Replace the pinned 2018 undertheseanlp wordlist with kaikki.org's current
wiktextract export of the Vietnamese Wiktionary, read with --kaikki. The
same authors and website, eight years fresher: 34,813 words instead of
26,845, with every graph metric up and bot game length unchanged.
The export is fetched fresh for every build and is not pinned, by the
owner's decision: kaikki keeps no dated snapshots, so a checksum would
break weekly. The builder therefore hashes the file as it streams it and
records source_sha256, source_rows and source_fetched_at in meta; the
fetch downloads to a .part name and renames on success; a truncated or
non-JSON body fails the build, and the --min-words floor rises to 30,000.
DICT_SHA256 and verify-dict are gone; the Makefile, Dockerfile and the
builder's URL constant are held in agreement by a test.
Current Wiktionary text is CC BY-SA 4.0, so the data licence returns to
4.0: data/LICENSE is restored, and NOTICE, ATTRIBUTION, README, the image
docs and the in-game footer credit Wiktionary tiếng Việt's contributors
and wiktextract/kaikki.org. builder_version becomes 3 for the changed
meta contract.
Both themes use #d97757 for the accent. Buttons carry the dark ink colour instead of white, since white on this orange falls short of readable contrast. The favicon follows.
Replace the 179 MB minhqnd SQLite aggregate with the 4.8 MB
undertheseanlp/dictionary JSONL, pinned by commit and SHA-256, reading
only rows tagged "wiktionary". The two other wordlists in that file are
never read: hongocduc is GPL and would force a relicense, tudientv is an
unlicensed derivative of a commercial dictionary.
build-dictionary gains --merged and --sources (names validated, default
wiktionary), a shared finish() tail, and meta rows for the source commit
and the sources kept and excluded. The SQLite --in path, its schema
auto-detection and their tests are removed. Fixture builds now record
that they carry no upstream data instead of inheriting a licence string.
The --min-words floor moves from 40,000 to 20,000; the corpus is 26,845
words, down from 48,216, all of the loss being words absent from the
2018 Wiktionary scrape. Capitalization is not a filter.
The data licence follows the source text: CC BY-SA 3.0 Unported, which
is what vi.wiktionary.org carried in 2018. LICENSE, NOTICE, ATTRIBUTION,
the README, the image docs, the builder's meta string and the in-game
footer all name Wiktionary tiếng Việt's contributors as the authors and
undertheseanlp as the intermediary. The Makefile/Dockerfile pin test now
also checks the commit the builder stamps into the database.
Measured against the shipped corpus rather than estimated: viwiktionary is
already a third of what we ship, and undertheseanlp is bigger and denser but
carries an unlicensed Vietlex derivative in one of its three branches.
The suite only ever talks to localhost, and a listener on every interface
is what makes Windows Firewall ask to allow the binary — which `go run`
rebuilds to a new temporary path each time, so answering never sticks.
Opening the game is an advantage — the first player picks from a whole
syllable, everyone after them plays what is left — and it went to whoever
created the room, making them favourite in every game of a series.
The seat list is rotated by a random offset before the engine sees it, so
everybody still plays in the order they sat down and only the starting
point moves. A bot room keeps the human opening.
A turn arriving pulled focus into the word field even mid-sentence in the
chat, dropping the rest of the message into the game. Focus is now only
claimed when no other text field holds it; syllable seeding is unchanged.
A room screen wide enough for it is two columns now: the game or the
lobby on one side and the room's conversation on the other, so neither
has to be scrolled past to reach the other. A phone keeps one column,
the game first, with the chat folded behind an unread count while a game
is on. One chat panel spans both phases by staying in the same place in
the markup — a remounted panel reopened having read nothing, and handed
the reader the lobby's conversation back as unread mail.
The room also keeps the score of the series. PlayerSlot carries the
games a seat has won since the room opened, credited before the lobby is
broadcast so the players see it the moment a game ends, and shown in the
lobby and beside each player's score on the board. It belongs to the
seat rather than to the room, because vacating one is exactly when the
name on it stops meaning the same person.
