Essentially reverts D1918 / d592bf9cb6.
Instead of copying path requests to workers, setup the result vector,
then setup an index, and compute 'in-place'.
To send messages, the result vectors are read in order. This makes the
order trivially constant no matter how many workers there are, and the
architecture overall makes it much easier to efficiently paralellise.
Tested by: Langbart, Stan
Differential Revision: https://code.wildfiregames.com/D3849
This was SVN commit r25256.
Components have some initialisation-order dependency. They could also
have destroy-order dependency, and from a RAII-like point of view, it
makes sense to do this.
Differential Revision: https://code.wildfiregames.com/D3843
This was SVN commit r25251.
This changes ParamNode to use UTF8 values internally (XMB files are UTF8
since cb9d0733ef).
This removes the need for a lot of conversions, speeding things up and
allows cleaning up the validator interface & a few other callsites.
ConstructJSVal could be a tad slower because of UTF8->16 conversions
within Spidermonkey; but the difference is unlikely to be noticeable in
practica.
Also:
- Changes `ToXML` to `ToXMLString` for clarity.
- Add a simple "op" test & show a particular behaviour of merge nodes
that I intend to change somewhat in D3830.
- Remove Component.h from simulation2 PCH - brought in too much.
Tested by: langbart
Differential Revision: https://code.wildfiregames.com/D3834
This was SVN commit r25228.
An actor file, as referenced by the VisualActor, can now define
different actors for different "quality level" setting.
In this initial version, the quality is handled directly by the object
manager.
Actor format impact:
- '<qualitylevels>' may be used as the root node, containing actor nodes
as children.
- such actor nodes can refer to a file, or to an inline actor, or
simply be inlined.
- such actor nodes may have a 'quality' attribute, specifying the
maximum quality level of this actor. By default, 255 (the maximum) is
implied.
- The actor format remains valid, but 'groups', 'variants', 'material',
'castshadow' and 'float' can be given a [minquality, maxquality[ range
via XML attributes. Outside of this range, the XML node is ignored
(making it possible to define, in a single actor file, several quality
levels).
Quality is a 0-255 value, with:
- Range 0-99 intended for lower level-of-detail actors (billboards,
etc.)
- Range 100-200 the 'normal' range for models. 100 is "low", 150
"medium", and 200 "high".
- Range 201-255 used for higher quality actors that might be used for
e.g. cinematics.
The range is wide to make it easier to add intermediate levels in the
future and it seemed easier given that an integer value of some kind was
required anyways.
Engine impacts:
- A new CActorDef class is introduced, wrapping an art/actors XML file
and its different quality levels. ObjectBase remains the definition of a
given 'actor', now at a given quality level.
- CActorDef imposes a maximal # of quality level for a particular actor
definition (5 currently).
- CUnit is made to refer to an Actor Definition explicitly, not a
particular ObjectBase.
- As a minor optimisation, variation keys are calculated on
pointer-to-sets-of-selections, instead of raw sets-of-selections, as
this reduces copying.
- some refactoring, including better const-correctness and hotloading
support via std::shared_ptr.
Differential Revision: https://code.wildfiregames.com/D3787
This was SVN commit r25210.
This basically does two things:
- Allow turrets on moving entities (fixes serialisation in cmpPos).
- Allow an entity to be present on another when the other is created.
That makes it basically a boiled down version of a prior patch, omitting
the part where orders could be passed on.
This does allow e.g. ranged units on chariots that fire while the
chariot moves.
Original diff by: @sanderd17
Redone by: @Stan
Differential revision: D1958
Comments by: @Alexandermb, @Angen, @bb, @Langbart, @Nescio, @Stan,
@wraitii
Refs. #2577 by implementing the entity on platform case.
This was SVN commit r25192.
The motion manager introduced in bae258f9a1 makes 'chasing' (e.g. an
entity targeting another entity, which also applies to fleeing)
behaviour symmetrical: both units see the initial state instead of it
being ID-dependent.
This allows removing hacks introduced in 6a66fb8205 (and refined in
7b88b1a0f9).
Differential Revision: https://code.wildfiregames.com/D3785
This was SVN commit r25185.
This implements a form of crowd movement that I've generally called
'unit pushing' in the last few years.
Essentially, any two units will push each other away when they're too
close. This makes it possible to ignore unit-unit obstructions, and thus
makes movement much smoother in crowds.
