Skip to content

Tier-2 MON-call oracle for NC — 123B RELES, 54B MDLFI, 503B DVINST

Resolving three SINTRAN III VSX/500 L07 MON-call divergences between two emulators, using carved L07 bytes (and live-recovered L07 code) as the tie-breaker. All addresses and values are octal unless a decimal is named. Every claim below is tagged VERIFIED (byte-proven or live-captured on real SINTRAN L07) or INFERRED (structure/semantics imply it).

Source layers used: - Carved segment listings: /mnt/d/ND/t/re/segments-ref/<seg>/<seg>.asm + .symbols.txt - Resident image: /mnt/d/ND/t/resident/SINTRAN-DATA_commoncode.bin (base 0) - Live-recovered reservation code: ../../OS/21-SEMAPHORES-RECOVERED-CODE.md - NPL L07 source: ../../NPL-SOURCE/NPL/RP-P2-N500.NPL - Existing ND-500 carve: /mnt/d/ND/t/re/mon-analysis/511B-DVIO/ - Error-code decode: ../../../Reference-Manuals/ND-60.145.7A EN ND FORTRAN Reference Manual.md (decimal | octal | text) and the Monitor Calls manual appendix A.

Error-code decode (decimal = octal = text), the three that matter: | dec | oct | meaning | |-----|-----|---------| | 5 | 5 | Device not reserved | | 46 | 56 | No such file name | | 124 | 174 | Illegal parameter |

Q1 — 123B RELES (ReleaseResource): release a NEVER-reserved device

VERDICT: succeeds silently — no error 5.

Releasing a valid device/resource that was never reserved is a designed idempotent no-op; it returns with no error. Error 5 ("Device not reserved") is not raised by RELES on this path — error 5 is raised by the I/O operations that require a reservation, not by the release.

Proof

The RELES handler bytes are not present in the static carve: RELES=37156B in SINTRAN-DATA_commoncode falls inside a 1024-word zeroed hole (37040B..41037B are 000000 in SINTRAN-DATA_commoncode.bin), i.e. the resident monitor image was paged out at carve time. So RELES cannot be answered from the static segment bytes alone.

It is answered from the live-recovered L07 machine code in ../../OS/21-SEMAPHORES-RECOVERED-CODE.md, captured with the CPU stopped inside the mapped monitor segment on a running SINTRAN III VSX/500 L system.

The release primitive BRELEASE (BRELE = 010610) decides the "not reserved" case explicitly:

010623  044400  LDA ,B 0     A := DF.RESLI          (reserve-link of the datafield)
010624  131073  JAZ *+73      RESLI = 0  ->  jump to 010717  (restore registers + EXIT)
...
010717  054270  LDX *-110    restore X
010720  034263  LDF *-115    restore T,A,D
010721  125265  JMP I *-113  return via saved L
  • VERIFIED (live L07): when the datafield's RESLI = 0 (the resource is not reserved), 010624 JAZ branches straight to the register-restore/return at 010717. No error code is loaded; control returns normally. The two error calls in BRELEASE are reached only for inconsistent or third-party-owned datafields, not for the plain not-reserved case:
    • 010627 JPL I *+75 — reached only if RESLI != 0 and RTRES = 0 (chained but ownerless — an internal inconsistency).
    • 010633 JPL I *+71 — reached only if the datafield is owned by a different RT program (X != owner, X != 0). A never-reserved datafield has RESLI = 0 and never reaches either — it exits at 010624.
  • The RELES monitor wrapper (RELES = 037156B, captured but annotation preliminary in chapter 21 §4) loads only file-oriented error codes on its captured branches — 037174 SAA 132 (=err 90 "No file opened with this number"), 037176 SAA 126 (=err 86 "Not opened for random read"), 037203 SAA 133 (=err 91 "Not mass storage file") — i.e. parameter-validation errors for the file-number form of the call. None of them is error 5. The device-release tail (past 037217, not in the capture) performs the actual release via the BRELEASE primitive shown above.

Emulator guidance: MON 123 (RELES) on a valid device number that the caller has not reserved must return success (error code 0) and leave the reservation state untouched. Emit an error only for a bad/undefined device or file parameter (the 86/90/91 family), never error 5, and never for the idempotent "already free" case.

