## Description:
clientId replay bugfix (was picking first clientID in the array)
https://discord.com/channels/1359946986937258015/1479543573404844042
## Please complete the following:
- [x] I have added screenshots for all UI updates
- [x] I process any text displayed to the user through translateText()
and I've added it to the en.json file
- [x] I have added relevant tests to the test directory
- [x] I confirm I have thoroughly tested these changes and take full
responsibility for any bugs introduced
## Please put your Discord username so you can be contacted if a bug or
regression is found:
w.o.n
## Description:
TL;DR: it's faster.
buildableUnits is called via PlayerView.actions from UnitDisplay (each
tick without TileRef), BuildMenu (each tick when open), MainRadialMenu
(each tick when open), PlayerPanel (each tick when open),
StructureIconsLayer (when placing a building from build bar),
NukeTrajectoryPreviewLayer (when placing nuke, on tick when tile
changes), ClientGameRunner (on click to attack/auto-boat or hotkey B or
G).
After https://github.com/openfrontio/OpenFrontIO/pull/3213 got merged,
the change with largest impact in
https://github.com/openfrontio/OpenFrontIO/pull/3193 was done in such a
different way that a new PR was needed
The idea in 3193 was to not always ask for Transport Ship from
buildableUnits. In such a way that very little extra data was send to
the worker. This had the biggest impact on performance (the idea was
months older btw, see
https://github.com/openfrontio/OpenFrontIO/pull/2295). Now, we do it the
other way around, by telling buildableUnits all unit types we want. Or
we want them all (undefined). The downside is more data is send in the
worker message. The upside is we have more options and can add more in
this PR.
This PR implements some of the leftovers in 3193 on top of 3213 and adds
further improvements.
(Some unrelated refactor/perf changes where moved out of this PR and
into already merged
https://github.com/openfrontio/OpenFrontIO/pull/3233,
https://github.com/openfrontio/OpenFrontIO/pull/3234,
https://github.com/openfrontio/OpenFrontIO/pull/3235,
https://github.com/openfrontio/OpenFrontIO/pull/3236,
https://github.com/openfrontio/OpenFrontIO/pull/3237,
https://github.com/openfrontio/OpenFrontIO/pull/3238,
https://github.com/openfrontio/OpenFrontIO/pull/3239)
- **GameRunner**, **WorkerMessages**: _playerActions_ and
_PlayerActionsMessage ._ Option to ask for no buildable units (null). It
now has 3 modes: get all actions and all buildings (units undefined),
get all actions and no buildings (units null), or get all actions and
specific building (units contains Unit Types).
- **GameRunner**: _playerActions_. fixes wrong assumption in PR 3213:
that only if units was undefined, we have to know canAttack.
ClientGameRunner wants to know both, in case of a click on non-bordering
land, to decide if it should auto-boat using a Transport Ship. So units
is not undefined (we only ask for Transport Ship now which has a
positive effect on performance for each click/tap) but we need canAttack
still.
Solved by removing the unit === undefined check before _canAttack_ in
_playerActions_.
- **GameRunner**, **GameView**, **WorkerClient**, **WorkerMessages**,
**Worker.worker**: added _playerBuildables_ / _buildables_ next to
existing _playerActions_ / _actions_. With above solved, there was still
no option to only get buildable units when the actions are not needed.
While **StructureIconsLayer**, **NukeTrajectoryPreviewLayer**,
**BuildMenu** and **UnitDisplay** need only that. To not make
playerActions more convoluted with more params or so, i've added a new
function _playerBuildables_ in **GameView** to only get buildable units
(**GameRunner** _playerBuildables_). _playerBuildables_ has 2 modes: get
all buildings (units undefined) or get specific buildings (units
contains Unit Types). Also update some comments that mentioned .actions
in **NukeTrajectoryPreviewLayer**.
