fix(as370): an undefined symbol is IFO188, and the instruction is zeroed (#82)

[mgrossmann]

Closes #82.

x_factor returned 0 for a symbol in no table, in pass 2 as much as in pass 1, and recorded nothing. IFOX00 rejects it — IFO188 <symbol> IS AN UNDEFINED SYMBOL, severity 8 (erms.asm:193 / jermsgcd.asm SEV188) — and assembles the whole machine instruction as zero.

The bytes matter as much as the message. IFOX00's all-zero instruction is an invalid opcode and program-checks when reached; as370's kept its opcode and ran — a branch to address 0, a load from base+0, an MVC into offset 0 of whatever the base register held.

Shape of the fix

statement kind	what happens
machine instruction	scan_undef_terms in the else if before the format switch — instruction zeroed, hasa1 (and hasa2 for SS) set to 0
DC/DS/EQU/ORG/USING/END	recorded from x_factor, pass 2 only — IFOX zeroes the value, not the statement
literal (=A(NOVAL))	not zeroed. IFOX assembles L 4,=A(NOVAL) normally and flags the pool; emit_lit flags the statement that wrote the literal, the way IFO158 already does

Because a flagged instruction never reaches resolve()/expr_val(), the two halves cannot report the same symbol twice.

The predicate is neither !sym_find nor !defined. sym_get enters symbols that are only referenced — EXTRN/WXTRN, a V-con, a =V literal — and those keep defined == 0 for the whole assembly. !defined alone would have flagged every external in the corpus. They carry S_ER, and nothing else undefined does: !s || (!s->defined && s->type != S_ER).

At the x_factor site the value and the relocation count are deliberately untouched — the change there is diagnostic only, which is why no deck moves.

Counted whether or not printed, per bce32d8: list stops at 512, the overflow says so (verified with 600 undefined symbols: 512 printed, ... and 88 further undefined-symbol diagnostics, RC 8).

Verification

	result
fixture vs. listref/ifox-listing-undefsym.txt	object bytes and ADDR columns identical; literal stays 5840 F018; RC 8, 5× IFO188
826 ecosystem modules	0 raise IFO188; the same 825 assemble as on the baseline
tests/corpus/check.sh	no module moved — 743 decks byte-identical
tests/run.sh	41 OK, 0 failures, ALL SAMPLES BYTE-IDENTICAL TO IFOX00 (incl. hello/litmove/arith/ksdsnatr with a SYS1.MACLIB extract)
tests/listref/check.sh	all four references still column-exact

Modules probed: libc370 (743), rexx370, ufsd, ftpd, httpd, httplua, httprexx, lua370, mvsmf, lstring370, nsf370.

The positive control. nsf370 has since fixed its eaten cards (9012263), so its current tree is clean. Assembled at 9012263^ — the state the issue measured — the two modules report exactly the symbols the issue names:

nsfctcio.asm  DCBDDNAM  IHADCB  DCBOFLGS  IHADCB  DCBOFOPN
nsfvsvc.asm   ANCVERNO  ASCBASID  DOUNSTG

L R3,REQFUNC(,R8) is absent from that list because the card was eaten whole — that one is #72's diagnostic, not this one.

Tests

The #82 tripwire is replaced by the oracle's numbers: RC 8, the five messages at their statements, the 14 zero bytes followed by an intact L, and the single RLD entry (so an undefined term invents no relocation).

A second case guards what must not be read as a symbol — X'40', C'A', B'1111', L'FIELD, =A(…) — with every symbol a forward reference, which is what proves the diagnostic stays silent in pass 1. Without the self-defining-term skip, MVI FLAG,X'40' would report an undefined symbol named X and zero a good instruction.

Known divergences, documented not fixed

• NUMBER OF STATEMENTS FLAGGED: IFOX says 4, as370 says 5 — as370's counter counts messages. Second recorder to inherit this knowingly; filed separately.
• A EQU B with B defined further down: IFOX resolves EQU in pass 1 and flags IFO188; as370's pass-2 gate keeps it silent and gives A the pass-1 value 0. Its own defect, adjacent to this one; filed separately.
• tests/undefsym.s is frozen — IFOX00's listing prints its five comment cards verbatim and numbers every statement from them, so its now-stale header cannot be corrected without the fixture and the oracle drifting apart. Noted in tests/run.sh and tests/listref/README.md.