perl6/nqp

…r corruption.

rakudo does not use them yet, and that step breaks the build on non-windows
we will figure out a proper fix after the rakudo release.

This reverts commit c7bed44.
It was wrong because building p5regex with stage0
involves the wrong serialization context.

…s to cover it.

…a few bits that will be unrequired.

… simple enough to be valid p5 and p6. Passes.

…s needed, looking to STD_P5 for inspiration.

tabs missing

child_objs is now allocated with mem_sys_allocate_zeroed, and a child object
is ignored if child_objs[i] is NULL. This fixes a bug where write-barriering
would trip over a bogus pointer when write-barriering a struct returned from C
and a field was unread before passing it to another C function.

…able difference to the output size of CORE.setting compilation, and a small but certainly worthwhile difference to CORE.setting compilation time. Causes no issues here; hopefully not for anyone else now either.

…on to nodes.

… be boxed. Outsourcing this to QAST has various benefits, including keeping this out of the QAST tree itself. Then inlining doesn't have to worry over this detail.

…on scheme.

…ng is switched to using it yet, though).

…an argument needs to be filled in when inlining. Implement a first cut of producing a tree with the placeholders substituted in.