Module replay 

Dual-engine replay: run a Program through the interpreter and the x86 recompiler and compare. Gated to linux/x86_64 (the recompiler needs the Hyperlight host stack, so the whole javm-bench crate is gated to it).

Both engines are driven through the same javm-bench BuiltCaps → invoke_cached path, so they receive byte-identical caps/initial-state — any divergence is a real engine disagreement, never a setup skew.

Structs

Diff

Both engines’ outcomes for one program.

Functions

diff

Run prog through both engines (sharing one BuiltCaps) and compare.

diff_batch

Run a batch through diff, returning (index, Diff) for each diverging program. One long-lived sandbox handles the whole batch — no rebuilds (those were the source of host-heap corruption; a single sandbox runs thousands of distinct programs cleanly).

diff_image

Run a pre-built Image through both engines and compare.

image_for

Build the Image for a program: its code (body + signature epilogue) plus the appended ecalli 0 terminator, entered at pc 0 with the program’s initial register seed.

image_with_ro

Build an Image from raw instruction words (+ ecalli 0 terminator) with a pinned read-only data cap of ro_bytes mapped at ro_start — for category-#3 read-only-cluster differential tests. Both engines materialize it PinnedCapRo (interp perm-RO, recompiler MatRange) and charge it per 2 MiB cluster.

image_with_ro_caps

Build an Image from raw words with several pinned read-only data caps, each (start, bytes) — for multi-cap read-only-cluster differential tests (e.g. two distinct caps sharing one 2 MiB cluster). Each cap takes its own cnode slot, so the recompiler resolves each as a separate PinnedCapRo MatRange with its own source PA, exactly as production does.

replay_interp

Interpreter outcome for prog.

replay_recomp

Recompiler outcome for prog.

seed_reg

Convenience: seed register xreg (by x-number) to val in a slot-keyed init map. Mirrors the generator’s seeding; handy for hand-built programs.