- **ClientGameRunner**, **PlayerPanel**, **BuildMenu**, **UnitDisplay**,
**StructureIconsLayer** and **NukeTrajectoryPreviewLayer**: Since PR
3213, **StructureIconsLayer** and **NukeTrajectoryPreviewLayer** ask for
specific types of units from **GameView** _actions_ (**GameRunner**
playerActions). Now have the other files do the same. For example
**BuildMenu** asks for the new _BuildMenuTypes_ when it calls
._buildables_ and **ClientGameRunner** asks for UnitType.TransportShip
when sending a boat
- **ClientGameRunner**: canBoatAttack now accepts BuildableUnit[]
instead of PlayerActions so we can send it either actions.buildableUnits
or just buildables. Have functions call myPlayer.buildables(tileRef,
[UnitType.TransportShip]) when we only need a buildable unit and no
actions. Or myPlayer.actions(tileRef, null) when we need actions but no
buildable units. Or myPlayer.actions(tileRef, [UnitType.TransportShip])
when we need both actions, like canAttack, and a buildable unit. Then if
needed send either actions.buildableUnits or buildables to to
_canAutoBoat_ / _canBoatAttack_.
- **MainRadialMenu**: needs all player buildable unit types including
Transport Ship, so the _actions_ call argument for unit types can stay
undefined (unchanged) there.
- **MainRadialMenu**: now that **BuildMenu** uses _playerBuildables_
instead of _playerActions_, we must put data in
_this.buildMenu.playerBuildables_. And since we're not putting the
(unneeded) full _actions_ in there anymore, we can now put only the
needed and expected _actions._buildableUnits_ in it.
- **Game**, **PlayerImpl**, **StructureIconsLayer**: Typesafety and some
added perf: new type _PlayerBuildableUnitType_ (see also the below point
for how it is formed). So callers of _buildableUnits_ can never ask for
the wrong type like e.g. UnitType.Train because it doesn't return data
for that type. This type is now used in **PlayerImpl**, **BuildMenu**,
**RadialMenuElements**, **StructureDrawingUtils** and **UnitDisplay**
for that reason. And **InputHandler**, **StructureIconsLayer** and
**UIState** (little more on that in point below).
- **InputHandler**, **StructureIconsLayer**, **UIState**: In order to
make type safety work for GhostUnit.buildableUnit.type too (line ~217 of
StructureIconsLayer), changed type of interface _BuildableUnit_ to
_PlayerBuildableType_. Which is only more accurate. Same for and
this.structures and uiState.ghostStructure and with the latter,
_renderUnitItem_ in **UnitDisplay** and _setGhostStructure_ in
**InputHandler**. All Structures are of PlayerBuildableType (there are
even some in PlayerBuildables that aren't Structures, but it is much
more confined than UnitType).
- **Game**: Typesafety and some added perf: added _BuildMenus_ and
_BuildableAttacks_ in the same fashion that the existing StructureTypes
was already used (simplified it a bit too, with it renamed
_StructureTypes_ to _Structures_ and removed _isStructureType_). They
can be used with .types or .has(). _BuildableAttacks_.has() is used in
**RadialMenuElements**. _BuildableAttacks_ and existing _Structures_ now
make up _BuildMenus_. Which is used in **BuildMenu**,
**StructureIconsLayer** and **UnitDisplay**. Then _BuildMenus_ together
with UnitType.TransportShip make up the _PlayerBuildables_. Which is
used in **PlayerImpl** _buildableUnits_ (see point below). And with
_PlayerBuildableUnits_ we get the new _PlayerBuildableUnitType_ (see
above point on Game / PlayerImpl).
- **RadialMenuElements**: replace non-central ATTACK_UNIT_TYPES in
**RadialMenuElements** with centralized _BuildableAttackTypes_ too. Use
_PlayerBuildableUnitType_ for more type safety (can't by mistake add
UnitType.Train to its build menu). Make use of _BuildableAttackTypes_
instead of adding items hardcoded line by line in _getAllEnabledUnits_,
just like we already did since PR 3239 with _StructureTypes_. And use
_BuildableAttacks.types_ in the same fashion that existing
_isStructureTypes_ (now Structures.types) was already used elsewhere.
- **PlayerImpl**: _buildableUnits_
-- would do Object.values(UnitTypes) on every call. Now for better perf
directly loop over player buildable units by using _PlayerBuildables_
(see above point). In this way we also exclude MIRVWarhead, TradeShip,
Train, SamMissile and Shell so there are less unit types to loop through
by default. Since a player doesn't build those by themselves, they are
only build by Executions which use _canBuild_ directly and not
_buildableUnits_.
-- for more performance, do for loop instead of using .map and .filter,
no intermediate array needed nor callback overhead. We just loop over
the given units (which if undefined will contain _PlayerBuildables_).
