Clear all variable pad slots after multi-variable foreach loop

[leonerd]

We need to remember to clear all the pad slots related to iteration variables, which means some number of additional ones after the main "itervar" when doing multi-variable iteration. This ensures we don't retain references or large scalar values longer than necessary.

I'll write a perldelta for this + the other recent fixes for foreach loops once they're all in, before the next point release.

[leonerd]

Oh dear. One failing test on macOS, dist/threads complains about a panic to copy freed scalar. There's a chance that actually is related to my change.

panic: attempt to copy freed scalar 152a1d168 to 152983a38 at t/free.t line 118.
../dist/threads/t/free.t ............................................. 
Dubious, test returned 60 (wstat 15360, 0x3c00)
Failed 28/29 subtests 

[leonerd]

Though, looking further at the test (dist/threads/t/free.t line 118) it really doesn't appear to be anything to do with multivariable foreach. So maybe not.

I'll rerun it and see if it happens again; it could be a spurious unrelated issue.

[leonerd]

Well that's slightly awkward. Just rerunning the test made it pass. I'm a little nervous about that because it doesn't explain why the test previously failed, but I've looked up and down the code and I really can't see how it relates to multivariable-foreach and also the failing test file runs reliably without upsetting valgrind for me locally, so I don't feel it is directly related to this work.

[mauke]

Perl_cx_dup also sets blk_loop.itersave. Would it need to initialize blk_loop.additional_padvars, too?

[leonerd]

I don't believe that's necessary. The Copy() operation will copy the value of this field from old to new; then various fixups happen afterwards to adjust pointer-like things to point at nested cloned structures. This field just stores an integer, so needs no further treatment.

[mauke]

Could cx->cx_type & CXp_FOR_PAD be written as CxPADLOOP(cx)?

[leonerd]

Indeed it can :)