This first iteration of this system only allows pushing between idle
units and between moving units (i.e. a moving unit does not affect an
idle one).
This is because the unitMotion logic to detect it is stuck & needs to
use the pathfinders starts breaking: units can fail to move because they
are pushed away from their intended movement, and the current logic
fails to handle this gracefully.
Thankfully, the most value of this patch in terms of player experience
is found in the improvements to group movements and shuttling.
Other impacts:
- As the short pathfinder is called less often, we can increase the
starting search range & reduce the # of max turns, both improving
collision recovery.
- The performance of idle units is slightly worsened, as they must be
checked for idle-idle collisions. If needed a 'sleeping' system, as used
in physics engine, could be implemented.
- In general, however, expect slight performance improvements, as fewer
short paths are computed.
- Gathering efficiency should increase slightly, since shuttling times
are likely reduced slightly.
- As a sanity change to improve some edge cases (units that say they're
moving, i.e. pushable, but don't actually move), the 'going straight'
logic is turned off if a short path has been computed. This requires a
few cascading changes to work correctly.
Technical notes:
- To reduce the cost of the n^2 comparisons that pushing requires, units
are only compared within a small square on a grid which is lazily
reconstructed each turn. The overhead seems rather small, and this is
much simpler than keeping an up-to-date grid.
- The design is intended to be parallelisable if needed someday.
- The pathfinder's CheckMovement ignores moving units in UnitMotion, as
that is now the spec. Idle units are not ignored, which is required for
the 'collision' detection to work correctly (see above).
Refs #3442 (not fixed - idle units are not pushed by moving units).
Fixes#5084 (the overlap can still happen, but units will push each
other away).
Differential Revision: https://code.wildfiregames.com/D1490
This was SVN commit r25182.
Change the OOS notification logic to remember the OOS-ness of each
client. Reset it on client leave.
The server will thus continue checking for OOS if the OOS-client leaves.
This is convenient to ignore observer OOS, or wait for an OOS player
without restarting the game.
Also add the turn number to the OOS dump, to fix#3348: particularly
following d4c2cf4430 the turn is likely to not be the same between
different clients.
Agree by: asterix
Differential Revision: https://code.wildfiregames.com/D3753
This was SVN commit r25170.
d8ea401a95 replaced the ExploreTerritories logic, but I did not account
for the territory grid being N cells wide when the LOS is N+1.
Therefore, we could get out of bounds.
Reported by: Vladislavbelov
Differential Revision: https://code.wildfiregames.com/D3765
This was SVN commit r25161.
Observers no longer lag the game for players. There is still some time
to serialise the game when sending it to a joining observer, and
depending on the chosen 'max lag' the game may stop while observers
sufficiently catch up, but this impact too is reduced.
- Make the NetServerTurnManager ignore players marked as 'observers' for
the purpose of ending a turn, effectively making it possible for
observers to lag without it affecting the players in any way.
- Add a config option (network.observermaxlag) that specifies how many
turns behind the live game observers are allowed to be. Default to 10
turns, or 2 seconds, to keep them 'largely live'.
- The controller is not treated as an observer.
- Implement a simple UI to show this delay & allow the game to speed up
automatically to try and catch up. This can be deactivated via
network.autocatchup.
- Move network options to the renamed 'Network / Lobby' options page.
- Do not debug_warn/crash when receiving commands from the past -
instead warn and carry on, to avoid DOS and "coop play" issues.
Refs #5903, Refs #4210
Differential Revision: https://code.wildfiregames.com/D3737
This was SVN commit r25156.
NaN values could not be serialised safely because of the multiple
possible NaN numbers.
Since NaN values are usually the result of bugs or dangerous code, it
seems simpler to refuse to serialise them.
(D3205 was a safe-serialization alternative, should the need arise).
Fixes#1879
Differential Revision: https://code.wildfiregames.com/D3729
This was SVN commit r25151.
9d82ae15af introduced logic to extend the search range of the short
pathfinder in some situations. This extension was unbounded, resulting
occasionally in search domains several hundred meters wide, and path
computation that can take several seconds.
Reported by: Vico (on the forums).
Differential Revision: https://code.wildfiregames.com/D3760
This was SVN commit r25150.
