Parser: detect name registration from user input during setup
(sender == name constraint prevents false matches on gameplay messages).
Bridge: extract _process_buffer() to handle leftover lines after
_wait_for_players returns, preventing messages from being stuck.
Player bots: proactively send name on join (not just first player
sending count) so they respond even if they missed the 'say me' prompt.
All tests pass (1551/1553 parser checkpoints, 38 player unit tests).
Bridge changes:
- Wait for at least one user to JOIN before starting monop
- Ensures observer is in channel to see all setup messages
Parser changes:
- Handle 'Player N, say me' even without prior 'How many players?'
- Infer num_players_expected from highest player number seen
- Emit state during setup phase
run_game.py changes:
- 3s stagger between bot joins so setup is visible in web UI
- Observer connects before bots to catch all registration messages
Parser changes:
- Track num_players_expected from user input after 'How many players?'
- Create placeholder players ('Player N') on 'say me please' prompts
- Replace placeholder names when real names appear in roll lines
- Emit state during setup phase (was returning None)
- Include phase and numPlayersExpected in state JSON
UI changes:
- Empty slots shown as dashed/dimmed panels with '?' token
- Placeholder players shown as 'Registering...'
- Registered players shown with checkmark and ,500
- Status bar shows 'Setting up · 2/3 players registered'
Same top-banner pattern as the stale indicator but with red accent
and 'Connection lost — retrying every 2s...' message. Page gets
subtle greyscale/dim instead of a blocking overlay.
Subtle top banner with dimmed page (not a blocking overlay like the
full disconnect state). Shows elapsed time since last update.
Separate from the hard disconnect which triggers after 3 fetch failures.
Debug panel: added 'Stale' button to test.
- Zero state: bouncing dice + message when no game exists
- Disconnected: greyed overlay after 3 consecutive fetch failures,
with retry countdown
- Game over: gold banner with winner name + confetti animation,
detected from log ('WINS!') or single remaining player
- Debug panel: Ctrl+Shift+D toggles hidden panel with buttons to
test all four states (zero, playing, gameover, disconnected, reset)
- Also fixed player panel to use p.number for current-turn matching
and property ownership display
The C code picks a starting player via dice roll but doesn't reorder
the player array. The parser now tracks _first_player_idx and rotates
the players list in get_state() so the UI shows them in actual turn
order.
The root cause: checkpoint handler queued .trade via say_delayed, but
monop sends '-- Command:' before processing the trade. The old handler
reset in_trade and queued .roll, so both commands got sent.
Fix: checkpoint handler no longer sends commands. The -- Command:
prompt is where the bot decides to trade or roll, matching how monop
actually works (-- Command: is the interactive prompt).
Also: no trades while in jail (jail has no -- Command: before roll).
Updated all tests to reflect the new flow.
- Handle 'Which player do you wish to trade with?' prompt when
in_trade is True (was silently returning, causing bot to roll
during an active trade prompt)
- Simplify property trade offers to always say 'done' (cash-only
trades) to avoid getting stuck on property list prompts
