Optimise foreach on builtin::indexed ARRAY

[leonerd]

Rather than generating an entire temporary list that is twice as big as the original array, instead set a flag on the OP_ITER that tells it to set one of the iteration variables to the current array index and use the same CXt_LOOP_ARY optimisastion that regular foreach over an array would use.

• Consider whether builtin::indexed LIST... should do the same optimisation. Likely the answer is "yes", but in a separate change.

• This set of changes requires a perldelta entry, and it is included

[leonerd]

This change does have the potential impact that, before it, any modifications to the iterated-on array that happen during the body of its own foreach loop would not get seen, whereas now they would. This is the same issue that single-variable foreach on a plain array already had, so it's not new. But it is a change for existing code that already calls this function.

[tonycoz]

builtin.c: In function ‘Perl_XS_builtin_indexed’:
builtin.c:495:35: warning: unused parameter ‘cv’ [-Wunused-parameter]
  495 | Perl_XS_builtin_indexed(pTHX_ CV *cv)
      |                               ~~~~^~

Normally suppressed by the attribute in :

#define XSPROTO(name) void name(pTHX_ CV* cv __attribute__unused__)

[tonycoz]

Deparse needs an update:

$ ./perl -Ilib -MO=Deparse -Mbuiltin=indexed -e 'for my ($i, $v) (indexed @ARGV) { print "$i: $v\n"; }' 1 2 3
use builtin (split(/,/, 'indexed', 0));
foreach my ($i, $v) (@ARGV) {
    print "${i}: $v\n";
}
-e syntax OK

[tonycoz]

Looks ok otherwise

[Grinnz]

This change does have the potential impact that, before it, any modifications to the iterated-on array that happen during the body of its own foreach loop would not get seen, whereas now they would. This is the same issue that single-variable foreach on a plain array already had, so it's not new. But it is a change for existing code that already calls this function.

Consistency seems good, and the foreach docs basically say "don't do this" already. But maybe this change should have a brief mention in perldelta, since this feature isnt experimental anymore?