pp_substr: avoid unnecessary work prior to inserting a replacement string

[richardleach]

There look to be a couple of inefficiencies in pp_substr when preparing to insert a replacement string:

        if (repl) {
            SV* repl_sv_copy = NULL;

            if (repl_need_utf8_upgrade) {
                repl_sv_copy = newSVsv(repl_sv);
                sv_utf8_upgrade(repl_sv_copy);
                repl = SvPV_const(repl_sv_copy, repl_len);
            }
            if (!SvOK(sv))
                SvPVCLEAR(sv);
            sv_insert_flags(sv, byte_pos, byte_len, repl, repl_len, 0);
            SvREFCNT_dec(repl_sv_copy);
        }

As per the two commit messages:

• repl_sv_copy = newSVsv(repl_sv); is overkill when we can do repl_sv_copy = newSVpvn(repl, repl_len);, which is a much more straightforward process.
• By the time we get to this block, sv should definitely be SvOK. (This check has been changed to an assertion, rather than just removed, out of an abundance of caution.)

---

• This set of changes does not require a perldelta entry.

[tonycoz]

The old code could COW the copy, this will never do so.

That said, we know at this point we need to convert the byte per code point of the source to utf-8, and we don't actually need an SV to do the sv_insert_flags(), so we can just use bytes_to_utf8_free_me(), which won't even copy the string if the upgrade is trivial, ie. all of the code points are invariants.

[richardleach]

Cool, thanks. I'll take a look at bytes_to_utf8_free_me().