tree 68bed5102e083e58f99733711be0d1cba62b0a92
parent 81ed871cc2759de7e996a851af76c09f4cd57ce2
author Todd Wang <toddw@google.com> 1461059191 -0700
committer Todd Wang <toddw@google.com> 1461089420 +0000

Prepare to remove non-nil-any (aka top-level-any) from vdl.Value

Our final goal is to remove non-nil any from vdl.Value.  The
idea is that vdl.Value already represents a statically-typed
"any" value, and we shouldn't allow vdl.Value.Type to return any,
unless the value is any(nil).

In the short-term, we keep the representation and API of
vdl.Value as-is, and let it distinguish between these cases:
  Value{Type: T, Rep: R}
  Value{Type: AnyType, Rep: Value{Type: T, Rep: R}}

However we change ValueEqual to consider these two cases to be
equal, to help us get closer to our final goal.  That should
allow our existing tests to pass, but no longer rely on
distinguishing these cases.  This should make it easier for us to
move to our final world where vdl.Value doesn't represent non-nil
any, and doesn't have an API for it either.

In addition we add the AssignIndex and AssignField methods to
vdl.Value.  If vdl.Value doesn't represent non-nil-any, we'll
need these methods to perform assignment of array and list
elements, and to perform assignment of struct and union fields.
The implementation of these methods still relies on the existing
top-level-any representation.

And we also clean-up the vdl.Value API, removing the *Value
return argument from all of the Assign* methods.  That was meant
as a convenience, but is more confusing than it is helpful,
especially in the new world.  We also add a *Type argument to
each of the *Value() helper functions, so that we can easily
create both named and unnamed scalars.

The net result of this group of CLs is that we should be able to
start converting existing code (e.g. the compiler) over to use
the new no-non-nil-any semantics.

MultiPart: 1/2
Change-Id: I665fc94dbe90c0d53a33c55495c28ac706f9cdc5