The conversation reads as a log — one line per message, "name: text",
each seat in its own colour — rather than as a stack of bubbles, which
stops being legible once four people are talking. ChatMessage carries
the seat that spoke, so a line is coloured by what the server said
rather than by matching display names, and a line whose seat has been
vacated is in nobody's colour.
The board's word button gets a test id of its own: it and the chat's
send button both read "Gửi", and the two are now on screen together.
A room holds up to four people and needs two to start. Both numbers are
server constants sent to the client in RoomState, so the lobby draws
whatever the server allows and widening a room is a server change alone.
Failing a turn eliminates that player rather than ending the game. The
syllable and the used words survive them, the turn passes to whoever is
next, and the last player standing wins. Two seats is that same rule seen
from close up, which is why there is one implementation of it and not two.
A dead end still costs the first player to face it their own clock, as
before: they get their turn, and lose it. Everyone behind them has already
seen that board, so they go out together rather than each sitting out a turn
limit they cannot use — which leaves the player who closed the position
standing, the same outcome two players get.
A player who is knocked out keeps their seat. They watch the rest of the
game, chat included, with only the word input gone, and everybody lands back
in the same lobby when it ends. The result screen is the whole table, ranked
by who outlasted whom, with each score reported beside the place rather than
deciding it.
The turn clock is deliberately not paused for a seat that has dropped, so a
player who loses their connection on their own turn loses it the way anybody
else would. Their reconnect window decides only whether they are still in the
game afterwards. Any number of windows can be open at once, settled by one
timer armed for the nearest of them.
Starting waits for every guest, not merely the first: a room of four that
began on one yes would have dealt three people a turn they never agreed to.
Kicking names a seat and is still refused on a player who is ready, and on
the owner's own — leaving is what an owner who wants out does, and it hands
the room on. Joining stays a lobby thing: a room with a game running turns a
latecomer away even with seats going spare, because there is no way to hand
somebody a game already in progress.
BREAKING CHANGE: RoomState, TurnUpdate and GameOver lose the fields that
could only ever describe a second player, OpponentLeft is retired in favour
of presence on RoomState, and suggestions move to the new PlayerEliminated —
they describe the position that beat a player, which by the end of a longer
game is nobody else's position. ProtocolVersion goes to 2, so a client built
against 1 is refused with a readable error rather than decoding a frame that
now means something else.
Chat belongs to the room rather than to a game, so it works in the lobby
while they agree on one, during it, and in the lobby it ends in.
A player is replayed what was said while they held their seat. That is what
a refresh restores, and it is also the boundary: a room code is pasted into
group chats by design, so somebody who redeems one starts at silence rather
than reading what the last two people said. A seat records where the
conversation stood when it was filled; the room keeps twenty lines and no
more, so a room that lives all day cannot grow.
Text is untrusted input rendered in a stranger's browser, so it goes through
the filter nicknames already used — now with a cap on stacked combining
marks, which that filter admitted. Twenty runes made mark stacking a
curiosity; two hundred make it a glyph cluster tall enough to cover the
board, and it would sit in the history being replayed to everyone who
followed.
Talking is not playing. A chat message does not reset the room's idle clock,
or one open tab could hold a room and its code for the life of the process
by typing into it once every nine minutes. It does not spend the move budget
either, and a line to a player who cannot keep up is dropped rather than
allowed to close their session — losing a line is recoverable, losing a
session mid-game costs them the game. A history is not droppable that way:
it is the frame that corrects a whole panel, and there is nothing behind it.
When a seat is vacated its words stay and its author goes, name included,
and the player who stayed is re-synced rather than left holding a name that
the next person through the door could ask for.
A room used to be a wrapper around one game: joining started it, and the
room died with it unless both players accepted a rematch inside thirty
seconds. It is now a lobby that outlives its games.