Following bae258f9a1, all units moved, then all messages were sent. This
could result in units overlapping, which cannot be resolved without unit
pushing. Fixing this makes for a fairer comparison with the pushing
diff, so it seems worthwhile.
Differential Revision: https://code.wildfiregames.com/D3746
This was SVN commit r25126.
bae258f9a1 implemented CCmpUnitMotionManager, which handles motion for
CCmpUnitMotion (as distinct from ICmpUnitMotion, the interface).
The tight coupling between these two components was awkward at the
interface level, leaking underlying implementation details.
This diff makes CmpUnitMotionManager explicitly manage CmpUnitMotion,
instead of any implementation of ICmpUnitMotion, and moves files around
as requir
ed.
This:
- Makes it impossible to accidentally try to have the wrong
IID_UnitMotion managed by the UnitMotionManager.
- Allows devirtualising the calls from the manager to UnitMotion itself
(and inlining, as they are compiled in the same TU).
- Cleans up the manager interface - MotionState is now part of
CCmpUnitMotionManager.
- Cleans up ICmpUnitMotion interface - no need to provide a private
interface for the manager.
Differential Revision: https://code.wildfiregames.com/D3732
This was SVN commit r25125.
This is the same commit as 4bb31f084e, reverted in 5d1899785a. Noted
issues were fixed.
Partial revert/fix of 2f19cf86d3 and 2567fee329.
Before this diff, it was possible for a mod containing a cache/ or a
config/ folder to get written to incorrectly.
The issue is VFS can map multiple directories to one 'virtual' path, for
reading mods. However, writing data is problematic: which path to
choose?
The only viable solution is to use a path relative to the highest
priority directory encountered in the VFS path, or write paths could be
'hijacked' by lower-priority mods.
This fixes these issues by:
- Adding a new lookup mode ('Real-path') that explicitly picks the real
path relative to the highest-priority subdirectory in the VFS Path.
- Preventing overwriting a real directory with a lower priority one in
general.
- Revert c0c8132dd4's GetRealPath change, re-introducing the function as
GetOriginalPath.
This also cleans up some duplication that led to empty mod folders in
the user mod path, and cleans up loading the 'user' mod.
It also makes it explicit that a directory must be passed to Mount().
Note that the new 'realpath' lookup can still be somewhat complex with
many mount points at various hierarchy levels, but it is at least
predictable/deterministic without having to be careful about populating
order.
Fixes#2553
Differential Revision: https://code.wildfiregames.com/D3728
This was SVN commit r25107.
Partial revert/fix of 2f19cf86d3 and 2567fee329.
Before this diff, it was possible for a mod containing a cache/ or a
config/ folder to get written to incorrectly.
The issue is VFS can map multiple directories to one 'virtual' path, for
reading mods. However, writing data is problematic: which path to
choose?
The only viable solution is to use a path relative to the highest
priority directory encountered in the VFS path, or write paths could be
'hijacked' by lower-priority mods.
This fixes these issues by:
- Adding a new lookup mode ('Real-path') that explicitly picks the real
path relative to the highest-priority subdirectory in the VFS Path.
- Preventing overwriting a real directory with a lower priority one in
general.
- Revert c0c8132dd4's GetRealPath change, re-introducing the function as
GetOriginalPath.
This also cleans up some duplication that led to empty mod folders in
the user mod path, and cleans up loading the 'user' mod.
Note that the new 'realpath' lookup can still be somewhat complex with
many mount points at various hierarchy levels, but it is at least
predictable/deterministic without having to be careful about populating
order.
Fixes#2553
Differential Revision: https://code.wildfiregames.com/D3217
This was SVN commit r25104.
Fixes bae258f9a1.
As reported by Freagarach, garrisoned units trigger error messages.
As reported by Vladislavbelov, noPCH was broken & there were some style
issues.
Differential Revision: https://code.wildfiregames.com/D3707
This was SVN commit r25078.
This new MotionManager handles movement for UnitMotion components (not
UnitMotionFlying).
This is a first step towards unit pushing, by giving a central place for
the relevant units to collide.
One important side-effect is that movement is effectively synchronous -
the positions are not actually updated until all units have moved for a
turn (refs 6a66fb8205).
As a side-effect, it's an optimisation: fewer messages are being sent
overall, which leads to a slight speedup (negligible without a lot of
units though).
This is a first step - ideally, the movement functions called from
UnitMotionManager would actually be moved there.