Honest caveat: BRELEASE's RESLI=0 -> silent exit is VERIFIED live; the RELES wrapper's device tail is not in the capture, but the primitive's explicit handling of RESLI=0 as the normal return path is the design proof that release-of-unreserved is a no-op, not error 5.

Q2 — 54B MDLFI (DeleteFile): empty vs nonexistent name

VERDICT: empty/blank name -> error 124 (Illegal parameter); nonexistent (well-formed) name -> error 46 (No such file name). The two cases differ.

Proof / structure

MDLFI = 106063B in 006-S3FS (load base 26000B) is carved (VERIFIED bytes). It is a shared entry that selects the file operation with two status bits and then joins the common file-function body:

106063  174070  BSET ZRO SSM     function selector bit SSM := 0
106064  174020  BSET ZRO SSK     function selector bit SSK := 0   (00 = delete)
106065  021111  STD I 111        stash the parameter pointer
...
106071  135106  JPL I 106  -> ptr @106177   call the name-parse routine (resident)
...
106114  135066  JPL I 66  -> ptr @106202    call the object/name lookup (resident)
106115  124057  JMP 57  -> 106174           on skip-return (lookup failed): error path
106116  044523  LDA ,B 123                  success: continue with function code
...
106174  004402  STA ,B 2         ILDWD: store the returned error code into B+2 (the caller's error word)
  • VERIFIED: the not-found error is propagated, not hard-coded, at 106174 (STA ,B 2 stores whatever error code the lookup routine returned in A into the caller error word).
  • VERIFIED: both candidate error constants are physically present in 006-S3FS: SAA 56 (=170456, error 46 "No such file name") and SAA 174 (=170574, error 124 "Illegal parameter") both occur in the segment.
  • The name-parse (JPL I 106 -> 106177) and the directory/object lookup (JPL I 66 -> 106202) reach through the segment's indirect pointer pool (106177..106211) into resident routines (e.g. pool word 106177 = 003752, a low resident address). That resident code is in the zeroed/uncarved region, so the exact branch that chooses 124 vs 46 is not statically in 006-S3FS.

Which code for which case — INFERRED (strong)

Grounded in (a) the two error-code meanings, (b) both constants living in the file-system segment, and (c) SINTRAN's fixed name-handling order: 1. Empty/blank filename is rejected by the name parser (the JPL I 106 step) before any directory search — a zero-length name is a malformed parameter → error 124 (Illegal parameter). It never reaches the lookup. 2. Well-formed name that matches no file passes the parser, reaches the object lookup (JPL I 66), which fails and returns error 46 (No such file name), propagated at 106174.

Emulator guidance: MON 54 (MDLFI) — empty/blank name string → 124; syntactically valid name with no matching file → 46. (Note "124" here is decimal = 174B; do not confuse with 124B = decimal 84 "Not opened for sequential read" in the octal-numbered manual appendix.)

Honest caveat: the empty-vs-nonexistent split is INFERRED — the deciding branch lives in the resident name-parse/lookup, which is not in the static carve. The MDLFI entry stub, the error-propagation site (106174), and the presence of both error constants in 006-S3FS are VERIFIED.

Q3 — 503B DVINST (InputString): break strategy + returned byte count

VERDICT: the terminating break character IS read, stored, and COUNTED; MaxNo is an inclusive ceiling. The returned count therefore includes the break byte. Of the two emulators, the one returning 14 (break byte counted) matches real SINTRAN; the 12 result under-counts by dropping the terminator (and stopping one read short of the inclusive max).