Whoever created the room owns it and the other seat is the guest. The guest
readies and the owner starts; the owner has no readiness of their own,
because starting is the same statement. A finished game returns both to the
lobby, where the next one is agreed exactly as the last was — the readiness
that started a game is spent with it.
A guest takes their readiness back before leaving, which is deliberate
friction: a player the owner is waiting on should have to say so before
walking away. The owner can free the seat of a guest who is not ready, and
not of one who is — readiness is a commitment, not an inconvenience. An
owner who leaves hands the room to whoever is left, unreadied, because they
are the one who starts now.
Something has to bound a room that outlives its games: the last player out
closes it, as does ten minutes in a lobby nobody started a game in. A
dropped connection is still not a player leaving — the seat is held for the
reconnect window in the lobby as well as mid-game, so a refresh no longer
costs somebody their room, and a resume lands in the lobby it left.
The rematch handshake is retired, and RoomCreated and RoomJoined go with it.
All three described part of what RoomState now describes in full, and three
messages for one lobby is three ways for a client to hold a view of it the
server never had. One snapshot per recipient, broadcast from the one place
that knows an input is finished, so no handler can forget to send it.
Closing the position no longer wins the game on the spot. The player handed
a syllable that starts nothing keeps the turn they were given and loses it to
the clock, the way they lose any turn they cannot answer — the win used to
land before they had seen the board at all. A turn lost to a position nobody
could answer is reported as the dead end it was rather than as time spent
thinking.
The bot is the exception, and stays one: it has no clock to spend, so the
room settles its dead end the moment the search comes back empty. NoMove
does that without dressing it up as a resignation the bot never chose.
GameOver now carries a few of the words the position still had, filled for
the losing player only — the winner was not the one who was stuck. An empty
list on a loss is the other half of the message: nothing could have been
played, so the panel says so instead of listing nothing.
Three things a player asked for after playing:
The field now arrives holding the syllable the next word has to start with,
so only the half that is still in question gets typed. It is written once per
turn and never over an existing draft, which keeps the field uncontrolled and
leaves Telex and VNI composition alone.
The chain lists the newest word first. The move that decides what to play
next is the last one, and it belongs where the eye lands rather than at the
end of a list that has to be scrolled.
A finished game can be downloaded as a plain-text transcript: the chain in
playing order, who played each word, and the final score. The export keeps
playing order because a chain read on its own only makes sense from the
opening word forward — newest-first is a reading aid for the live board, not
the shape of the game.
A Windows checkout with core.autocrlf enabled handed buf a CRLF schema, so
an empty comment line reached the codegen plugin as a carriage return rather
than an empty string. The plugin pads non-empty lines with a space, and the
committed output kept a trailing space that an LF checkout never emits,
breaking the generated-code sync check.
One distroless image of about 25 MB carries the binary, the built frontend and
the derived dictionary. The 179 MB upstream release is downloaded in a builder
stage and never reaches the final image; the derived wordlist is copied in as
its own layer alongside its licence, attribution and notice, because CC BY-SA
4.0 applies wherever that data is distributed and an image is distribution.
FIXTURE_DICT=1 builds the same Dockerfile against the checked-in word sample,
so the image is built and smoke-tested on every push rather than only at
release. An image built only at release time is an image that breaks at release
time.
CI runs the Go suite under race detection, the frontend type check and tests,
the browser suite, and the image with its licence assertions. The wire contract
keeps its own workflow; the test steps it duplicated were removed from it.
docs/deployment.md covers configuration, the reverse-proxy settings that each
break the game in a way that looks like something else, and what a restart
costs.
Twenty-one Playwright specs against the real binary serving the real bundle: a
bot game with its rejections, an online game across two browser contexts, and
the reconnect path. The server plays against the fixture dictionary, so nothing
here downloads the upstream release.
