Commit graph

4 commits

Author SHA1 Message Date
jvanvik
9597d89079 Server-run engine: review fixes
The review of track T2 found these; each has a test in tests/mp-sim-*.

- Built-ins: an arg, array entry or draft that is '__proto__', 'constructor'
  or 'prototype' is refused by the envelope check and again by the realm
  (free text may still say them). After every realm call the trampoline
  checks the realm's built-ins against what they were at load; a realm
  whose built-ins changed answers {srvErr, polluted} and is replaced from
  the worker's mark like one that timed out. S as one JSON per key is a
  null-prototype object, so the epilogue's for…in never reads a prototype.
- A realm that cannot be built again is never used again: rebuild() tries
  once more without the packed R, and when that fails too the network is
  dropped (NO_REALM), and the engine tells the host (opts.onLost).
- A mark that fails or runs out of time changes nothing: it clears the old
  mark first, the realm is built again from its own save, never from an
  older mark, and the reply names the mark. A day rebuilds only from its
  own attempt's mark.
- A day whose alone or nets is not a list is refused whole (ERR); a throw
  in a load or day step answers that call, never ends the worker.
- An exit counts against the session the worker was running (the worker
  says so before each job and step), else as before. Workers of their own
  are capped (opts.maxOwn, 4): past that a session is given up.
- A worker that exits before it is ready is started again after a pause
  that doubles; after 5 in a row every call answers SIM_DOWN.
- The game text is read and compiled once, in createEngine: a src/ that
  does not compile throws there, and every worker, also one started after
  an exit, runs the game engine.simId names.
- A damaged route memory never breaks a save that loads (routed from
  nothing, logged), and only a server network loads (S.mp.sim, Normal).
- Routing memory (UI.rk, UI.rkR) is kept by the in-place restore and
  carried in rt (rk), so a reloaded or rolled-back network skips the same
  routings as the one that kept going.
- A fresh realm takes over the fx of the one it replaces (the mark's after
  a failed call, all of them after a read): badges, windows and toasts
  waiting for a page are not lost to a timeout.
- Windows: the stand-in elements answer querySelector from what was
  written (a [data-x] marker) and have a working classList, never the
  inert proxy that said every window is open. A request's ctx may name the
  open window, ctx.win = {k, ui} (mp-sim-acts.js WINDOWS: bstd, routing,
  charter, ops, pie), with the view state only the page holds (the
  intern's POP, the pie window's tab and POP), checked in the realm. This
  is an addition to the contract of design 5.8 that the page (T4) sends.
  The stand-in is hidden again after an act.
- xpBuild's via is a TEXT arg (256): a six-span express fits.
- actDone, askDone and pulled carry ms, the worker time of the call;
  engine.mem() and engine.has(code, nid) are public; prefs() passes a
  realm's err through; after close() every call rejects CLOSED.
- The fake engine fails as the engine does (NO_REALM, ERR, a load without
  a save, a day of an unknown session or naming an unloaded network,
  GAVE_UP, SIM_DOWN, CLOSED, onLost), its greet says its one-shots once and
  never 'told'; a test runs the same edge calls on both.
- SIM_DRAFT is built from DRAFT_TYPES.
2026-11-09 22:16:52 +01:00
jvanvik
72f31e809b Server-run engine: every call answers, and a crashing load gets its own worker
A load whose networks did not all check out could hang its session for good, a
worker could spin on a job it could never run, a session whose load crashed its
worker never got a worker of its own, and an unload followed at once by a load
(a language change) lost the session's place or never answered. Fixed:

- Worker: a load call is checked whole before it touches anything. One that
  names no network, or one twice, is refused (ERR) and every realm stays as it
  was. A network without a save now answers err (broken) like one whose save
  does not load, and the others load. A network marked loading with no load
  under way answers NO_REALM instead of waiting for ever.
- Worker: it goes on only while something ran. If a job were ever still
  waiting when nothing could run, it is answered (ERR, logged) rather than left
  pending, and the worker no longer busy-loops on it.
- Engine: a session is on its worker from the moment its load is sent. A worker
  exit counts against the sessions whose load was under way (else against every
  session it held), so a save that crashes its worker gets a worker of its own
  after two exits and is given up (GAVE_UP) after a third, while the sessions
  that only shared the worker lose their realms but not their place in the pool.
  Loads under way count for placement.
- Engine: an unload no longer drops the placement of a session whose next load
  is under way, so an unload and a load sent back to back stay on one worker; a
  worker of its own is retired only when no call is pending on it (a pending
  call there used to never settle).
- Engine: a message that cannot be sent rejects (ERR) and leaves nothing pending.
- Realm: a what-if puts UI.hotMap back too (teSlop's trial left its own).
- Realm: a network in a language this server has no dictionary for plays in
  English, as the page does, instead of loading broken.

Tests (tests/mp-sim-engine.test.js): a malformed load changes nothing and every
call of the session answers within 3 s, with the workers idle after; an unload
and a load back to back keep the session on its worker; a session whose load
takes its worker down twice gets a worker of its own, then GAVE_UP, while the
bystander is never blamed; an ask leaves UI.hotMap as it was; a network in an
unknown language loads in English. Each fails on the engine before this change
(the worker's alone: the hang; the engine's alone: the placement and the exits;
the realm's alone: teSlop's hotMap and the language).

The two drafts no test played are played now (tests/mp-sim-acts.test.js): a
premium border crossing into Hong Kong (buildLink with cnPrem) and the AI's own
script deployed as it wrote it while it drops most of the traffic (teDeploy with
slop, on a small network with one capped transit port), each compared byte for
byte with the page's clicks, and both read in the DRAFTS check (nothing is left
unplayed).

Runs, each file alone under an 8G cap: mp-sim-engine 16/16 (11.6 s),
mp-sim-contract 8/8, mp-sim-acts 5 + 1 todo (10.9 s), mp-sim-determinism 6/6
(53.4 s).
2026-11-09 20:47:31 +01:00
jvanvik
c5c7ea9508 Server-run engine: a restored network plays on exactly as before
A network loaded from its checkpoint drifted from the one that never stopped:
after a checkpoint on day 1, 2 or 5 its R differed at once (bizDem at Ashburn
6426 against 26, then cityDem, citySrv, tNet), and from the next day S too
(cash, eyeball growth). Three causes, each shown by a probe:

- R after a day is the R of the day's last routing, made before the day's end
  changed S (customers who left or signed, moods, bills, transit history) and
  routed from the memory of the R before it. srvLoad routed the saved S again
  from R's own memory: another R, which the next day reads (sfOf for eyeball
  growth, the peering share). Every compute() in the realm now keeps, for the
  R it makes, S as it routed it (one JSON per top-level key, 0.8 ms big) and
  the memory it started from. rt carries that routing point (pt: the memory,
  the keys that changed since, the keys S did not have then), and srvLoad
  routes that S again from it: the same R, byte for byte. The old restore test
  checkpointed after day 10, where a failure or event routed again at the end
  of the day and nothing changed after it: it passed by luck.
- The sales desk's tried offers (UI.slTried, an offer that did not fit waits a
  week) live outside S. A fresh realm forgot them and the desk tried other
  offers. rt carries them (sl); a day mark keeps them and the in-place restore
  after a throw puts them back (before, a parsed S dropped them).
- A save and a reload give some keys back in another key order (a customer's
  defaults go last, the intern's week tally in its own order), and the week's
  report lists the tally in that order. At a checkpoint rt carries the live
  text of each such key (ord, srvGate's round trip, about 8 ms), taken on load
  only where it holds what sanitize() made of the save. A fresh realm after a
  timeout takes the mark's own text the same way (live).

The plain game drifts across a save and a reload too (jsdom: from the first
day after it; the page routes a reloaded S with an empty R and keeps no desk
memory): a game property, so the engine carries what is missing.

