* Build/Misc: Add a few *San CMake flags
Add the following flags for the related tools:
- MSAN for Memory Sanitizer
- UBSAN for Undefined Behavior Sanitizer
- TSAN for Thread Sanitizer
* Remove unused parameter
* Fix UBSan reported issue
* Disable G3D buffer pools when using Thread Sanitizer as it has its custom locking mechanisms
* Code cleanup
* Move threads from maps to tiles
* Move tile building logic to TileBuilder class
* Fix memory leak in TileBuilder
* Fix build
* Store TileBuilder as raw pointer for now, it will be changed later on to use modern C++ constructs
* Fix crash on shutdown
* Revert pvs-studio change
* Fix generating 1 single tile not closing the program
(cherry picked from commit a4e93d779c)
Re-generating MMAPs IS required.
Partially reverts 995a443da2 .
Having 2 different slope angles of 55° and 85° created too many polygons to fit in the current mmtile structure. This caused some polygons to become disconnected from each other, creating the old "invisible walls" effect. Because of this and because of the performance hit when loading a mmtile caused by the increase of polygon numbers, this commit reverts the recent changes and sets by default the slope angle to 55°.
Feel free to restore the previous behaviour by running .\mmaps_generator --maxAngle 85 --maxAngleNotSteep 55 , specifying the map id as number if a single map should have different slope values.
This is the last commit that will change MMAPs version to force re-generating them. Any future change that will affect only the generation settings will be optional (but recommended).
(cherry picked from commit 3947e4cb57)
* Some research about mmaps load performances
* Tools/MMapsGenerator: Add some more input parameters
Add some more input parameters which affect the resulting mmaps:
- --smallOutputSize: sets both walkableSlopeAngle and walkableSlopeAngleNotSteep to 55°, reducing by a lot the .mmtile size and the number of polys. A rcFilterLedgeSpans() call is included too to filter out some polys. Default to false.
- --maxAngle: Restore this parameter removed some commits ago. Specifies the max walkable angle of a Creature when in combat. Default to 85
- --maxAngleNotSteep: Max walkable angle of a Player and of a Creature out of combat. Must be less or equal than --maxAngle. Default to 55.
A performance profiling comparison highlighted that .mmtile load grows exponential with the number of polys. A small server with GridUnload enabled and without preloading all grids on startup will be more affected than a big server with GridUnload disabled and preloading grids on startup.
(cherry picked from commit b6e205f7ef)
* Tools/MMAPs: Remove input argument --maxAngle
Remove input argument --maxAngle . This should be just hardcoded in MapBuilder::GetMapSpecificConfig() so all settings are easily found in the same place instead of being spread around.
* Tools/MMAPs: Add new area type NAV_AREA_GROUND_STEEP
Add new area type NAV_AREA_GROUND_STEEP for ground with slope in the range (55, 70] .
NAV_AREA_GROUND is used for ground with range [0, 55] .
NAV_AREA_GROUND_STEEP takes priority over NAV_AREA_GROUND.
* Tools/MMAPs: Fix NAV_GROUND_STEEP flag not being saved in the mmtile
* Core/PathFinding: Implement NAV_GROUND_STEEP flag
Implement NAV_GROUND_STEEP flag, used only by Creatures that are in combat or evading.
* Distinguish between RC_WALKABLE_AREA and NAV_AREA_GROUND.
* Allow mobs in combat to walk on steeps up to 80°
* Allow mobs in combat to walk on steeps up to 85°.
Disable rcFilterLedgeSpans() filter as it removed long steep spans.
* Increase cost of steep spans to try making creatures walk around obstacles instead of walking on them
* Revert last commit
(cherry picked from commit 995a443da2)
* Additionally he PathGenerator will now check for realtime movement flags instead of creature based template data only so scripted flying movement and players will now correctly get their path calculated
(cherry picked from commit fdcb7388a6)
* Core/Auras: Attempt to fix infinite loop with aura 18950 on map unload
* Core/Auras: Log a detailed error and assert when failing to remove all auras after a few tries
* Code cleanup
* Fix build warnings
* Fix more build warnings
(cherry picked from commit bd2d60c267)
* Core/MMAPs: Adjust walkable climb
Adjust walkable climb as recast using walkableClimb also to find the poly from a position, giving priority to polys that are below the position but closer than walkableClimb.
* Temporarily disable static collision (the whole check should be removed)
* Core/Spells: removed deprecated mmap path check for TARGET_DEST_CASTER_FRONT_LEAP
* Core/Objects: allow flying units to use the helper as well (flying units casting radius based spells)
* Code refactor
* Handle raycasts that end in a point with no height in the mmap mesh as PATHFIND_NOPATH
* Walk back a bit from raycast hitpoints as sometime the 2D result point is outside of the polygons due to floating point errors.
* Remove whitespace
* Revert 4a197ba22a as a raycast point path should have the Z retrieved with getPolyHeight(). Raycast will only return a 2-point path with Start and Hitpoint/End
* Cleanup PathGenerator raycast case
* Fix PathGenerator raycast broken if start and end are on same poly.
Fix PathGenerator raycast broken if no wall is hit.
Remove unused case of using raycast with an existing previous path (can be added back properly if needed).
Remove forcing poly length to 2 when we actually have already the right number.
* Use closestPointOnPolyBoundary on the second try of finding a point on poly for raycast.
Note that in this case the mesh height is not used which might cause issues. The poly boundary height will be used instead.
* Handle cases where getPolyHeight() fails because the point is on polygon border (and caused by floating point imprecision)
* Add far from poly flags
* Set PATHFIND_INCOMPLETE in raycast case if startFarFromPoly or endFarFromPoly
* Fix blink close to walls with no valid polygon behind the wall
* Require to re-extract mmaps
Co-authored-by: Ovah <dreadkiller@gmx.de>
(cherry picked from commit c0b75bf40d)
actually port back the previous implementation, now correctly
this implies that a unit that has a different victim than the one its chasing will stand still till either:
- a new movement replaces the current, for w/e reason
- the chase target is again the current victim
probably a pause implementation on the generator is a more elegant solution...
updates #24600
(cherry picked from commit 472e1fd8ae)
