andrew/libakgl
1
0
Fork 0
Code Issues 67 Pull Requests Actions Packages Projects Releases Wiki Activity
Files
32dd48f28ffe0c1362cf11bdadf0546d525ef382
libakgl/src/collision.c

451 lines
15 KiB
C
Raw Normal View History

Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
/**
* @file collision.c
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
* @brief The narrowphase, and the only place libccd is visible.
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
*
* Nothing else in the tree includes `<ccd/ccd.h>`. The translation between
* libakgl's shapes and libccd's support-function protocol is `static` here, so
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
* changing narrowphase library is a change to one file and no header -- and so
* that `akgl.pc` never has to name a dependency that is compiled in.
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
*/
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
#include <math.h>
#include <string.h>
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
#include <akerror.h>
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
#include <akgl/collision.h>
Add a pluggable broad phase, and the incremental uniform grid behind it akgl_Partitioner is a record of function pointers and an initializer, the same shape as the render and physics backends. `move` is its own slot rather than remove-then-insert, and that is the whole design: a uniform grid's move is a comparison and a return when the proxy has not left the cells it was in, which is the steady state for a walking actor and the permanent state for a static one. Spelling it as remove-and-insert would turn the incremental grid into the rebuild-every-frame tree that PERFORMANCE.md already measured and rejected. The grid is a dense 128x128 array of cell heads over the world, cells keyed on tile size, with two intrusive chains per entry -- one through the cell so a query can walk it, one through the proxy so removal unlinks a whole span without searching. Everything is an int16_t index rather than a pointer, so the structure is relocatable and clearing it is a memset. A proxy covering more cells than AKGL_COLLISION_GRID_MAX_SPAN spills onto one chain every query walks, which is what makes the cell pool's ceiling provable rather than hopeful. tests/partition.c compares every query against a linear scan over the same proxies and asserts containment in one direction, not equality. That asymmetry is the contract: over-reporting costs a narrowphase call, under-reporting is a wall an actor walks through. The suite is a table over partitioners so the same assertions run against every implementation, which is what the second one landing later will be held to. Three deliberate breaks, and two of them were not caught the first time: - Removing the min-cell rule so a shared-cell pair reports repeatedly: caught. - Dropping the unlink in `move`: **not caught**, initially. Stale entries do not produce wrong query answers, because the bounds re-check filters them out -- they leak the cell pool until the grid stops working, on a timescale no unit test reaches by accident. Counting live pool entries across 200 moves is the only thing that sees it, and now does: 505 entries against an expected 5. - Swapping floorf for a truncating cast: **not caught, and correctly so.** Everything below zero is clamped to cell 0 either way, so today the two are indistinguishable. The comment claiming truncation is a defect was overstating it and now says what is actually true: the clamp is the only thing hiding it, and a negative world origin or a relaxed clamp would make it real with no test standing in front of it. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:29:02 -04:00
#include <akstdlib.h>
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
#include <akgl/collision_arena.h>
#include <akgl/error.h>
#include <akgl/types.h>
#include <ccd/ccd.h>
#include <ccd/quat.h>
#include <ccd/vec3.h>
/**
* @brief Ceiling on narrowphase iterations.
*
* libccd's own default, set by `CCD_INIT`, is `(unsigned long)-1` -- an
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
* unbounded loop inside a frame. A pair that reaches this cap is reported as not
* colliding, which is the same answer a missed collision gives, but it
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
* terminates. Bounding the iterations also bounds what the arena can be asked
* for, since the polytope grows with them.
*/
#define AKGL_COLLISION_MAX_ITERATIONS 100
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
/** @brief Below this, a normal is treated as having no length at all. */
#define AKGL_COLLISION_EPSILON 1e-6f
/** @brief One positioned shape, in the form libccd's callbacks read. */
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
typedef struct {
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
uint8_t kind; /**< AKGL_COLLISION_SHAPE_*. */
ccd_vec3_t pos; /**< World centre: owner position plus shape offset. */
ccd_vec3_t half; /**< Half-extents. `v[0]` doubles as the radius for a circle. */
} collision_ccdobj;
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
/**
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
* @brief Furthest point of a shape in a given direction.
