pp_split: reify using NULL rather than PL_sv_undef (gh#18077)

[richardleach]

Non-existent array elements have been represented by NULL rather than &PL_sv_undef since ce0d59f. As discussed in #18077, an instance of reification in pp_split seems to have not been updated to match.

I'm not sure what ideal Perl code for performance testing is, but trying with @_ suggested that there's little difference between:

        for (i = AvFILLp(ary); i >= 0; i--)
	    AvARRAY(ary)[i] = NULL;

and

        Zero(AvARRAY(ary),AvFILLp(ary) + 1,SV*);

I just went with the latter because it results in a very slightly (16 bytes) smaller pp.o.

make test succeeds locally (64 bit Linux).

[khwilliamson]

This LGTM, but I'm not an expert here.

Any reason not to apply this patch?

[atoomic]

The Zero solution seems also better IMO than the for loop.
I wonder if we could not use this trick in other locations too shown in ce0d59f

[atoomic]

Please add a protection for AvFILLp(ary) >= 0

[richardleach]

@atoomic -

Please add a protection for AvFILLp(ary) >= 0

As per review comment above, I'm now wondering if the existing loop is redundant and should be removed.

The Zero solution seems also better IMO than the for loop.
I wonder if we could not use this trick in other locations too shown in ce0d59f

I've used Zero() in #18072, which hasn't received any review comments to date.

[richardleach]

D'oh, I added the originally-requested protection for AvFILLp(ary) >= 0 and force-pushed, and only then noticed the additional review comment stating that it's not necessary. I'll do another push.

[richardleach]

Additional push done. Functionally that takes this PR back to its original form, but there is now the additional comment about potentially redundant code. (I may well submit a follow-up PR to remove it.)