Every wait is on a state the server produced — the turn indicator, the required
syllable, a word appearing in the chain — never a sleep. Moves are chosen by
reading the same wordlist the server has rather than scripting a fixed
sequence, because the bot is entitled to pick a different reply.
Two things that do not work are recorded where the next person will look.
Offline emulation does not cut a loopback socket, so a test that went offline
kept playing happily. Routing the socket cuts it for real, but the close does
not reliably reach the server, so the grace window never starts: those tests
take the page away instead.
Create a room and share a six-character code or an invite link; join by either.
The waiting room shows the code until an opponent arrives. During a game the
opponent's disconnection raises a banner with the grace countdown, and their
return clears it. When the game ends, the rematch prompt appears only while an
offer is actually open, so there is never a button that can only produce an
error.
The board is now one component both modes render, which is what makes online
play the same screen as the bot game rather than a second implementation of it.
Room codes are normalized by stripping separators and reading the code out of a
pasted invite link, and nothing else. Filtering out every character outside the
alphabet would turn any sentence into six plausible letters and send the player
to a room that was never theirs.
The socket now opens when the player asks for a room, not on arrival: the
handshake carries the nickname once, and a socket opened before they typed it
introduced them under whatever was stored beforehand. A tab that already holds
a session reconnects immediately instead, so refreshing mid-game returns to the
game. A resume the server cannot honour is dropped quietly rather than opening
the lobby with an error the player did nothing to cause.
The client also notices a socket that fails without closing. Nothing did
before, so the page kept showing a live connection and a running countdown over
a socket nothing could reach. A tick that fired late is ignored, because a
backgrounded tab has its timers throttled and would otherwise close a healthy
connection every time the player switched away.
build-dictionary gains a --words mode that reads a plain list instead of the
upstream database. Everything after that — filtering, alias generation, writing
and verification — is the code the real build uses, so a fixture cannot drift
into being shaped differently from what the server loads.
testdata/fixture-words.txt is hand-written rather than extracted, so no test
artifact carries the upstream release's licence. Its graph is built around one
hub syllable: it is the only one with enough continuations for the server to
open a game on, so every game starts on a word ending in it and a scripted test
always knows the first answer.
A finished room now stays alive to ask both players whether they want another,
and restarts with a fresh opening word once both do. Only a room with two
connected humans offers one: a bot has nothing to negotiate, so a bot room
closes the moment its game ends rather than leaking a goroutine and an engine
per finished game.
turn_seq no longer restarts at one. A rematch reuses the same connections, so a
submission still in flight from the previous game could otherwise match a turn
in the new one and be applied to it.
The end-of-game decision sits after the whole select, so every way a game can
end reaches it. Opening the offer from the message arm alone meant the turn
clock — the most common natural ending — closed the room with nothing to accept.
A resume is refused into a finished game, including one waiting on a rematch
answer, so the room now retires the connection being replaced only once it has
agreed to the swap. Retiring it up front ended the game the client was trying
to rejoin, which a duplicated tab was enough to trigger.
attach releases the room it is leaving. Nothing else told that room the
connection had gone, so a session asking for several rooms stranded all but the
last, each parked in select holding a goroutine and a room code for the life of
the process.
Also: RequestRematch is rate limited, because it is the only client message
that fans out to both players and an unbounded one lets a burst fill the
opponent's outbox until their session is closed for falling behind. And a
resume announces itself to the opponent, who was otherwise left watching a
disconnect banner for someone already playing again.
RequestRematch asks to play the same room again; RematchState tells each
player where both answers stand, rendered per recipient so neither client has
to work out which acceptance is whose.
There is no decline message. Leaving is the decline, and the server already
learns about that from the socket closing, so one message and one timeout
cover every way a rematch does not happen.
Both tags are new, nothing existing moved, and a phase-6 client that has never
heard of rematch_state ignores it rather than failing to decode.
Assets under _app/immutable carry a content hash in the name, so a changed file
is a changed URL and the old one can be cached forever. The shell cannot: it
names those hashed assets, so a copy cached across a deploy points at files that
no longer exist and the app loads into a blank page with nothing in the log.
