fix(interact): widen click dedup window to stop cross-tick double-firing

[Arbousier1]

Problem

Players observed that a single physical click on any ray-based table control (join seat, hand tile, ready toggle, viewer commands) was being processed twice. For hand tiles this was especially destructive because the click toggle is select → discard, so one press discarded immediately.

Root cause

PlayerUseUnknownEntityEvent (fired by the Sparrow Heart client-only click proxies) does not implement Cancellable, so onClientInteractionProxy cannot suppress the event. Paper then fires a follow-up PlayerInteractEvent for the same physical click, which onFlatDisplayInteract processes again. The previous dedup window was 40ms, which is shorter than one server tick (50ms); when the two events landed on different ticks the second one slipped through and the action ran twice. The right-click + ARM_SWING pairing has the same cross-tick failure mode.

Fix

Raise the dedup windows in TableEventCoordinator and MahjongTableManager from 40ms to 150ms (3 ticks) so the follow-up event is always caught:

• \TableEventCoordinator.DUPLICATE_DISPLAY_ACTION_WINDOW_NANOS\ — first dedup layer, covers all action types
• \MahjongTableManager.DUPLICATE_DISPLAY_ACTION_WINDOW_NANOS\ — second dedup layer
• \MahjongTableManager.DUPLICATE_HAND_TILE_CLICK_WINDOW_NANOS\ — third dedup layer, hand-tile specific

The fix is uniform across all \DisplayClickAction\ types because \sameDisplayAction\ compares \�ctionType + tableId + ownerId + tileIndex + seatWind + command, so JOIN_SEAT / TOGGLE_READY / PLAYER_COMMAND benefit from the same window.

Safety

• Sichuan exchange phase already bypasses hand-tile dedup (!session.isSichuanExchangePhase(ownerId)), so rapid exchange clicks are unaffected.
• 150ms is well below human double-click perception for distinct actions; legitimate consecutive clicks on different tiles / seats remain distinguishable because \sameDisplayAction\ requires field-level equality.
• No behavior change to game rules — only the dedup timing window moves.

Verification

• \compileJava\ passes.
• \TableEventCoordinatorFlatInteractionTest\ passes, including:
  • 
    ightClickAndArmSwingBothCancelButExecuteTheSameActionOnlyOnce\ (existing dedup guarantee)
  • \ownedClientProxyUsesTheSameExactRayActionAndSwingsForRightClick\ (proxy path)
• \SessionHandSelectionCoordinatorTest\ passes (select → discard toggle semantics preserved).
• \MahjongTableManagerTest, \TableEventCoordinatorDisconnectTest, \TableEventCoordinatorSeatAttendanceTest\ pass.

This is a behavior-preserving bug fix; no AB gate needed.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 082f1adcb6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve fast same-tile discard clicks

With these windows at 150ms, a legitimate quick second click on the same hand tile is treated as a duplicate before it reaches the normal select→discard path in SessionHandSelectionCoordinator.clickHandTile: both clicks have the same table/owner/tileIndex, so clickTile returns as if handled and never calls session.clickHandTile to discard. This only affects players who intentionally double-click the selected tile within 150ms, but that is a plausible way to discard; the proxy follow-up needs to be filtered without debouncing deliberate second physical clicks for this long (and the matching display-action window in TableEventCoordinator has the same effect).

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve fast same-tile discard clicks

With these windows at 150ms, a legitimate quick second click on the same hand tile is treated as a duplicate before it reaches the normal select→discard path in SessionHandSelectionCoordinator.clickHandTile: both clicks have the same table/owner/tileIndex, so clickTile returns as if handled and never calls session.clickHandTile to discard. This only affects players who intentionally double-click the selected tile within 150ms, but that is a plausible way to discard; the proxy follow-up needs to be filtered without debouncing deliberate second physical clicks for this long (and the matching display-action window in TableEventCoordinator has the same effect).

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Deduplicate by tick/source instead of wall-clock time

If a server tick stalls for more than 150ms, the proxy event and its follow-up PlayerInteractEvent/PlayerAnimationEvent can still land on consecutive ticks but exceed this wall-clock window, so isDuplicateDisplayAction returns false and the same physical click is processed twice. Since the duplicate being fixed is explicitly a cross-tick event pair, this should key off the event source/tick (or another deterministic click marker) rather than assuming three normal 50ms ticks is an upper bound.

Useful? React with 👍 / 👎.