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
*
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
* The support function *is* the shape as far as libccd is concerned. One
* function dispatching on kind rather than one per kind, because a pair may mix
* kinds and `ccd_t` has no way to choose per object beyond this.
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
*
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
* There is no rotation, here or anywhere in this library yet -- `util.h` says so
* and both example games depend on it. Adding it means rotating @p dir into the
* shape's local frame at the top of this function and rotating the answer back,
* and nothing else in the narrowphase changes. See `TODO.md`, "Actor rotation".
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
*/
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
static void collision_support(const void *obj, const ccd_vec3_t *dir, ccd_vec3_t *dest)
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
{
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
const collision_ccdobj *shape = (const collision_ccdobj *)obj;
ccd_real_t planar = 0.0;
ccd_real_t len = 0.0;
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
switch ( shape->kind ) {
case AKGL_COLLISION_SHAPE_CIRCLE:
/*
* A circle extruded along z is a cylinder, not a sphere. A sphere's caps
* curve away from the plane and would let a shape slide past a corner
* that a 2D game expects to catch.
*/
len = (ccdVec3X(dir) * ccdVec3X(dir)) + (ccdVec3Y(dir) * ccdVec3Y(dir));
if ( len > AKGL_COLLISION_EPSILON ) {
len = CCD_SQRT(len);
planar = ccdVec3X(&shape->half) / len;
ccdVec3Set(dest, ccdVec3X(dir) * planar, ccdVec3Y(dir) * planar, CCD_ZERO);
} else {
ccdVec3Set(dest, CCD_ZERO, CCD_ZERO, CCD_ZERO);
}
dest->v[2] = ccdSign(ccdVec3Z(dir)) * ccdVec3Z(&shape->half);
break;
case AKGL_COLLISION_SHAPE_CAPSULE_X:
// A segment on x with a circle of radius hy swept along it.
ccdVec3Set(dest,
ccdSign(ccdVec3X(dir)) * (ccdVec3X(&shape->half) - ccdVec3Y(&shape->half)),
CCD_ZERO,
CCD_ZERO);
len = (ccdVec3X(dir) * ccdVec3X(dir)) + (ccdVec3Y(dir) * ccdVec3Y(dir));
if ( len > AKGL_COLLISION_EPSILON ) {
len = CCD_SQRT(len);
planar = ccdVec3Y(&shape->half) / len;
dest->v[0] += ccdVec3X(dir) * planar;
dest->v[1] += ccdVec3Y(dir) * planar;
}
dest->v[2] = ccdSign(ccdVec3Z(dir)) * ccdVec3Z(&shape->half);
break;
case AKGL_COLLISION_SHAPE_CAPSULE_Y:
ccdVec3Set(dest,
CCD_ZERO,
ccdSign(ccdVec3Y(dir)) * (ccdVec3Y(&shape->half) - ccdVec3X(&shape->half)),
CCD_ZERO);
len = (ccdVec3X(dir) * ccdVec3X(dir)) + (ccdVec3Y(dir) * ccdVec3Y(dir));
if ( len > AKGL_COLLISION_EPSILON ) {
len = CCD_SQRT(len);
planar = ccdVec3X(&shape->half) / len;
dest->v[0] += ccdVec3X(dir) * planar;
dest->v[1] += ccdVec3Y(dir) * planar;
}
dest->v[2] = ccdSign(ccdVec3Z(dir)) * ccdVec3Z(&shape->half);
break;
default:
// A box, and the fallback for anything unrecognised: the corner in the
// direction's octant.
ccdVec3Set(dest,
ccdSign(ccdVec3X(dir)) * ccdVec3X(&shape->half),
ccdSign(ccdVec3Y(dir)) * ccdVec3Y(&shape->half),
ccdSign(ccdVec3Z(dir)) * ccdVec3Z(&shape->half));
break;
}
ccdVec3Add(dest, &shape->pos);
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
}
/**
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
* @brief Centre of a shape.
