One thing I've learned from maintaining open-source projects is that
someone will always be unhappy no matter what you do. :-) The patch you've
suggested
improves upon the current state of the code and add extra flexibility, so
we probably shouldn't dwell too much over the details in it and simply
install it.


On 16 January 2014 20:40, Dmitry Gutov <dgutov@yandex.ru> wrote:

> On 16.01.2014 16:26, Bozhidar Batsov wrote:
>
>> By "current implementation" I meant the current patch. Yeah, I think
>> that some way to have defs align to the start of an expression only for
>> modifier expressions might be a good idea in the interest of maximum
>> flexibility.
>>
>
> They are plain method calls anyway. Like mentioned, `private', etc, are
> not keywords.
>
>
>  I meant to say that in
>> the current implementation of the feature in MRI assigning the result to
>> a variable doesn't make much sense, but in the future that might change
>> (if the return value becomes a method object instead of a symbol).
>>
>
> I don't think it's likely to change. Previously, the return value was
> undefined. Now that it's specified to be method name symbol, it probably
> won't ever change, in the interests of backward compatibility.
>
>
>  In Rubinius defs have always returned an object:
>>
>> def foo; end
>> => #<Rubinius::CompiledMethod foo file=(irb)>
>>
>
> Ah, I see. It's probably considered an implementation detail, though.
>
>
>  Anyways, I think that the current patch will be good enough for most
>> users. I was just pointing out some potential problems in case they were
>> overlooked by you.
>>
>
> Thanks, but the occasional need for more flexibility is often apparent.
> The question I'm usually trying to answer is whether we can get away with
> fewer options, and still keep users happy.
>