-- `b = a` then `b = <new>` must NOT clobber `a`.
--
-- The VM compiler resolves `Expr::Ref(a)` to `a`'s register directly with
-- no MOVE (free read). Pre-fix, the `Stmt::Let` new-binding path then
-- aliased `b` to that same register, so a later `b = 99` reassignment
-- overwrote the shared slot and corrupted `a`. Cranelift inherited the
-- bug because it lowers VM bytecode. Tree-walker was always correct.
--
-- The fix: when the compiled RHS register is already owned by an existing
-- local, allocate a fresh register and emit OP_MOVE so each user-visible
-- binding has its own slot. One extra MOVE per shadow-from-ref, no impact
-- on the common `b = +a 1` / `b = mset a k v` shapes (those already
-- allocate fresh registers from the arithmetic/builtin ops).
-- Number: the original repro from the Zipf-slope corruption report.
number-shadow>L n;z=3.14;t=z;t = *t 2;[t z]
-- Map: same shape, different value type. mset returns a fresh map but
-- `b = a` first must give b its own slot.
map-shadow>t;a=mset (mmap) "k" 1;b=a;b = mset b "k" 99;fmt "{}|{}" (mget a "k") (mget b "k")
-- List: classic shadow followed by accumulator rebind.
list-shadow>L L n;a=[1 2];b=a;b = +=b 99;[a b]
-- Text: shadow followed by self-concat.
text-shadow>t;a="x";b=a;b = +b "y";fmt "{}|{}" a b
-- run: number-shadow
-- out: [6.28, 3.14]
-- run: map-shadow
-- out: 1|99
-- run: list-shadow
-- out: [[1, 2], [1, 2, 99]]
-- run: text-shadow
-- out: x|xy