Nate the prolific Perl6 RFC producer and commentator wrote:
> I don't think suggesting that we eliminate globals altogether is a good
> or feasible idea. It's an unnecessary crutch. They are useful in many
> circumstances.
So here's my order of preference:
1) Eliminate dynamic variables entirely.
2) Eliminate dynamic variables from default Perl, but allow their use
with a pragma.
3) Provide a new strict pragma to disable dynamic variables. (Note
that I have an ulterior motive, here; strict is the de-facto standard
for module writing. If I get this into strict, then everybody will
feel obligated to write their modules in this style. :) Of course,
module writers are supposed to be good enough to not use unnecessary
crutches, right?)
CONTEXT ALERT: All of this is still optional to RFC 106, which can
provide for making lexical variables the default without having to do
anything to dynamic variables. Don't have a cow if you were worried
about that.
> I think this is shaping up to be a really good proposal. Here's what I
> hear:
>
> 1. All variables are liberally lexically scoped by
> default. Either scope them package-wide or do the
> nearest block idea.
>
> 2. You can explicitly scope variables with my or our.
>
> 3. Global variables can still be declared with 'use vars'.
>
> 4. Dynamic keywords (like local) still only work on
> global variables.
>
> Sound accurate?
Everything except #4. I explicitly believe in allowing dynamic
assignment on lexical variables. Way back before this list, when the
first version of RFC 6 came out, Tom Christiansen commented, "One
could even make Chip happy and allow one to C<scopetemp> [one of many
proposed new names for local()] a lexical." I'm not saying Tom was
behind the idea, but it is a possibility, and I'm behind it.
Apparently, Chip (the same one who started Topaz, I presume?) was,
too, at some earlier point in Perl history.
> Exactly how point #1 is implemented can be left till
> later, just as long as this:
>
> $x = 1;
> {
> $y = $x + 1;
> }
> print "$y";
>
> Works without any predeclarations.
Agreed. Implementation suggestions (both of them) will be relegated
to the implementation section.
> P.S. I meant "package-wide" in the previous email. The package should
> always be the outermost logical container, I think.
But packages can span multiple files. This boogers up lexical
scoping. Two different files should both be able to plop
subroutines/methods (and, yes, dynamic variables if they still exist)
into a package, but they need to depend upon their _lexical_ variables
not clashing. Otherwise, some of the more esoteric Perl modules might
not be able to function (those that deliberately tweak the contents of
other packages).
J. David
