Removing subroutines from symbol tables

Ovid has asked for the wisdom of the Perl Monks concerning the following question:

I'd swear I've asked this question, but since I can't find the post, here's hoping I didn't :)

Basically, deleting subroutines from a symbol table seems a bit buggy, but I don't know if this behavior is documented or not.

#!/usr/bin/perl -l

sub this { 'this' }
print this;
undef &main::this; # undefine the code slot
print main->can('this') ? 'main->can("this")' : '! main->can("this")';
eval {print this};
print $@;
undef *main::this; # undefine the entire glob
print main->can('this') ? 'main->can("this")' : '! main->can("this")';

__END__
# result
this
main->can("this")
Undefined subroutine &main::this called at sub.pl line 7.

! main->can("this")
[download]

The first time I try to remove the subroutine via undef it still leaves a CODE slot in the symbol table's *this glob. I can't just use delete because it's not an array or a hash (it fails if you try). This is bad in procedural code but it's awful in OO code because $object->can('this') will succeed even if the method has been deleted.

This leaves me with having to undefine the entire glob but that's a horrible solution because the other slots in the glob, if any, are blown away. Is this a known/documented bug/feature in Perl? Is there a workaround?

Cheers,
Ovid

New address of my CGI Course.

Comment on Removing subroutines from symbol tables Download Code

Replies are listed 'Best First'.
Re: Removing subroutines from symbol tables by chromatic (Archbishop) on Nov 19, 2005 at 07:44 UTC
I've done this before. I usually end up creating a new glob, with either Symbol's `gensym()` or `my $glob = do { local *GLOB };`. Then I copy every slot from the old glob, except `CODE` of course, into the new glob, then overwrite the old symbol table entry with the new glob. I have code somewhere for this; Devel::TraceMethods perhaps. It probably belongs in its own module somewhere.	[reply] [d/l] [select]
Re^2: Removing subroutines from symbol tables by Aristotle (Chancellor) on Nov 19, 2005 at 07:48 UTC
It probably belongs in its own module somewhere. How about right in Symbol? Seems like the right place for that. Makeshifts last the longest.	[reply]
Re: Removing subroutines from symbol tables by Aristotle (Chancellor) on Nov 19, 2005 at 07:42 UTC
That doesn’t surprise me at all. It’s a feature, not a bug: the same behaviour makes `sub foo;` predeclarations work, as well as things like `my $x = \&foo; foo = sub { 'yep' }; print $x->(), "\n"; __END__ yep` [download] The only solution for what you want that I can think of is to save all the slots from the glob, then blow it away, and finally restore the other slots. Makeshifts last the longest.*	[reply] [d/l] [select]
Re: Removing subroutines from symbol tables by Steve_p (Priest) on Nov 25, 2005 at 22:00 UTC
Rather than blasting the entire glob, couldn't you just blast away the CV* inside the glob. `#!/usr/bin/perl -w use Inline C; use Devel::Peek; sub foo{1;} {main::Foo} = \&foo; print Dump Foo; clear_cv(Foo); print Dump Foo; __END__ __C__ void clear_cv(SV* gv) { if(SvTYPE(gv) == SVt_PVGV) { GvCV(gv) = Nullcv; } }` [download]	[reply] [d/l]


XP is just a number
	PerlMonks