The envelope keeps its shape {s, uidN, rt, last, day}; rt keeps R's own route
memory, so an older checkpoint loads as before (close, not exact). Big network:
the checkpoint grows from about 42-49 KB gzip to 54-79 KB (rt 7-29 KB; the
whole R would be 375 KB), srvSave from 2-2.5 ms to 7-8 ms, a load by one
routing. A day with its mark costs the same (big 39.8 -> 37.8 ms, busy
90.0 -> 88.8 ms, within noise: the day mark joins per-key JSON now, which pays
for the routing point).

Tests: the restore test checkpoints after days 1, 2, 5 and 10 of the scenic big
network and of the same network busy (intern on every order, the desk signing
on its own, failures, events, an attack), those after days 2 and 10 right after
an action and a what-if (sent to the copies too), each compared byte for byte
at once and 10 days later (SIM_LONG: after every day to 30, 60 checkpoints). It
fails on the old engine, and without any one of the three parts. The retry test
plays busy networks (fails without the desk memory in srvRestore). The market
checkpoint compares bytes too; canon() is gone.
2026-11-09 20:16:38 +01:00
jvanvik
065c922287 Server-run play, part 2: the engine
The engine that runs every network of a server session (server-sim design,
section 4): each network in its own vm realm, realms in a small pool of
worker threads, a promise API for the session host. Nothing uses it yet;
the plain build and the game are unchanged (no file under src/).

- scripts/lib/mp-sim-realm.js: one network in a DONT_CONTEXTIFY realm
  running the plain page's game (never the MP page's client folder), with
  recording DOM stand-ins, no-op timers that count, and the srv... calls:
  load from a checkpoint (route memory seeded, then routed), begin a new
  network, marks and restores, the day as a page runs it, the action
  pipeline with drafts, arming, the sanitize gate and rollback, what-ifs
  that put S, R, the dice and uidN back, pulls with top-level key diffs,
  slim R and flow records on demand, fx captured and bounded, saves, cards.
- scripts/lib/mp-sim-worker.js: realms per session in one worker, days one
  network at a time with other messages between them, loads and unloads
  waiting for a session's day, time limits; a throw goes back to the mark
  in place, a timeout gets a fresh realm (new epoch) carrying the old R, a
  day is retried once and the network broken after a second failure.
- scripts/lib/mp-sim-engine.js: createEngine() with the fake engine's API,
  placement by realm count, every pending call rejected when its worker
  exits, a worker of its own after two exits in an hour; snapWith() joins
  the host's clock, roster and since to a realm's snapshot.
- Contract: a pulled snapshot is the realm's half (the host adds clock,
  roster, since); haptic fx carry milliseconds; the fake engine follows,
  with close() and snapWith(). The button crawl moved to tests/lib.
- Tests: realm vs jsdom byte for byte over 60 scenic big-network days with
  actions; restore; retry after a throw or a timeout; realm order; a timer
  audit and the gate over a sweep of sent buttons (SIM_LONG=1: a year and
  every button); captures, rollbacks, asks, selection reports, ordering,
  bounds, pollution, worker exits; the act tables against the game.

The gate found two keys the game writes in a shape a reload repairs (buyTE's
S.te, mkLeave's S.monkey): game fixes for the integrator, kept visible as a
todo test and a list in the audit.

Rebased onto the 3.3.0 merge (61e31c5): the engine itself needed no change. A
realm still routes in the game's own code (no route worker, no saves), the
route cache that now keeps list orders from day to day leaves a checkpoint and a
retry the same, and the capacity market and moved customers play in a realm as
on a page, under the strict gate. The tests needed three things: an old save's
first days earn the new traffic badges, which the unwatched network's snapshot
keeps; the restore test compares JSON text with sorted keys instead of a
deep-equal of two networks (a diff of gigabytes when they differ); and the crawl
opens the windows that move a customer, with a new test playing the market's and
the path moves' acts, 40 days of sales and a checkpoint against the page.
2026-11-09 19:30:38 +01:00