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
*
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
* MPR requires this where GJK does not: `ccdMPRPenetration` calls `center1` and
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
* `center2` unconditionally, so leaving either `NULL` is a null dereference on
* the first contact rather than a degraded answer.
*/
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
static void collision_center(const void *obj, ccd_vec3_t *dest)
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
{
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
const collision_ccdobj *shape = (const collision_ccdobj *)obj;
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
ccdVec3Copy(dest, &shape->pos);
}
/** @brief Put a proxy into the form the support functions read. */
static void collision_to_ccd(akgl_CollisionProxy *proxy, collision_ccdobj *dest)
{
dest->kind = proxy->shape.kind;
ccdVec3Set(&dest->pos,
(ccd_real_t)(proxy->x + proxy->shape.ox),
(ccd_real_t)(proxy->y + proxy->shape.oy),
(ccd_real_t)(proxy->z + proxy->shape.oz));
ccdVec3Set(&dest->half,
(ccd_real_t)proxy->shape.hx,
(ccd_real_t)proxy->shape.hy,
(ccd_real_t)proxy->shape.hz);
}
/** @brief Configure the solver. Identical for every query. */
static void collision_configure(ccd_t *ccd)
{
CCD_INIT(ccd);
ccd->support1 = collision_support;
ccd->support2 = collision_support;
ccd->center1 = collision_center;
ccd->center2 = collision_center;
ccd->max_iterations = AKGL_COLLISION_MAX_ITERATIONS;
}
/**
* @brief Box against box, in closed form.
*
* Overlap on each axis, take the smallest, and the normal is that axis signed
* away from @p b. Exact rather than converged, which matters for a resting
* actor: an iterative solver answers `(0.0001, -0.99999, 0)` where this answers
* `(0, -1, 0)`, and that difference accumulates into a slow sideways creep along
* a floor.
*/
static akerr_ErrorContext *collision_box_box(akgl_CollisionProxy *a, akgl_CollisionProxy *b, akgl_Contact *dest, bool *hit)
{
float32_t delta[3];
float32_t overlap[3];
float32_t acentre[3];
float32_t bcentre[3];
int axis = 0;
int i = 0;
PREPARE_ERROR(errctx);
acentre[0] = a->x + a->shape.ox;
acentre[1] = a->y + a->shape.oy;
acentre[2] = a->z + a->shape.oz;
bcentre[0] = b->x + b->shape.ox;
bcentre[1] = b->y + b->shape.oy;
bcentre[2] = b->z + b->shape.oz;
delta[0] = bcentre[0] - acentre[0];
delta[1] = bcentre[1] - acentre[1];
delta[2] = bcentre[2] - acentre[2];
overlap[0] = (a->shape.hx + b->shape.hx) - fabsf(delta[0]);
overlap[1] = (a->shape.hy + b->shape.hy) - fabsf(delta[1]);
overlap[2] = (a->shape.hz + b->shape.hz) - fabsf(delta[2]);
*hit = false;
for ( i = 0; i < 3; i++ ) {
if ( overlap[i] <= 0.0f ) {
SUCCEED_RETURN(errctx);
}
}
for ( i = 1; i < 3; i++ ) {
if ( overlap[i] < overlap[axis] ) {
axis = i;
}
}
dest->nx = 0.0f;
dest->ny = 0.0f;
dest->nz = 0.0f;
/*
* Away from b. If b's centre is to the right of a's, then a leaves to the
* left and the normal is negative on that axis. A delta of exactly 0 means
* the two are concentric here and either direction is as good; pick one
* rather than emit a zero-length normal, which would be a wall that moves
* nothing.
