{
my $glob = gensym() ;
${*$glob} = 123 ;
print "SCOPED: " . "${*$glob}\n" ;
}
print "OUT: " . ${*Symbol::GEN0} . "\n";
I don't think you quite understand what that is doing.
At compile time, that *Symbol::GEN0 in the OUT line causes
the GEN0 glob to be created and referred to directly by
the compiled code. Then at runtime, gensym fetches and deletes it from the symbol table (skipping creating it since it's already there). Adding to your test:
use Symbol;
print "Symbols: ", join(",",keys %Symbol::), "\n";
{
my $glob = gensym() ;
${*$glob} = 123 ;
print "SCOPED: " . "${*$glob}\n" ;
}
print "OUT: " . ${*Symbol::GEN0} . "\n";
print "Symbols2: ", join(",",keys %Symbol::), "\n";
print "OUT2: ", ${"Symbol::GEN0"} . "\n";
gets you:
Symbols: qualify,geniosym,gensym,GEN0,EXPORT,EXPORT_OK,import,ungensym
+,qualify_to_ref,EXPORT_FAIL,delete_package,ISA,BEGIN,VERSION
SCOPED: 123
OUT: 123
Symbols2: qualify,geniosym,gensym,EXPORT,EXPORT_OK,import,ungensym,qua
+lify_to_ref,EXPORT_FAIL,delete_package,ISA,BEGIN,VERSION
OUT2:
About the undef of types, I don't remember ever seeing your actual tests, and I wasn't able to deduce anything useful from your Safe::World. If undefing the individual parts of the glob is freeing more than undefing the whole glob, it is a bug and sample code showing it in action would be very welcome.