# Two-phase refactor — slice 1 (`twoPhase` flag + AST scaffolding)

Branch `ast-stage-c` off `1c93f19` (`v3-two-phase-plan`).
Suite **178/178**; self-hosting fixpoint **OK** (**2,957,466 bytes**;
+56,684 vs 2,900,782 — the AST module is now linked into the compiler).

## What landed

`docs/plan-two-phase.md` §4 slice 1: the parser now builds AST nodes
**alongside** the legacy emit path, and the AST module is part of the
compiler build.  The nodes are **inert** — nothing reads them, so the
generated `.ssa` is unchanged (stage2 == stage3 byte-for-byte).

| File | Change |
| --- | --- |
| `compiler/src/M2.atg` | `IMPORT … , AST`; `VAR twoPhase : BOOLEAN`; the `M2` root production runs `AST.Init; twoPhase := TRUE;`; the `integer` expression leaf also builds `NkIntLit` via `AST.MakeLeaf`; `Fact` gets a throwaway `astScratch : AST.Node` |
| `compiler/build.sh` | compile and link `AST` (after `FileIO`) |
| `compiler/src/M2.lst` | regenerated CR listing |

`M2P.mod` now imports `AST` and calls `AST.Init` once per parsed unit
(the root production runs per source file), so every unit starts with a
fresh arena.  The new `AST.def/.mod` are picked up transitively by
M2make during stages 2/3 of the fixpoint.

## Notes

- `twoPhase` is a grammar-level BOOLEAN, set TRUE in the root
  production.  Later slices gate the new lowering walk on it; for now it
  only guards the inert `NkIntLit` build.
- `Compilation ended with LL(1) errors.` from CR is **pre-existing**
  (benign) and unchanged.
- No new test: every existing test already exercises the path (each
  integer literal builds a node), and there is no observable behaviour to
  assert yet.

## Next

Slice 2 (`docs/plan-two-phase.md` §4): `Expr` / `SimExpr` / `Term` /
`Fact` build `NkBinExpr` / `NkUnary` / `NkIdent` / literals in addition
to emitting; add `Lower` for expressions, exercised by a test-only
`.ssa` byte-compare.