Dan Sugalski wrote: > my $foo; > $foo = 12; > print $foo; > $foo /= 24; > print $foo; > > may well have the vtable pointer attached to the PMC for $foo change with > every line of code. Probably will, honestly.
Well, there's only two assignments there, so I assume that print is going to upgrade the scalar to have a string rep. Are you thinking of scalar representations something like: undef integer integer_with_string real_with_string (dirty) real_with_string I like the principle of automatically changing representations, but shouldn't the compiler be a little smarter? In this example every expression has a static type. We're burning memory and CPU just to do stuff at run-time that the compiler could have eliminated. For example, Perl could use reps like: number (undef) number (int) number (int) number (float) number (float) Building inlined "number" vtable ops seems like a big win if Perl can use them. Which brings up an interesting question. Does Perl the language require just-in-time representation changes or is it legal for the compiler to choose an efficient representation in advance? - Ken
