Fast math is not required, but it speeds up some calculations at the expense of accuracy. There are some functions like dtMathIsfinite that use floating point functions that become undefined behavior when compiled with fast-math, so we need to conditionally short-circuit these functions when compiled with that flag.
-Wnan-infinity-disabled is complaining about the isfinite call in dtMathIsfinite.
This also sets the linux runner explicitly to Ubuntu 24.04, since ubuntu-latest defaults to 22.04 for some reason. This also updates gcc and clang to the latest versions in apt and logs their version to the run output. We need at least clang18 to disable the -Wnan-infinity-disabled warning for Catch.
Finally, this also removes some unused code that was throwing a warning (and thus an error) on newer compiler versions.
* Changed asserts to be explicitly disabled with RC_DISABLE_ASSERTS rather than using NDEBUG
This gives explicit control to users over when recast asserts are enabled or disabled. This is useful when users wish to disable asserts in debug mode or enable them in release mode. e.g. it's common to compile game code in debug alongside third party code in release.
* Added Integration.md with documentation on the integration process for Recast
- Upgraded doxyfile to latest format
- Fixed a bunch of warnings and errors in docstrings that doxygen was complaining about
- Added the doxygen-awesome theme to modernize the output
- Removed a duplicate screenshot we'd had in the docs folder
- Moved the RecastDemo screenshot into the Docs/images folder
- Changed the doxygen main page to be the README.md since the dedicated main page file we had was nearly identical but out of date
- Added CONTRIBUTING.md so it's part of the generated doxygen output
- Removed duplicate license file to avoid confusion
- Combined the FAQ with the Recast_api.txt, since the FAQ was essentially just better docs for the members of `rcConfig`
* Fixes for many low-severity compiler warnings
Mostly a lot of pedantic things like end-of-file newlines and commas on the last enum value. However some fixes like the weak v-tables warnings fixes might help with code gen. It's unclear if the linker is able to elide multiple copies of the same type's v-table in different translation units, (sometimes it can) but these fixes ensure we don't have to rely on it.
* Validate constraint: #polygons per tile
When adding a tile to a NavMesh, ensure that the number of polygons in that tile fit in the poly ID address space.
* Improve dtNavMeshParams field documentation
* Make dtClosestHeightPointTriangle use no epsilon (except for checking
denominator)
* When point is outside poly, directly find closest height from boundary
detail edges instead of first interpolating and then finding height from
detail tris.
* When point is inside poly, if all dtClosestHeightPointTriangle queries
failed, then find height from detail edges. This should only happen if
the point is right on top of an internal detail tri edge.
This adds a method to dtNavMeshQuery that allows users to retrieve paths
from the nodes explored by the previous Dijkstra search. This is useful
after Dijkstra searches to retrieve arbitrary paths inside the explored
area.
dtNavMeshQuery now fails initialization if too many nodes are requested
in the node pool. This could cause wrong paths and infinite loops to
happen if the node indices started overflowing dtNodeIndex or
dtNode::pidx.
Fix#178
This adds a version of queryPolygon that takes an interface which is
passed batches of polygons in the search area. This new API is then used
from the old queryPolygons to collect polygons, and from findNearestRef
to find the nearest poly in an arbitrarily sized area. Previously
findNearestPoly had an arbitrary poly limit of 128 which could cause it
to return potentially wrong results.
Fix#107
* Added virtual qualifier to a few destructors in subclasses.
* Removed a few empty destructors that didn't need to be explicitly defined.
* Fixed a few typos
* Removed unnecessary empty virtual destructors in some classes with parent's who derrived from a class with a virtual destructor
External links representing off mesh connections between layers in the
same tile location would not be correctly removed. These links had
side = 0xff which were skipped during tile removal under the assumption
that these links were internal.
Also correct some comments in the same place.
Fix#113
This makes it simpler to investigate exactly what is blocking a ray - for example, if an agent cannot move onto a polygon due to QueryFilter flags, then this provides a way for the agent to find out which flags it needs to have and to perform a state change which alters its movement flags.
"Options for dtNavMeshQuery::FindPath" is not correct, the option is used in SlicedFindPath only
DT_FINDPATH_LOW_QUALITY_FAR is not used anywhere
I would also add, that even though raycast considers costs, does so not very well :-(
