From: Daniel Berger <djberg96@...>
Date: 2007-02-15T00:05:05+09:00
Subject: Re: Question about rb_cvar_set

On Feb 13, 7:17 pm, "George Ogata" <george.og...@gmail.com> wrote:
> On 2/14/07, Daniel Berger <djber...@gmail.com> wrote:
>
> > Hi,
>
> > Ruby 1.8.5 p12
>
> > It looks like there was a prototype change in rb_cvar_set() at some
> > point in the 1.8.x branch. It now takes a 4th parameter (an int). What
> > is the 4th parameter supposed to be?
>
> Hi Dan,
>
> If you poke your nose into the rb_cvar_set() function, you'll see it's
> just a warning enabler flag.  Specifically, if this argument is
> nonzero, then in verbose mode (-v, -w) you'll get a warning if you
> modify a class variable from a subclass.  As in:
>
> $ ruby -w -e 'class B; @@x = 1; end; class D < B; @@x = 5; end'
> -e:1: warning: already initialized class variable @@x
>
> The flag appears to be used to distinguish between cases such as:
>
>   class B
>     @@x = 1
>   end
>
>   class D < B
>     @@x = 2    # warning
>
>     def f
>       @@x = 3  # no warning
>     end
>
>     def self.f
>       @@x = 4  # no warning
>     end
>   end
>
> Here, although the last 3 assignments to @@x all modify B's @@x, the
> ones inside a def are considered "assignment-style", whereas the "@@x
> = 2" is "declaration-style".  Only the latter generates a warning.

Ah, thanks George. It looks like you're supposed to test for the
RB_CVAR_SET_4ARGS macro to stay backwards compatible, too.

Dan