*/
if ( axis == 0 ) {
dest->nx = (delta[0] > 0.0f) ? -1.0f : 1.0f;
} else if ( axis == 1 ) {
dest->ny = (delta[1] > 0.0f) ? -1.0f : 1.0f;
} else {
dest->nz = (delta[2] > 0.0f) ? -1.0f : 1.0f;
}
dest->depth = overlap[axis];
// Midpoint of the two centres. For two boxes that sits inside the overlap.
dest->px = (acentre[0] + bcentre[0]) / 2.0f;
dest->py = (acentre[1] + bcentre[1]) / 2.0f;
dest->pz = (acentre[2] + bcentre[2]) / 2.0f;
*hit = true;
SUCCEED_RETURN(errctx);
}
/**
* @brief Flatten a normal into the xy plane, and rescue a degenerate one.
*
* See #AKGL_COLLISION_TEST_PLANAR. The fallback matters more than it looks: two
* shapes at exactly the same centre have no planar direction to separate along,
* and level authors put things on top of each other constantly. Answering with a
* zero-length normal there would move an actor by nothing and report success,
* which is a wall that does not stop anything.
*/
static void collision_flatten(akgl_CollisionProxy *a, akgl_CollisionProxy *b, akgl_Contact *dest)
{
float32_t len = 0.0f;
float32_t dx = 0.0f;
float32_t dy = 0.0f;
dest->nz = 0.0f;
len = sqrtf((dest->nx * dest->nx) + (dest->ny * dest->ny));
if ( len > AKGL_COLLISION_EPSILON ) {
dest->nx = dest->nx / len;
dest->ny = dest->ny / len;
return;
}
// Nothing planar survived. Separate along whichever axis they overlap least.
dx = (a->shape.hx + b->shape.hx) - fabsf((b->x + b->shape.ox) - (a->x + a->shape.ox));
dy = (a->shape.hy + b->shape.hy) - fabsf((b->y + b->shape.oy) - (a->y + a->shape.oy));
if ( dx <= dy ) {
dest->nx = (((b->x + b->shape.ox) - (a->x + a->shape.ox)) > 0.0f) ? -1.0f : 1.0f;
dest->ny = 0.0f;
dest->depth = dx;
} else {
dest->nx = 0.0f;
dest->ny = (((b->y + b->shape.oy) - (a->y + a->shape.oy)) > 0.0f) ? -1.0f : 1.0f;
dest->depth = dy;
}
}
akerr_ErrorContext *akgl_collision_test(akgl_CollisionProxy *a, akgl_CollisionProxy *b, uint32_t flags, akgl_Contact *dest, bool *hit)
{
collision_ccdobj obja;
collision_ccdobj objb;
ccd_t ccd;
ccd_real_t depth = 0.0;
ccd_vec3_t dir;
ccd_vec3_t pos;
int result = 0;
PREPARE_ERROR(errctx);
FAIL_ZERO_RETURN(errctx, a, AKERR_NULLPOINTER, "NULL first proxy reference");
FAIL_ZERO_RETURN(errctx, b, AKERR_NULLPOINTER, "NULL second proxy reference");
FAIL_ZERO_RETURN(errctx, dest, AKERR_NULLPOINTER, "NULL contact reference");
FAIL_ZERO_RETURN(errctx, hit, AKERR_NULLPOINTER, "NULL hit flag reference");
memset(dest, 0x00, sizeof(akgl_Contact));
dest->tilex = -1;
dest->tiley = -1;
dest->tilelayer = -1;
*hit = false;
if ( (a->shape.kind == AKGL_COLLISION_SHAPE_NONE) ||
(b->shape.kind == AKGL_COLLISION_SHAPE_NONE) ) {
SUCCEED_RETURN(errctx);
}
/*
* The bounds are already computed and already stored, so the cheapest
* rejection in the system costs four comparisons and no arithmetic. Most
* candidate pairs a broad phase hands over die right here.
