First cut of enforcing return types of subs and methods. Both of Foo and returns Foo forms work - also did some more general infrastructure for trait verbs on routines to make this work. Limitation in this patch is it only works for explicit calls to return, not yet implicit returns by falling off the end of the sub. .of and .returns on a Sub also work as a result of this, since it's done by having the Callable[::T] role holding the return type. Because when we get the current Sub using interpinfo we end up with the Parrot-level sub rather than our reblessed one, need to maintain a reference on that to the re-blessed one that carries the type info. Hopefully we can drop that in the future, somehow.

Move Pair.key back out of the Setting. Rationale: we use Pair.key when doing is export, and thus end up calling a Perl 6 sub before it's had chance to be re-blessed, which is about to become a bad thing. Happily, I think this is all that will need moving.

Add a type mapping for Iterator, so @foo.itertator.WHAT doesn't blow up.

Can't .include the parametric roles macro at runtime, so just inline the PIR (best I can think of for now; don't see to have Q:PIR{ } in NQP, which may allow something shorter).

Associative becomes a parametric role.

Positional becomes a parametric role.

Some cleanups and tweaks to Callable[::T].

Callable role becomes parametric (like role Callable[::T = Object] { }).

For registering parametric roles in the stage one set of built-ins, !meta_compose must not depend on having a constructed Perl6 Compiler object yet.

Remove Junction's deendence on Array and List, then move it to an earlier spot in the built-ins list.

Make Signature not be dependent on List and Failure, to break a dependency chain that got in the way of the bootstrap.

Forgot a file in the last commit.

Factor parametric role creation PIR out into a macro, which neatens actions.pm a little and means we can use it to create parametric roles from PIR too.

First cut at implementing lexical multi subs.

Merge branch 'master' of git@github.com:rakudo/rakudo

Implement lexical subs for the non-multi case.

Parse my multi foo() { }, as STD.pm does.

Make various built-ins that should use $*OUT and $*ERR actually use them.

Add S16-filehandles/unlink.t to the spectests.

Fix up exporting (and as a result, importing) of multis.

Revert an accidential change from earlier debugging.

Fix importation of modules in nested namespaces.

Make .assuming work on multi-methods. Add the now-passing S06-currying/assuming-and-mmd.t to spectest.data.

Move .reverse and .list of List into the setting.

Get gen_setting_pm.pl to generate use statements at the end, which import the various classes in the setting. We also have to make sure we note them in %*INC so it doesn't go trying to load Any.pm, etc from disk. This is fine for when we're compiling - just use a BEGIN - but since those aren't persisted we need to make sure we set up %*INC in the pre-compiled version too. Did it with a sick hack for now; hopefully someone thinks of a good way...

Fix a braino in my first cut of import; we may only want to load the module once, but we may need to import many times into many different scopes, so checking we only imported the first time was completely bogus.

Support renaming of named parameters in the signature. This also takes us a couple of steps closer to STD.pm.

Make sure that prompt in void context actually prompts.

Make sure recursive calls to the compiler through use get a @?BLOCK.