Also pre-allocate the results array to get the most out of it, even if
V8 might already be very good at this.
-- cache config, railNetwork and inSpawnPhase so they can be re-used
inside the for loop.
-- cache cost inside the loop
-- it would check twice for tile!==null to decide to call
findUnitToUpgrade and canBuild. Now once.
-- eliminated double/triple checks for the same thing. It called
_findUnitToUpgrade_ (and with that _canUpgradeUnit_) and then _canBuild_
which both check if player has enough gold for the cost of the unit
type. And they both check if the unit type is disabled. Now we call
private functions _canBuildUnitType_, _canUpgradeUnitType_ to first do
checks on unit type level for early returns, and
_findExistingUnitToUpgrade_ to find existing unit without doing anything
extra. in a specific order to check everything only once. The public
functions _findUnitToUpgrade_ and _canBuild_ have an unchanged
functionality and we don't call them from _buildableUnits_ anymore.
-- would get _overlappingRailRoads_ and _computeGhostRailPaths_ when
canBuild was true. But this data is only meant for
**StructureIconsLayer** and it logically only uses it when placing a new
unit, not when upgrading one. Which is also commented on line 351 of
**StructureIconsLayer**. So, we now only get overlapping railroads and
ghost rails if we're not hovering to upgrade an existing unit.
- **PlayerImpl**: _findUnitToUpgrade_: unchanged functionality, but have
it call new private function _findExistingUnitToUpgrade_ to find
existing unit.
- **PlayerImpl**: _canBuild_: unchanged functionality, but have it call
new private function _canBuildUnitType_ to do the checks it first did
itself. And then new private function _canSpawnUnitType_ for the rest of
the checks. This way we can call _canBuildUnitType_ and
_canSpawnUnitType_ from _buildableUnits_ in a specific order to prevent
double/triple checks.
- **PlayerImpl**: _canBuildUnitType_: new private function to be shared
by _buildableUnits_, _canBuild_ and _canUpgradeUnit_ to be able do unit
type level checks in a specific order to prevent double/triple checks.
Via parameter knownCost, _buildableUnits_ can send it the cost it
already fetched so that it doesn't have to be fetched again. For caller
_canUpgradeUnit_, the isAlive() check (which was previously only done in
canBuild) is new but harmless, maybe even better to have also check
isAlive() on upgrade now that Nations are also upgrading which might
prevent some edge case bugs.
- **PlayerImpl**: _canUpgradeUnitType_: new private function to be
shared by _buildableUnits_ and _canUpgradeUnit_ to be able do unit type
level checks in a specific order to prevent double/triple checks.
- **PlayerImpl**: _canSpawnUnitType_: new private function to be shared
by _buildableUnits_ and _canBuildUnit_ to be able do unit type level
checks in a specific order to prevent double/triple checks.
- **PlayerImpl**: _findExistingUnitToUpgrade_: new private function to
be shared by _buildableUnits_ and _findUnitToUpgrade_ to be able do unit
level checks in a specific order to prevent double/triple checks.
- **PlayerImpl**: _isUnitValidToUpgrade_: new private function to be
shared by _buildableUnits_ and _canUpgradeUnit_ to be able do unit level
checks in a specific order to prevent double/triple checks.
- **PlayerImpl.test.ts**: because of the isAlive() check in which is new
for _canUpgradeUnit_ (see above at _canBuildUnitType_), the tests needed
to have the players be alive at the start, in order to pass.
- **BuildMenu**: use .find instead of .filter in canBuildOrUpgrade, a
function we already needed to change. This is faster and prevents an
allocation.
**PERFORMANCE**
As for calling ._buildables_ instead of unnecessarily getting
._actions_, there is an obvious win because there's less to send
calculate and recieve.
Also asking for only the needed buildings helps a lot (especially if
TradeShip isn't needed, see the difference in benchmark in original
#3193).
But the real-world impact is hard to measure. gave it a try in #3193 and
those results should be even better now.
