Pass 1 now records every field and method signature (access, type,
parameter types) in bank-1 member and parameter tables, rejects duplicate
fields or methods per type, invalid init/destroy shapes (E151/E152, D122),
and `var` after a method statement or inside a block. Each signature and the
program open a typed local scope. Pass 2 scans statement lines: unqualified
names must be locals, parameters, or fields of the enclosing class (D123),
and member chains through locals, fields, self, new, calls, and indexing
report unknown members (E061) and private access (E120) at the member.

The parser test PRG grew past $4000 and crashed: mmu_init copies the COMMON
load image in the system mapping, where BASIC ROM hides $4000+. COMMON now
loads right after STARTUP and mmu.s asserts both stay below $4000.

Verified in x128: 28 fixtures, member tables against a source-derived
oracle, privacy and scope mutations caught; map, source, lexer, names, and
diagnostics suites pass on the new layout.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

parser.s runs two passes over the cached source. Pass 1 parses interfaces,
classes (implements lists, fields, method headers), and the one program,
records top-level types in a bank-1 table, tracks method/program/if/while/
for blocks, checks duplicate locals, and reports syntax, ordering, and
duplicate errors with line-level recovery. Pass 2 reports type references
that name no declared type, so forward references are legal.

Verified in x128 against 24 fixtures through the real ERR files: 7/7
accepted, 10/12 rejected grammar fixtures at their sidecar positions
(line_cont and multi_stmt are statement-level, pending E03-S03), and five
parser fixtures matching their complete diagnostic lists. A pass-order
mutation is caught. D121 (proposed): method names may be keywords, as
spec 5.1's `method print()` requires.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

names.s gives each distinct identifier a 1-byte ID (1-255), comparing
case-insensitively across both PETSCII letter ranges. The table lives in
bank 1: 64 hash-chain heads plus 20-byte records holding the original
spelling. name_declare replaces the first-use spelling with the first
declaration's spelling, so forward references keep the declared form.
A full table returns NAMES_ERR_FULL and writes nothing.

Verified in x128: the C128 dump equals a host model byte for byte across
case variants, an 8-name collision chain with an identical-hash pair, and
a 260-name overflow, with a guard after the table intact. A mutation that
skips the text compare is caught. D120 (proposed): 255-name limit.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

src_load reads a SEQ file from device 8 once into a bank-1 cache (D118).
It enforces the 80-byte line, 599-line, and 18,999-byte limits as bytes
arrive and reports stable codes with exact positions. src_rewind and
src_next_line replay the cache one logical line at a time for every pass.
diskio.s wraps the KERNAL file calls; channels close on every path.

Verified in x128 with a 1581: eight fixtures, including the exact limits,
each limit error, a missing file (DOS 62), and a missing final $0D, all
match host-computed line counts, Fletcher-16 checksums, and positions.
Adds a shared PowerShell test library for x128 scripts.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Add cfg/compiler.cfg, the production mmu.s, c128.inc, and the test mailbox,
proven by a capacity-checking skeleton in x128:

- 8K bottom common RAM, shared code at $1300, private zero page at bank 0
  page $1A via P0, bank-1 tables at $2000-$BFFF, test mailbox at $1BF0
- ld65 rejects overflows of the ZP, COMMON, MAIN, SWSTACK, and TABLES areas
- a CIA #2 timer NMI taken under bank 1 with the private zero page returns
  cleanly through the NMI guard (closes the E01-S03 NMI risk)
- decision D117 records the map; RESTORE is ignored while compiling

Also pin newLang/ to LF in .gitattributes. A branch switch under
core.autocrlf had rewritten the byte-exact fixtures with CRLF.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>