From: Caleb Clausen on
On 7/19/10, Brian Candler <b.candler(a)pobox.com> wrote:
> Caleb Clausen wrote:
>> On 7/17/10, Gavin Sinclair <gsinclair(a)gmail.com> wrote:
>>>>
>>>> Your code runs without errors for me with ruby 1.9.1-p376.
>>>>
>>>> Stefano
>>>
>>> Thanks for testing it. I should have specified my version, of course:
>>>
>>> ruby 1.9.2dev (2010-07-02 revision 28524) [i386-cygwin]
>>
>> It seems like this could be a regression.
>
> Seconded. The program works OK for me with ruby 1.9.2dev (2009-07-18
> trunk 24186) [i686-linux]

Nobu says this is not a bug, and his explanation makes perfect sense
to me, now that I understand it. instance_eval passes self as a
parameter to the block in 1.8 and 1.9.2 but didn't in 1.9.1. Also,
lambda panics if it doesn't receive the expected number of params,
whereas proc does not.

See ruby-core:31336

From: Gavin Sinclair on
>>
>> It seems like this could be a regression.
>
> Seconded. The program works OK for me with ruby 1.9.2dev (2009-07-18
> trunk 24186) [i686-linux]

Data so far:
* works on 1.9.1-p376 [Caleb, Gavin]
* works on 1.9.2dev (2009-07-18) [Brian]
* doesn't work on 1.9.2dev (2010-07-02) [Gavin]

That agrees with Nobu's comment on ruby-core (which I should have
relayed here earlier, sorry):

No, it not the point. Now instance_eval yields the self.

This code (to boil the problematic case down) fails...

code = lambda { 42 }
instance_eval &code

...because instance_eval passes 'self' as a parameter to the block.
Lambdas are fussy about parameters being accounted for, hence the
ArgumentError. Procs are not fussy about parameters, so replace
'lambda' with 'proc' in the above code and it works.

For that matter, _this_ works too:

code = lambda { |s| 42 }
instance_eval &code

From what I gather:
* in 1.8.x, instance_eval passed self but it was an undocumented feature
and lambdas weren't fussy so it didn't matter
* in 1.9.1 instance_eval stopped passing self (don't know whether that change
was intentional)
* heading towards 1.9.2 it started passing self again, from about December 2009

For a (still?)-undocumented feature, the message "ArgumentError: 0 for
1" is not very helpful.

Gavin