Differential Revision: https://code.wildfiregames.com/D3509
This was SVN commit r25071.
Fixes a1dc9cadd8: if the speed doesn't change, UnitMotion doesn't update
the visual actor. Unfortunately, if another component has in the
meantime reset the animation to 'Idle', the unit will now move while
Idle. This can happen when leaving formation to do something else,
though it'srare.
This fixes that by instead always calling VisualActor, which does its
own checking to avoid redundancy. It's a bit less efficient, but not too
much.
Note that this relies on UnitMotion::UpdateMovementState being called
after any UnitAI code that could reset the animation to IDLE.
Differential Revision: https://code.wildfiregames.com/D3619
This was SVN commit r25011.
To hide network latency, MP turns send commands not for the next turn
but N turns after that (introduced in c684c211a2).
Further, MP turn length was increased to 500ms compared to 200ms SP
turns (introduced in 6a15b78c98).
Unfortunately, increasing MP turn length has negative consequences:
- makes pathfinding/unit motion much worse & unit behaviour worse in
general.
- makes the game more 'lag-spikey', since computations are done less
often, but thus then can take more time.
This diff essentially reverts 6a15b78c98, instead increasing
COMMAND_DELAY from 2 to 4 in MP. This:
- Reduces the 'inherent command lag' in MP from 1000ms to 800ms
- Increases the lag range at which MP will run smoothtly from 500ms to
600ms
- makes SP and MP turns behave identically again, removing the
hindrances described above.
As a side effect, single-player was not actually using COMMAND_DELAY,
this is now done (can be used to simulate MP command lag).
Refs #3752
Differential Revision: https://code.wildfiregames.com/D3275
This was SVN commit r25001.
- Replace ScriptInterface::RegisterFunction with
ScriptFunction::Register
- Mostly removing unused cmpPrivate*
- Some usage introduces specific getters (mapgenerator, AIWorker,
XmppClient,...)
- Several passthrough functions are simply removed in favour of calling
the original, reducing duplication
- Make use of ScriptRequest/ScriptInterface capabilities where
relevant.
- Make JSI_* headers only expose necessary functions, lightening them
considerably and reducing duplication
- Reuse namespaces in JSI_* implementations directly, reducing visual
noise there
Follows f3aedf88a6
Differential Revision: https://code.wildfiregames.com/D3626
This was SVN commit r24983.
Introduce a LOS_TILE_SIZE, to replace usage of TERRITORY_TILE_SIZE in
the LOS code.
This makes it possible to change the resolution of LOS/Terrain without
affecting the other component.
Additional refactoring:
- LosTile has been renamed LosRegion (it's more comparable to the
hierarchical pathfinder regions/spatial subdivisions)
- LosState explicitly refers to "los vertices" instead of terrain
vertices.
Refs #5566
Differential Revision: https://code.wildfiregames.com/D3076
This was SVN commit r24980.
The new methods:
- aren't included in ScriptInterface.h directly, lightening that header
- don't use boost CPP
- don't need argument types or number or constness to be specified
- can work with object methods somewhat transparently
- support optional cmptPrivate (allowing removal of many UNUSED macro)
- support optional const ScriptRequest&, which is safer.
This first diff changes only some of the JSI files & the component
manager. Further diffs will update other files and finally delete the
current code.
Differential Revision: https://code.wildfiregames.com/D2818
This was SVN commit r24969.
The spawn code should not assume that obstructions will never be
0-sized.
Reported by: Zack
Fixes#6039
Differential Revision: https://code.wildfiregames.com/D3583
This was SVN commit r24945.
Obstructions can reactivate after being destroyed, which leaves them
dangling forever. This is similar to 54f7a09a54.
This fixes this issue by setting a temporary "is destroyed" variable to
true.@
Reported by: seeh
Tested by: langbart
Reviewed By: Freagarach
Fixes#6016
Differential Revision: https://code.wildfiregames.com/D3558
This was SVN commit r24894.
5d96346ac5 proved unsufficient to fix formation 'waltzing'. This is a
better fix, which makes sure units actually try to reach their
designated offset in the first place.
Further, it removes code that recalculated offsets un-necessarily, which
led to an issue with "sloppy" formations such as open and closed orders.
Fixes#5997
Differential Revision: https://code.wildfiregames.com/D3543
This was SVN commit r24865.
