Beagle SCM

RON-001: RONToTime clamps out-of-range slots (never BADARG) + generator must not emit ms≥1000

A ron60 whose calendar slots overflow (ms≥1000 from a same-second monotonic bump) is a legal 60-bit value/string but NOT a valid time; RONToTime (abc/RON.c:91) returns BADARG, and ron._date — the one caller that HARD-throws — crashes the whole jab status. Two fixes: (A) RONToTime clamps each slot to the nearest valid value so rendering / mtime restore never fail; (B) the ron60 stamp generator (ULOG monotonic max(now,last+1) bump) must not overflow the ms field when >1000 rows land in one second. Repro string: 26702A0RFd (ms=1000). Method Issues.

Input

Context

• Found debugging a jab status crash in a store-backed beagle clone:

  JS exception: Error: ron._date: bad ron60.

• jab get wrote 1349 put wtlog rows in ONE second; the ULOG monotonic

  stamp max(RONNow(), last+1) increments the PACKED ron60 integer, so the ms field (low two RON digits, ms = d1*64 + d0, 0–999; abc/RON.c:66) counts 0…999 then rolls to 1000,1001… — legal ron60 ints, illegal times.

• Boundary verified: 26702A0RFc = ms 999 → renders "10:00"; 26702A0RFd =

  ms 1000 (0x3e8) → RONToTime BADARG (abc/RON.c:110-111). Row 181 of the clone .be: put dog/test/KEYW.c.

• Callers: ron._date (js/codec.cpp:164, HARD-THROWS, passes ms=NULL yet still

  hits the ms check), dog/HUNK.c:47/:1664 (fall back to now/"?"), abc/FILE.c:421 io.setMtime (falls back to zero timespec). Only the JS date path is fatal.

Goals

• (A) RONToTime clamps every out-of-range slot to nearest valid (ms→999,

  sec/min→59, hour→23, mon→1,12, mday→1,31, year digits→9) and returns OK; never BADARG once t != NULL. The three callers then fail only on ts==0.

• (B) the ron60 stamp generator (js/ulog.hpp _ulog_now, native ULOG /

  keeper keep_next_pup_key) carries a same-second overflow into the next SECOND (or widens/serialises sub-ms) so it NEVER emits ms≥1000 — stamps stay round-trippable and monotonic.

Constraints

• (A) abc/RON.c only; repro-first (CLAUDE.md §17) as an abc unit test;

  existing RON round-trip tests stay green; ASan clean. RONOfTime input validation stays STRICT (writers must produce valid times).

• (B) ULOG appender layer (spans js/ + dog/ULOG + keeper) — separate landing.
• The clamp is READ-side robustness; it does NOT repair the stored corruption

  — (B) is the root cause.

WIP

Design decisions

• (A) replace each return BADARG in RONToTime with a clamp to the nearest

  valid value (overflow→max, underflow→min); keep RONOfTime strict so this is a decode-tolerance change only, not a widening of what writers may emit.

TODOs

• (A) repro: abc RON test — clamp cases in dog/abc/test/RON.c (ms 1000→999

  time unchanged, mon 14→12, hour 25→23, min/sec 60→59; all return OK).

• (A) clamp patch in RONToTime (abc/RON.c:103 slot-by-slot clamp) — landed

  abc ffc0624b / beagle pin a4094f7f "out-of-range timestamps do not throw".

• (B) generator: carry same-second ms overflow into the next second

  (_ulog_now / keep_next_pup_key); never emit ms≥1000; add a >1000-rows-in-one-second repro. follow-up

• (B) optional: sweep/repair existing wtlogs already carrying ms≥1000 stamps.

Blockers and bummers

• (B) spans ULOG (js/dog/keeper) — larger, its own landing after (A).

Outcome

• (A) LANDED: abc ffc0624b / beagle pin a4094f7f "RON-001: out-of-range

  timestamps do not throw" — RONToTime clamps every slot (abc/RON.c:103), RONOfTime stays strict; clamp cases in dog/abc/test/RON.c.

• (B) STILL OPEN (audit 2026-07-07): keeper keep_next_pup_key

  (keeper/KEEP.c:152) still bumps the PACKED ron60 (max(now,last+1)), and jab _ulog_now (ulog.hpp JABCulogNow) returns raw RONNow() — nothing prevents ms≥1000 when >1000 stamps land in one second.