Fix a couple of subtle bugs.
- the VM was conditionally reading from the instruction stream, so that
other bugs got masked by decoding errors.
- push-n-bytes was clobbering eax.
No support for combining characters. Graphemes are currently just utf-8
encodings of a single Unicode code-point. No support for code-points that
require more than 32 bits in utf-8.
1000+ LoC spent; just 300+ excluding tests.
Still one known gap; we don't check the entirety of an array's element
type if it's a compound. So far we just check if say both sides start with
'addr'. Obviously that's not good enough.
There's a question of how we should match array types with a capacity on
ones without. For now we're going to do the simplest possible thing and
just make type-match? more robust. It'll always return false if the types
don't match exactly. For ignoring capacity we'll rely on the checks of
the `address` operation (which don't exist yet). This means we should do
this to pass an address to an array to a function f with signature `f (addr
array int)`:
var a: (array int 3)
var b/eax: (addr array int) <- address a
f b
rather than this:
var a: (array int 3)
var b/eax: (addr array int 3) <- address a
f b
Similar reasoning applies to stream types. Arrays and streams are currently
the only types that can have an optional capacity.