Fix Perl_av_store off by one error in key comparison

[bbrtj]

This change addresses a long-standing but rather uncommon bug in checking for replacing scalars in a readonly array. In rare cases, modifying the last scalar in a readonly array will croak with Modification of a read-only value attempted. This happens because the comparison treats the last index of the array as if it was array size.

Here is a script which reproduces this bug:

use feature qw(refaliasing);
my @xs = qw(a b c);
Internals::SvREADONLY(@xs, 1);

\$xs[1] = \42;
\$xs[2] = \42;

Prints:

Aliasing via reference is experimental at test.pl line 5.
Aliasing via reference is experimental at test.pl line 6.
Modification of a read-only value attempted at test.pl line 6.

---

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

[khwilliamson]

Thank you for this.

I would propose though phrasing this in terms of av_count which we invented because of how easy AvFILL is to misuse, as this bug demonstrates. I probably should have tried expunging most occurrences of AvFILL at the time. That might have caught this one.

[bbrtj]

I think AvFILL actually makes sense in this place. x > AvFILL seems more natural here than x >= av_count, because we actually want to check against the last index and need to replace > with >= when checking against the array count. But if you still think it's better to have it use av_count instead, then I can of course do that.

[tonycoz]

Could you please add a regression test?

[bbrtj]

Added a test, I hope I've put it in the right place.