[ASDAlexander77]

Follow-up 9 from the own-raytrace rulings ("an interface-typed global given another global's value is never counted").

Bug

let a: I = K.w gave the global no count of its own under rc: claimGlobalInitializers looked through casts and unions to the value a global is given, but not through ts.NewInterface, the view a class object takes for an interface-typed global. Giving either side a new value freed the block the other still read (a = new V(3) freed K.w; rc printed 999 for 1).

Wider than filed:

• the same for a global, a field or an element (let a: I = G / h.v / arr[0]) and const a: I = G (rc);
• let a: I = mk() and let a: I = new V(8): the call's or the new's reference was given back as a discarded temporary at the end of the initializer, so the global was born pointing at freed memory - under rc and own (own exited 127 on main).

Fix

claimGlobalInitializers looks through ts.NewInterface to the object (getThisVal()), so a fresh object is claimed and a read gets its retain.

Tests

• tests/00global_interface_initializer.ts (gc, rc, none; AOT + JIT): each source and each interface global given a new value, after allocations of the same size.
• own/own_global_interface_fresh.ts under own, rc, none and gc.
• Teeth: without the fix rc fails a call's result (exit 127), own fails the same.

Verified

• Windows Release: 4990/4990.
• Linux (WSL): 4972/4972; the new tests again under MALLOC_PERTURB_=165 MALLOC_CHECK_=3: 14/14.
• Both on 787b4ce; the branch merges cleanly into current main (own: compile the ray tracer under -mm=own #655 landed after).

Found, not fixed here

let opt: I | null = flag ? mk(5) : null reads garbage under gc and rc: the C | null -> I | null conversion is a plain cast with no ts.NewInterface.

🤖 Generated with Claude Code