Proof (L07 NPL, corroborated by the 026-S3IMPIT carve)

The character-move + break loop for DVINST (SMCNO=503) and DVIO (SMCNO=511) is IBTBREAK / IIBM in ../../NPL-SOURCE/NPL/RP-P2-N500.NPL:

IIBM:   (130246)
  130262  CALL IOTRANS; GO FAR TMWT      read ONE char (break/echo tables applied by term driver); none -> wait
  130264  X:=...5FYLLE; *SBYT            store the byte in the ND-500 buffer at offset 5FYLLE
  130270  A:=X+1 =: 5FYLLE              increment fill count  <-- count bumped BEFORE any break test
  130273  GO IBTBREAK

IBTBREAK: (130325)
  130325  A =: D                         D := new 5FYLLE
  130326  IF X.SMCNO=511 THEN X.11MXBRK ELSE X.MAXBYT   max := MAXBYT (DVINST) / 11MXBRK (DVIO)
  130335  IF D>=A GO N5RST              5FYLLE >= MAX  ->  done   (INCLUSIVE: the char that hit max is already stored+counted)
  130337  IF T:=RSISTE>=0 THEN
  130342     IF HENTE=T GO N5RST        this char was the break char (matched break table) -> done
          ELSE
  130346     IF BRECHOFL BIT 5BREAK GO N5RST   break flag set by input driver -> done
  130351  GO IIBM                       otherwise read the next char

N5RST: (130353)  ... restart user, return the count:
  130406  A =: D                         D = number of chars = 5FYLLE
  130407  IF X.SMCNO = 511 THEN A:=0; AD=:X.11NOCRET   (DVIO)
  130416  ELSE A:=0; AD=:X.NOCHRET                      (DVINST) <-- returned byte count = 5FYLLE
  • VERIFIED (L07 NPL): order is store byte → increment 5FYLLE → test break/max (130264 → 130270 → 130325). The break/terminating character is written into the buffer and the counter is incremented before the loop decides to stop. Hence the returned count (NOCHRET, 130416) includes the break character.
  • VERIFIED (L07 NPL): the max test 130335 IF D>=A is >= against the post-increment count, so MaxNo is an inclusive ceiling (the char reaching MaxNo is kept and counted).
  • Break-char detection happens inside IOTRANS/the ND-100 terminal input driver using the break strategy + break tables (BreakStrat=8 → user table BreakT1..BreakT4, 128 bits, bit set = break). The ND-500 side (IBTBREAK) only observes the result via RSISTE/HENTE (a matched break char) or the 5BREAK flag. Terminating chars are exactly those flagged by the break table; a break also occurs on reaching MaxNo.
  • Echo is independent of break: whether the (break) char is echoed is decided by the echo strategy/echo table in the input driver, not by this counter — echo does not change NOCHRET.
  • Corroboration (carved L07): the carved DVIO analysis at /mnt/d/ND/t/re/mon-analysis/511B-DVIO/README.md documents the sibling DVIO handler DVIO=141027B in 026-S3IMPIT (byte-proven), whose input phase is this same XNINSTR/IBMOVE path; its write-back mask and NOCHRET/11NOCRET split match the NPL above (DVINST writes NOCHRET, DVIO writes 11NOCRET).

Breaking the 12-vs-14 tie

Real SINTRAN never stops before storing and counting the terminating byte, and MaxNo is inclusive. So the correct return is the higher count that includes the break character: 14. The 12-result emulator is wrong in one (or both) of the exact ways this loop forbids: (a) it excludes the break terminator from the count, and/or (b) it treats MaxNo as exclusive and stops one read early.

Honest caveat: the rule (store+count-then-test; break byte counted; MaxNo inclusive) is VERIFIED from L07 NPL and consistent with the 026-S3IMPIT carve. Pinning the literal "14" to NC's run assumes NC's break table flags the terminator inside the data window (the normal case); the loop guarantees that whichever byte NC's break table selects as the terminator is itself read, stored, and counted.

Summary of verdicts

Q Call Divergence Verdict Decisive evidence
1 123B RELES error 5 vs silent Silent success (no error 5) BRELEASE 010624 JAZ (RESLI=0 → exit 010717), live L07
2 54B MDLFI 124 vs 46 empty → 124; nonexistent → 46 MDLFI 106063B carved; err propagated 106174; both SAA 56/SAA 174 in 006-S3FS
3 503B DVINST 12 vs 14 14 (break byte counted; MaxNo inclusive) IBTBREAK RP-P2-N500.NPL: store+incr (130264/130270) before break test (130335); count → NOCHRET (130416)