Following 847f3a9995,
Units in formation can get very small movement offsets, that nonetheless
require large rotations, and thus at least 2 turns to accomplish.
However, following 847f3a9995, PossiblyAtDestination fires() only after
the first rotation, so the unit ends up 'waltzing' in place.
Before that diff, the unit never even moved since PossiblyAtDestination
fired straight away.
This is also noticeable since IDLE formation re-order their members
since 71a61d5f50.
The fix here is to ignore rotation time for very small offsets, which
lets units accomplish the movement in one turn and fixes the issue.
Reported by: wowgetoffyourcellphone
Reviewed By: Freagarach
Tested By: langbart
Differential Revision: https://code.wildfiregames.com/D3518
This was SVN commit r24831.
I'm hoping this will give us more information on what's going wrong.
Also remove comments which were outdated since 7460d0e56e
Refs #5987
Differential Revision: https://code.wildfiregames.com/D3501
This was SVN commit r24811.
D3230 / 847f3a9995 introduced range checking at turn start, and removed
a hack that made units predict the position of their target too far
ahead. This worked fine when in "straight movement" mode, unfortunately
I failed to recognise that ranged units would never use that mode. This
meant that ranged-unit chasing was broken.
There is a straightforward fix however, since we can simply change
TryGoingStraightToTarget to be used by ranged units. It fixes the issue
efficiently and improves movement for ranged units in general, so it
probably should have been done from the start.
Refs #5936
Differential Revision: https://code.wildfiregames.com/D3489
This was SVN commit r24803.
- Because units slow down when turning, and JPS paths often begin with a
J-shape, chasers can fail to catch up to slower chasee, because the
latter don't recompute paths as often. To fix this, ignore the first
waypoint if it's close by and the next is accessible.
- Don't interpolate the target position when interpolation isn't
necessary (i.e. when not processing the MT_Update_Motion* message), as
that resulted in the "follow known bad path" hack to active
un-necessarily.
- Tweak PathingUpdateNeeded, it will return true when it has no path to
follow
- Remove the direct-range consideration in the "distance uncertainty"
calculation.
Refs #5936
Differential Revision: https://code.wildfiregames.com/D3485
This was SVN commit r24800.
Units movement is currently "all or nothing". This means that a chasing
entity that moves fast enough is likely to collide with its target, if
the latter is moving also. This means that it might fail to get in range
if the max range is smaller than the movement speed over a turn.
This happens to be very much the case in MP, as cavalry range is 4,
melee cav speed is ~20 and turns are 500ms.
This problem depends on which unit moves first (i.e. which unit is
lowest-ID).
To fix this, ignore the obstruction of the target, if it is moving, when
moving. This however means sometimes chasers will 'overshoot' and block
their target pathing, making the chase easier than it probably should
be.
Fleeing units don't suffer from this problem since they also ignore
their target (and their code handles it).
This new problem introduced in this diff is heavily dependent on the
exact speeds and ranges at play, and a further diff will improve the
situation to acceptable levels.
Reported by: FeldFeld
Refs #5936
Differential Revision: https://code.wildfiregames.com/D3482
This was SVN commit r24798.
Unit Motion currently checks if the unit is at destination during the
MT_Update_Motion* step, which happens late in the turn (notably, after
Timer.js) and moreover happens while entities are being moved (e.g.
entities with lower IDs have moved already, entities with higher IDs
have yet to do so).
This changes UnitMotion to instead check at turn start, which:
- benefits from in-turn path computations for more fluid movement
- ensure that distance checks aren't done against an entity that has
already moved for the turn.
The latter issue led to units failing to get in range of their target
when chasing them, in some situations.
As a side effect, this means that UnitAI move requests always take one
turn to succeed, so orders should be updated to check for range (or
they'll waste a turn). This is done for garrisoning, other orders were
already doing so.
Also includes a small tweak to avoid units rotating randomly when they
have no movement to accomplish.
Patch by: bb
Reviewed By: wraitii
Refs #5936
Differential Revision: https://code.wildfiregames.com/D3230
This was SVN commit r24797.
f2b3c4d8ec broke deserializing because it did not reset the mapsize on
Deserialize(), and thus the water renderer would not regenerate data.
Differential Revision: https://code.wildfiregames.com/D3461
This was SVN commit r24778.
