, and he thought the idea had merit.

+1
Interesting. But it is so large change for the language feature to add
to 1.9.2. I do not include this feature in 1.9.2.
But I also think it can be added to 1.9.x. What do you think, matz?
Do you want add it to ruby 1.9.x?

-- Yuki Sonoda (Yugui)

=end

[shyouhei (Shyouhei Urabe)]

• Status changed from Open to Assigned

=begin

=end

[naruse (Yui NARUSE)]

• Project changed from Ruby master to 14
• Category deleted (core)
• Target version deleted (3.0)

[naruse (Yui NARUSE)]

• Project changed from 14 to Ruby master

[mame (Yusuke Endoh)]

• Description updated (diff)
• Target version set to 3.0

[mame (Yusuke Endoh)]

• Status changed from Assigned to Rejected

I'm closing this ticket since the discussion has been stalled completely.

I still think this is an interesting proposal. If you (or anyone else) are interested in this feature, please create a new ticket. It would be good to reimplement it for trunk, and to investigate its merit on real-world applications and the compatibility problem.