Now testing only _buildableUnits_ performance in a synthetic benchmark,
we get these results. This is after other performance improvments so the
base is already better than it was in original #3193:
**BEFORE** (only buildableUnits itself)
<img width="602" height="96" alt="image"
src="https://github.com/user-attachments/assets/7770c0fa-a35e-42fc-90de-1de83242ec23"
/>
**AFTER** (only buildableUnits itself)
<img width="603" height="91" alt="image"
src="https://github.com/user-attachments/assets/a1578382-7010-4160-937c-7117bad18beb"
/>
## Please complete the following:
- [x] I have added screenshots for all UI updates
- [x] I process any text displayed to the user through translateText()
and I've added it to the en.json file
- [x] I have added relevant tests to the test directory
- [x] I confirm I have thoroughly tested these changes and take full
responsibility for any bugs introduced
## Please put your Discord username so you can be contacted if a bug or
regression is found:
tryout33
---------
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
## Description:
Removes the client-driven heartbeat loop and switches worker tick
execution to a worker-owned drain scheduler with batched game update
delivery.
## Why
The previous flow required the client to send a `heartbeat` every
animation frame just to keep the worker progressing turns. That had two
costs:
1. Simulation progress was coupled to browser frame cadence.
2. Catch-up periods produced many single `game_update` messages,
increasing message overhead and main-thread wakeups.
## What Changed
### 1) Remove heartbeat protocol
- Deleted `heartbeat` from `WorkerMessageType`.
- Removed `HeartbeatMessage` from `MainThreadMessage`.
- Removed `sendHeartbeat()` from `WorkerClient`.
- Removed the `requestAnimationFrame` keep-alive loop in
`ClientGameRunner`.
Files:
- `src/client/ClientGameRunner.ts`
- `src/core/worker/WorkerClient.ts`
- `src/core/worker/WorkerMessages.ts`
- `src/core/worker/Worker.worker.ts`
### 2) Add batched worker-to-client updates
- Added `game_update_batch` message type and `GameUpdateBatchMessage`.
- Worker now emits one batch message containing multiple tick updates.
- `WorkerClient` handles `game_update_batch` by replaying updates to the
existing callback in order.
Files:
- `src/core/worker/WorkerMessages.ts`
- `src/core/worker/WorkerClient.ts`
### 3) Move tick draining into worker
- Added a scheduler (`scheduleDrain`) and drain loop (`drain`) in
`Worker.worker.ts`.
- On each `turn` message, worker enqueues turn and schedules drain.
- Drain executes up to `MAX_TICKS_BEFORE_YIELD = 4` ticks per cycle,
then yields with `setTimeout(..., 0)`.
- Tick updates are collected into a batch and sent once with
transferables:
- `packedTileUpdates.buffer`
- `packedMotionPlans.buffer` (when present)
- If backlog remains, drain reschedules itself.
File:
- `src/core/worker/Worker.worker.ts`
## Behavioral Notes
- No server protocol changes.
- Ggame update callback contract remains the same (still receives one
`GameUpdateViewData` at a time in order).
- Ordering is preserved: `WorkerClient` iterates batch entries in
sequence.
- Error updates are still filtered from update delivery in the worker
batch path (same effective behavior as before for normal update flow).
## Expected Impact
- Fewer `postMessage` calls during backlog and burst turn delivery.
- Lower message overhead and fewer main-thread interrupts.
- Less dependence on UI frame timing for worker progress.
- Better catch-up stability due to explicit periodic yielding.
## Risk Areas
- Drain scheduling edge cases (re-entrancy / lost wake-ups).
- Mitigated with `drainScheduled`, `draining`, and `drainRequested`
flags.
- Larger per-message payloads due to batching.
- Bounded by `MAX_TICKS_BEFORE_YIELD`.
- Any assumptions in downstream code about receiving only `game_update`.
- Handled by adding `game_update_batch` support in `WorkerClient`.
## Please complete the following:
- [x] I have added screenshots for all UI updates
- [x] I process any text displayed to the user through translateText()
and I've added it to the en.json file
- [x] I have added relevant tests to the test directory
- [x] I confirm I have thoroughly tested these changes and take full
responsibility for any bugs introduced
## Please put your Discord username so you can be contacted if a bug or
regression is found:
DISCORD_USERNAME
## Description:
The ghost structure calls player actions each frame, which is costly
since it's checking for all possible actions.
This add a unit list filter in actions so if there are units it only
checks for buildability of those units.
