# V3 step 1.2 — dynamic symbol storage (done 2026-09-19)

Follow-up to step 1.1 (scope tree + per-scope BST): the fixed-size
pools are gone, replaced by `Storage`-allocated linked structures.
`SymTab.def` unchanged apart from dropping the now-meaningless
`MaxSyms`; `M2.atg` untouched; suite stays 6/6.

## What changed (`compiler/src/SymTab.mod` only)

- Symbols: heap nodes (`SymPtr`, `ALLOCATE` + `TSIZE`) linked as a
  BST per scope instead of `nodes[0..255]`. No capacity limit;
  `Enter` fails only on duplicates.
- Scopes: heap nodes (`ScopePtr`) forming the scope tree, plus a
  creation-order `link` chain so `PrintTable` still dumps every
  scope (in-order per scope, alphabetical).
- Record fields: heap list (`FieldPtr`, filtered by `owner`) instead
  of `fields[0..511]`.
- Kept bounded on purpose: `pend[]/pendF[]` (transient, per single
  declaration, max 64) and type descriptors (`tform[]/tref[]` —
  `TypeIndex` is an INTEGER woven through the grammar, so
  descriptors stay indexed for now).
- Nothing is ever freed: one compilation per process, the whole
  table is discarded at the next `Init`. `DEALLOCATE` intentionally
  unused (single-pass compiler, no steady-state growth).

## Verification

Rebuilt clean (LL(1)-clean, no gm2 warnings introduced), 6/6 green:
insert/find/dup-detection all exercised (`t_bad_dup` → 200,
`t_bad_undecl` → 201, 3 run tests byte-identical `.ssa` — the
backend is untouched).

## Note for step 7 (syslib port)

`FROM Storage IMPORT ALLOCATE` is now the only gm2-hosted import in
`SymTab` (plus `SYSTEM.TSIZE`, directly lowered later). The port
swaps it for the self-hosted `Storage` without touching the tree
logic — one reason the pool removal happened before the port.