Add special loading progress bar.
It should be fast enough to not keep loading screen for noticably long but
will provide better pathfinding for actors inside interior cells.
This AABB tree required when need to filter out input mesh that has not
influence navmesh tile output. This filtering is already done before. Each
recast mesh corresponds to a single navmesh tile and has appropriate bounds.
It fills areas.data with walkable triangles but right after it's overwritten
and rcClearUnwalkableTriangles is called instead that fills areas.data with
unwalkable triangles.
To avoid triggering NavMesh update when RecastMesh change should not change
NavMesh.
Based on the following assumption:
Given a set of transformations and a bounding shape for all these
tranformations, a new object transformation that does not change this
bounding shape also should not change navmesh if for all of this object
transformations resulting navmesh tiles are equivalent
The idea is to report back to RecastMeshManager all changes of NavMesh if there
are any assiciated with RecastMesh version. So we know the last time when
RecastMesh change resulted into the NavMesh change. When later report shows
that there was no NavMesh change for a new RecastMesh version we can assume
that any object transformation within the same bounding box should not change
NavMesh.
mLastBuildRevision == mRevision with current use cases does not happen. But
even if this will happen when another use case will be added it does not save
much computation. The most expensive operation is not adding objects to the
MeshBuilder but to optimize resulting navmesh that will happen anyway in
MeshBuilder::create call.
When distance between start and end point is greater than max radius of area
possibly covered by navmesh there is no way to find path via navmesh. Also if
distance is greater than cell size navmesh might not exists withing mentioned
area because cell is not loaded therefore navmesh is not generated. So minumum
of these values is used to limit max path distance. Assuming that path
actually exists it's possible to build path to the edge of a circle. When
actor reaches initial edge path is built further. However it will not be
optimal.
Serialization into a vector of chars produces inconsistent results that leads
to reduced cache hit rate. Using a structured object is a more clear solution
and allows to remove serialization and nontrivial key compare logic with more
straigt forward structured object comparison.
To fix msvc error:
components\detournavigator\objectid.hpp(14,13): error C2440: 'reinterpret_cast': cannot convert from 'const T' to 'size_t'
with
[
T=unsigned long
]
When there is only one actor (player) on a scene and it moving to other
cell first it will be removed from navigator then added. Remove cause
navmesh removing for its half extents. After it is added navmesh for
same half extents is created and added. While this all happens there are
still jobs for old navmesh are processing. Old navmesh still exists
because it is stored by shared pointer. So jobs take tiles from cache
and place them into old navmesh. After that other jobs take same tiles
from cache (half extents and coordinates are equal) and place them into
other navmesh. dtNavMesh changes tile data on add and remove. Adding tile
to two dtNavMesh corrupts tile in both nameshes.
When update method is called for not changed object befor this change
all object tiles were considered as not object tiles and were removed.
Also this marked those tiles as changed. This lead to alternation
between remove and add each tile update method was called. Problem was
detected by using Animated Containers mod.
Every updated object should produce a set of changed tiles where it is
placed. Before this change only current object tiles were updated. If
object was moved to another set of tiles then navmesh were not changed
in new tiles.
TileCachedRecastMeshManager::updateObject should add all new tiles if object
was moved and remove all no more used tiles. Both new and old tiles should be
marked as changed.
Also add tests to show desired result for add, update, remove.
* use c++11 std::align from <memory>
* for Ubuntu, use gcc5 instead of 4.8
* use travis to set gcc to 5
eval
and sudo
* use eval in .travis.yml
* use gcc-8
* replace precise with trusty llvm toolchain, because we have been using trusty for awhile now
* push things to matrix, so we can support multiple releases if we want
* we should not be allowing for failures, we are ready to start trusting clang and its analyzer
* scan-build was pushed to another package
* use gcc-8 still but wrap in scan-build
* travis.yml cleanup, have output of scripts go to stdout, make search for substring a regex
use double []
fix missing ,
use bash to use regex
black spaces matter
* set human readable names for our various builds, split out our static analysis between openmw and openmw-cs
* test if not set, then set otherwise ignore
* use quotes
* do not eval it, set it in travis env
* no more &&
* what does clang7 have to say?
* use sourceline for now
* use clang-7 instead of clang-7.0
* yes, llvm-toolchain-trusty-7 not llvm-toolchain-trusty-7.0
* for static analysis, openmw is compiled and checked on its own while openmw-cs is build with all the rest. this might change in the future.
and actually do it the other way around
Use LRU modification to hold currently used items. Use RecastMesh binary
data for item key.
Store original pointer of btCollisionShape in user pointer to make available
it as an identifier within all duplicates. Use pointer to heights data array
for btHeightfieldTerrainShape.
When player move fast enough, tiles update for specific area square
couldn't catch player move. Tiles to be removed are left in the queue
with lower priority then tiles to be added which are nearest to player.
This can lead to overflow for amount of tiles. So we try to do remove
first. But we detect change type approximately using mixed change type,
because even if we do it precise, change type could change while job
is in queue.