Before:
Player actions takes 20-30% of the worker
After:
<img width="825" height="342" alt="image"
src="https://github.com/user-attachments/assets/36e47547-5028-4dc9-bc42-e17df4a87200"
/>
Player actions takes 1-3% of the worker
Both performances are relevant only when a ghost structure is selected
## Please complete the following:
- [x] I have added screenshots for all UI updates
- [x] I process any text displayed to the user through translateText()
and I've added it to the en.json file
- [x] I have added relevant tests to the test directory
- [x] I confirm I have thoroughly tested these changes and take full
responsibility for any bugs introduced
## Please put your Discord username so you can be contacted if a bug or
regression is found:
Mr. Box
## Description:
Make the unit display bar a proper unit build bar
Add shortcuts for all structures and units
Add ranges for ranged structures and units
Changed the shortcuts to use the key instead of the code for
internationalization purposes
<img width="285" height="517" alt="image"
src="https://github.com/user-attachments/assets/91bb01e6-e48c-4255-ace1-306af9cdc25b"
/>
## Please complete the following:
- [x] I have added screenshots for all UI updates
- [x] I process any text displayed to the user through translateText()
and I've added it to the en.json file
- [x] I have added relevant tests to the test directory
- [x] I confirm I have thoroughly tested these changes and take full
responsibility for any bugs introduced
## Please put your Discord username so you can be contacted if a bug or
regression is found:
Mr.Box
---------
Co-authored-by: evanpelle <evanpelle@gmail.com>
Co-authored-by: icslucas <carolinacarazolli@gmail.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
## Description:
Makes so that when clicking on the attack warning message in chat, the
camera focuses on the "average position" of the attack, instead of just
the player.
The average position is calculated by taking the average position of all
attacking border cells. It makes the calculation for every AttackUpdate,
which adds some calculations every tick, but it shouldn't affect
performance that much, as it's just a sum of coordinates.
If you have a better way of getting the averagePosition information
(calculating it only when necessary instead of every tick), it would be
great.
closes#703
## Please complete the following:
- [x] I have added screenshots for all UI updates
- [x] I confirm I have thoroughly tested these changes and take full
responsibility for any bugs introduced
- [x] I understand that submitting code with bugs that could have been
caught through manual testing blocks releases and new features for all
contributors
## Please put your Discord username so you can be contacted if a bug or
regression is found:
leo21_
---------
Co-authored-by: Scott Anderson <scottanderson@users.noreply.github.com>
Co-authored-by: evanpelle <evanpelle@gmail.com>
## Description:
Taken from PR #506
Improve transport source tile by considering border extremums
Only calculate better spawn tile for humans, and have the sender
calculate it and send the src tile in the intent for better performance.
## Please complete the following:
- [x] I have added screenshots for all UI updates
- [x] I confirm I have thoroughly tested these changes and take full
responsibility for any bugs introduced
- [x] I understand that submitting code with bugs that could have been
caught through manual testing blocks releases and new features for all
contributors
## Please put your Discord username so you can be contacted if a bug or
regression is found:
<DISCORD USERNAME>
evan
Co-authored-by: evan <openfrontio@gmail.com>
## Description:
The server stores all players that have joined, and once the game starts
it sends a list of players to all clients. Players cannot join after the
game has started. Server now generated the PlayerID instead of the
client.
The is necessary for team mode, we need to know how who is playing the
game before it starts so we can properly assign teams based on clans.
## Please complete the following:
- [x] I have added screenshots for all UI updates
- [x] I confirm I have thoroughly tested these changes and take full
responsibility for any bugs introduced
- [x] I understand that submitting code with bugs that could have been
caught through manual testing blocks releases and new features for all
contributors
## Please put your Discord username so you can be contacted if a bug or
regression is found:
evan
Tried to implement feature to outline player under mouse cursor.
Intention is to improve gameplay and provide a way to estimate players
territories for strategic planning / avoid attacking wrong players with
confusingly similar colors.
When you stop cursor at the player - his borders will be drawn in white
color.
To focus on other player - click or move cursor to other player.
To hide outline - click on empty space (water, land).
"Focus" UI feature also triggers outline for the target player (easier
to see who exactly is that you are looking at).
Done via getting player under mouse cursor, setting it as global
focusedPlayer and painting borders in white color.
Tried to maintain minimal changes and utilize existing rendering queue.
Also added hover delays to avoid excessive redraws and provide better
experience. Redraws only happens when focusedPlayer changes - one time
for old focused player to clean outline and one time for new focused
player.
ralfisalhon