The board, the home screen and the game-over panel, in Vietnamese, served by
the Go binary as a static single-page app.
The store is a reducer over ServerMessage and computes nothing. Validity, turn
order, scores and the result are read from the wire, which is what lets one
screen serve the bot now and online play later. Every Vietnamese string lives in
one module, including the map from RejectReason to a message, so the server can
send UI keys instead of prose.
The word field is uncontrolled. A Telex or VNI input method composes a diacritic
across several keystrokes, and writing the value back on each one cancels the
composition. It is read on submit and cleared only there.
The countdown is drawn against the server's clock, estimated from the ping round
trip, and settles 300ms early so the ring never claims more time than the server
allows.
The screen owns the socket and the game while it is mounted, and gives the
current game up on the way out. Asking for a game is stored intent rather than a
condition inferred from the board being empty: clearing the board for a rematch
is that same condition, so the inference sent a second StartBotGame and the
server built two rooms that then destroyed each other.
A test greps the built bundle for dictionary words and holds a size budget, so
the wordlist cannot reach the browser unnoticed. Another reads the error codes
out of the Go transport and fails when one has no Vietnamese message.
One goroutine owns each room and its engine. The room goroutine starts before
anyone is seated and seating is itself a message, so reading run() is a complete
proof of the concurrency contract rather than a convention to uphold. The bot
searches a frozen copy of the board instead of the live engine.
Reads carry no deadline; liveness is ping-based, because a read timeout cannot
distinguish a healthy player idling in the lobby from a dead socket.
The hub no longer binds a joiner to a seat before the room decides whether to
seat them. Anyone holding a room code could previously resign or play on a
seated player's behalf, and the room code is the only credential online 1v1 has.
cmd/noitu-server serves the API and, when NOITU_WEB_DIR is set, the built
frontend, with unknown paths falling back to index.html for client routes. All
configuration is environment-only and every variable has a working default.
proto/noitu/v1/game.proto is the single source of truth for every WebSocket
message. buf generates Go types into server/gen and JavaScript types into
web/src/lib/proto; both trees are committed so building needs no codegen
toolchain.
The Go suite emits binary fixtures into proto/testdata and the JavaScript suite
decodes the same bytes, so the two generated clients are checked against one
artifact rather than against each other's assumptions. CI lints the schema,
rejects breaking changes against main, and fails when the committed generated
trees drift from the schema.
game.NumRejectReasons and game.NumEndReasons let the mapping tests prove every
engine reason has a wire value without guessing where the enum ends.
The engine is transport-free: no sockets, no protobuf, and no clock of its
own. Callers pass the current time in and read the deadline back, so every
rule is testable without a timer. One goroutine owns a game.
Validation resolves the word before checking the chain link. Roughly a
third of dictionary aliases move the first syllable, so "sy hai" resolves
to "si hai"; checking the link against what the player typed would reject
legal moves. The used-word set is keyed on the canonical form, which also
stops the same word being played twice under two spellings.
An opening whose last syllable starts nothing is refused. It would hand
the first player a game already lost, with no move and no reason, that
resolves only when the turn timer expires and then reports a timeout.
Three bot difficulties, separated by how far they look ahead rather than
by how willing they are to win: random, one-ply greedy, and depth-limited
negamax with alpha-beta. Strategies see a read-only view of the board, so
a bot cannot bypass the same validation a human's move goes through.
Simulation on the real corpus, alternating sides across 60 games per
pairing: hard beats easy 98%, medium beats easy 87%, hard beats medium
65%. Decisions take at most 19ms against a 150ms budget.
Three defects surfaced only under simulation. Withholding the winning
move from the middle bot, as first designed, made it lose to the random
bot 97% of the time. The search evaluated leaves with an inverted sign,
so it hunted for positions where it was about to be trapped. And the
rate limit on taking an instant win did nothing, because declining the
shortcut let the search rediscover the same move.
