On 5/25/05, Deborah Pickett <[EMAIL PROTECTED]> wrote:
> I'm afraid that because of the dynamic parse/execute nature of Perl, it
> may be a theoretically intractable problem to parse Perl safely.
Yep. It's not really possible for the parser to distinguish between:
BEGIN {
%main::{'&foo'} = sub { print "Hello!" };
}
And:
BEGIN { for 1... {} }
Or even:
BEGIN { system 'rm -rf /' }
So, like most things in Perl, we can trust the author of the code not
to do such things, or we can not trust them and lose something while
we're at it (the choice is given to the IDE user, not to us
designers). BEGIN blocks should never do INITialization, opening
filehandles, etc., precisely for this reason. This becomes even more
important in Perl 6 where bytecode will be more prevelant.
You may certainly parse Perl syntax trees without executing BEGIN
blocks, but you have to bite the bullet if the BEGIN changes the
syntax in a way that you can no longer parse it if you don't execute
it. And if you can't accept that, I see only one option: give up
BEGIN blocks. And we're not going to do that (but you're free to
place that restriction on whoever's code you're parsing).
Luke
