# Session checkpoint — Turbo-Modula-2 16→64 MCD translator + mc4x VM

Date: 13 sept. 17:4x. Working dir: `/home/eric/Projets/Projets-Modula2/MyWork/M2compiler/Resources/mc64`
(real absolute roots vary; use `find` when resolving .MCD inputs).

## State: WORKING
- `trans8to64` : 16-bit `Foo.MCD` -> 64-bit `Foo64.MCD` (mc64 VM image).
  - Build: `gm2 -fiso -o trans8to64 trans8to64.mod FileIO.o Console.o` (add `-g` for dbg build).
  - `Extended.MCD` -> clean rc=0 now (was aborting rc=134). All 9 spec modules translate;
    only `MCODE.MCD` fails BY DESIGN (114-B entry stub, no 304-byte descriptor plane).
- `mcint` : 64-bit VM that loads a translated image and runs it.
  - Build (full object set — do NOT prune):
    `gm2 -fiso -o mcint mcint.mod Loader2.o Interpreter.o Instruction.o Local.o Stack.o
          Memory.o Global.o Extended.o Console.o FileIO.o`
  - Verified: `mcint /tmp/Stack64h.MCD` -> rc=0 (end-to-end).

## Last bug fixed (this session)
gm2 `FOR j := 0 TO vcnt - 1` with `vcnt = 0` underflows an binds loop 65536x ->
`Get16`/descriptor overrun on `Extended.MCD` (rc=134). Fix: wrap each of the two
var-sizes loops in `IF vcnt > 0 THEN ... END` in `trans8to64.mod` (read+write both),
and the proc-scan loop in `IF nEnt >= 2 THEN`. Also patched `Interpreter.mod` 0CDH:
default/out-of-range case now `Push (eot + t)` (was `Push (eot)`) — honours 16-bit
return offset.

## Known limits / next
- `Extended64`/`Array`/`Local` translate but are `dcnt=1` (depend on loader-chain
  modules); cannot run standalone under `mcint` yet -> needs dependency-chain loading.
- Original 16-bit stack images themselves run under the 16-bit VM, not `mcint`.

## Files
- `mc64/trans8to64.mod`, `mc64/Interpreter.mod`, `mc64/mcint.mod` — the code.
- Binaries: `trans8to64`, `trans8to64dbg`, `mcint`.
- Test outputs: `/tmp/Stack64h.MCD`, `/tmp/Extended64g.MCD`, `/tmp/Array64*.MCD`...