Games against the hard bot end after about three moves versus fifteen
for two random bots: with 1,814 dead-end syllables an instant win is
usually available. Tunable, and flagged for playtesting.
Load the whole dictionary into maps at Open and close the database before
Open returns. The plan called for per-lookup SQLite, but a round-trip
benchmarked at 55us against 8.9ns for a map hit, and the hard bot in a
later phase explores hundreds of candidates inside a 150ms budget. The
in-memory form is also simpler: no connection pool, no prepared
statements, no tail latency. Costs ~70ms and ~7.8MB at startup.
Resolve returns the canonical word, never the spelling the player typed.
Canonicalization moves either end: about half the aliases differ in the
last syllable and more than a third in the first, so "sy hai" resolves to
"si hai". FirstSyllable and LastSyllable report the canonical's ends, and
the engine must chain on those or it will reject legal moves.
WordsStartingWith yields an iterator rather than the backing slice. A
caller could otherwise sort, shuffle or append into dictionary state:
verified that a write landed in the store, that most buckets have spare
capacity for append to scribble into, and that concurrent callers race.
Open validates what it loaded against the builder's recorded word count,
cross-checks every out-degree against the words actually indexed, and
rejects orphan aliases. A truncated database otherwise opens cleanly and
the server starts, rejects every word, and fails every room creation.
RandomOpeningWord picks from a pre-sorted slice by binary search instead
of rebuilding a filtered copy per call, cutting room creation from 374us
and 720KB to 18ns and no allocation.
Escape the database path when building the URI: SQLite reads # as a
fragment delimiter, so an unescaped path opens a different file and
reports a misleading schema error.
The store does not log. A library writing to the global logger fights
structured logging later, and the caller has WordCount, AliasCount and
License to state the CC BY-SA attribution itself.
Reduce the 179 MB minhqnd/dictionary SQLite release to a ~3 MB game
wordlist: 48,216 Vietnamese words of two or more syllables, indexed by
first and last syllable with an out-degree table for dead-end detection.
Source schema is auto-detected rather than hardcoded, since it is someone
else's release artifact; explicit flags override it and are validated
against the real tables, because SQLite silently reads an unknown
double-quoted column as a string literal.
Accept a word only if every syllable fits Vietnamese phonotactics. An
alphabet check is not enough: "credit card" and "come out" use only
letters Vietnamese has, and the multilingual source tags them as
Vietnamese. Onset matching backtracks so the gi digraph does not swallow
the nucleus of common words like "gi", "gin" and "gi" (rust).
Record accepted spelling variants in an alias table rather than solving
tone placement at runtime. Tone shifting applies only to open oa/oe/uy
syllables, since "hoan" and "hoai" have a single correct spelling, and
"qu" is a consonant onset. The i/y alternation uses an onset allowlist
plus explicit pairs, because it is lexical rather than productive.
Variants are generated as a cross product over syllables so a word with
two variable syllables still offers the fully modern spelling.
Build to a temporary file and rename only after commit, so a failed run
cannot leave an empty database where a good one was, then re-open the
result and verify its invariants on disk.
Licensing: the derived data is CC BY-SA 4.0 and stays a separate artifact
from the Apache-2.0 code, loaded at runtime and never embedded. Ships
NOTICE, data/LICENSE and an attribution file recording every change.
Research the Vietnamese noi tu word-chain game and plan a 7-phase web
implementation: SvelteKit frontend, Go backend, WebSocket transport with
Protobuf framing, and a server-authoritative dictionary over SQLite.
The server validates every move so the browser never holds the wordlist,
which keeps player-vs-player cheat-resistant and lets the bot and PvP
modes share one rule implementation.
Records the validated decisions: words of two or more syllables linking
on first and last syllable, a 20s turn limit, user-typed nicknames, and
Docker deployment behind a reverse proxy.
tiennm99