*/
if ( !SDL_HasRectIntersectionFloat(&a->bounds, &b->bounds) ) {
SUCCEED_RETURN(errctx);
}
if ( (a->shape.kind == AKGL_COLLISION_SHAPE_BOX) &&
(b->shape.kind == AKGL_COLLISION_SHAPE_BOX) ) {
PASS(errctx, collision_box_box(a, b, dest, hit));
if ( (*hit == true) && ((flags & AKGL_COLLISION_TEST_PLANAR) != 0) ) {
collision_flatten(a, b, dest);
}
SUCCEED_RETURN(errctx);
}
collision_to_ccd(a, &obja);
collision_to_ccd(b, &objb);
collision_configure(&ccd);
// One query, one arena. Resetting on entry rather than on exit means a query
// that returns early still leaves it clean for the next one.
akgl_ccd_arena_reset();
/*
* MPR and not GJK+EPA. MPR allocates nothing at all, converges in fewer
* iterations, and its weakness -- a coarser contact *point* -- is on a field
* the blocking resolver never reads. EPA stays compiled and available for a
* caller who one day wants an accurate manifold and will pay the arena for
* it.
*/
result = ccdMPRPenetration(&obja, &objb, &ccd, &depth, &dir, &pos);
if ( result == -2 ) {
FAIL_RETURN(
errctx,
AKGL_ERR_COLLISION,
"Collision arena exhausted at %zu of %d bytes; raise AKGL_CCD_ARENA_BYTES",
akgl_ccd_arena_highwater(),
AKGL_CCD_ARENA_BYTES
);
}
if ( result != 0 ) {
SUCCEED_RETURN(errctx);
}
/*
* libccd hands back the direction that separates the *second* object. This
* contact points at the first, so that a caller moving `a` along the normal
* by the depth undoes the overlap -- which is what every resolver wants, and
* saves each of them from remembering the sign.
*/
dest->nx = -(float32_t)ccdVec3X(&dir);
dest->ny = -(float32_t)ccdVec3Y(&dir);
dest->nz = -(float32_t)ccdVec3Z(&dir);
dest->depth = (float32_t)depth;
dest->px = (float32_t)ccdVec3X(&pos);
dest->py = (float32_t)ccdVec3Y(&pos);
dest->pz = (float32_t)ccdVec3Z(&pos);
if ( (flags & AKGL_COLLISION_TEST_PLANAR) != 0 ) {
collision_flatten(a, b, dest);
}
*hit = true;
SUCCEED_RETURN(errctx);
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
}
akerr_ErrorContext *akgl_collision_arena_selftest(float32_t separation, bool *hit, float32_t *depth)
{
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
collision_ccdobj a;
collision_ccdobj b;
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
ccd_t ccd;
ccd_real_t ccddepth = 0.0;
ccd_vec3_t dir;
ccd_vec3_t pos;
int result = 0;
PREPARE_ERROR(errctx);
FAIL_ZERO_RETURN(errctx, hit, AKERR_NULLPOINTER, "NULL hit flag reference");
FAIL_ZERO_RETURN(errctx, depth, AKERR_NULLPOINTER, "NULL depth reference");
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
collision_configure(&ccd);
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
a.kind = AKGL_COLLISION_SHAPE_BOX;
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
ccdVec3Set(&a.pos, 0.0, 0.0, 0.0);
ccdVec3Set(&a.half, 1.0, 1.0, 1.0);
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
b.kind = AKGL_COLLISION_SHAPE_BOX;
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
ccdVec3Set(&b.pos, (ccd_real_t)separation, 0.0, 0.0);
ccdVec3Set(&b.half, 1.0, 1.0, 1.0);
akgl_ccd_arena_reset();
/*
Answer whether two shapes overlap, and how to undo it akgl_collision_test takes two proxies and fills in a contact: a unit normal, a depth, and a point. Three paths, cheapest first. The proxies' bounds reject most pairs in four comparisons and no arithmetic, which is free because the bounds were computed when the proxy was synced. A box against a box is answered in closed form. That is not only cheaper than the general solver, it is *exact*, and the difference shows up where it matters most: an actor resting on a floor wants a normal of precisely (0, -1, 0), and an iterative solver converges to something like (0.0001, -0.99999, 0), which accumulates into a slow sideways creep. A tile game is almost entirely boxes, so this is the path that runs. Everything else goes to ccdMPRPenetration. MPR rather than GJK+EPA because MPR allocates nothing at all, converges in fewer iterations, and its one weakness -- a coarser contact *point* -- is on a field the blocking resolver never reads. EPA stays compiled and available behind the arena for a caller who one day wants an accurate manifold. The header says the point is approximate so nobody builds a damage falloff on it. The normal points out of the second shape and toward the first, so a caller moving `a` along it by `depth` separates the pair. libccd hands back the opposite convention; converting once here saves every resolver from remembering the sign, and getting it backwards would compile, would pass any "do these collide" test, and would drag actors into walls. There is a test that asserts the direction, and inverting the conversion turns it red. AKGL_COLLISION_TEST_PLANAR flattens the normal into the xy plane. The extrusion the setters apply already makes z the most expensive axis, so this is a net for a caller who set a depth by hand -- and the failure it catches is silent: the narrowphase reports a contact, the resolver pushes the actor into the screen, and the actor does not move on screen while remaining inside the floor. It also rescues the degenerate case, two shapes at exactly the same centre, where there is no planar direction at all and a zero-length normal would be a wall that stops nothing. Level authors stack things constantly. Three things the tests found rather than assumed: - The guard has two copies, one per path, and the box one is where it earns its place. Removing both is caught; removing only the iterative one is not, because the iterative solver is seeded from the line between the two centres and so converges to a planar answer on its own for a planar offset. That is measured and written down in the test rather than papered over with a fixture contrived to force it. - A concentric pair has no preferred sign on the degenerate axis, so the test asserts the magnitude of the normal rather than its direction. The first version asserted -1 and failed against an equally correct +1. - The box fast path and the iterative solver are two implementations of one answer, so they are driven over the same arrangements and required to agree on the decision and the axis. A shortcut nothing cross-checks is a shortcut waiting to diverge. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:18:09 -04:00
* ccdGJKPenetration and not ccdMPRPenetration, deliberately. This is the EPA
* path, and EPA is the half of libccd that allocates -- so it is the half
* that exercises the arena. The narrowphase proper uses MPR, which does not.
Give libccd a static arena so libakgl still allocates nothing libakgl does not call malloc at runtime. That is stated in seven places in the manual and is AGENTS.md's second standing rule, and every object comes from a fixed array in akgl/heap.h. libccd's EPA path does not know that: it builds its expanding polytope out of realloc and free, and ccdGJKPenetration documents a -2 return for when that fails. The alternative was to compile only the three files MPR needs and leave EPA out of the build. That works and it puts a copy of somebody else's source list in this repository, where it rots silently on the next submodule bump. This instead compiles all of libccd and points its allocator at a bump allocator over static BSS, which keeps the promise literally true -- and keeps EPA available rather than amputated, for the day a precise contact manifold is worth having. The arena is reset at the top of each query rather than freed block by block, so the lifetime is one narrowphase call, `free` is a no-op, and allocation is a pointer bump. Exhaustion returns NULL, which libccd already handles by unwinding to -2, which becomes AKGL_ERR_COLLISION naming the high-water mark -- a loud failure with a number in it rather than a silently missed collision, which would be a floor an actor falls through reported as success. One GJK/EPA box pair costs 7,264 bytes, measured and printed by the suite. The 64 KB ceiling is that with about nine times headroom, and the high-water mark is reported so the next reader can re-derive it rather than trust this line. The redirect lives in src/ccd_arena_shim.h, injected with -include, and not in CMake's COMPILE_DEFINITIONS. It was in COMPILE_DEFINITIONS first, and that is a mistake worth recording: CMake cannot carry a function-like macro through a -D, so it dropped __CCD_ALLOC_MEMORY without a diagnostic. libccd went on calling the C library's realloc while the shim's `free` quietly discarded the results -- strictly worse than doing nothing, and invisible, because it leaks rather than crashes. What caught it was insisting the test prove the wiring rather than the outcome. The first version asserted the arena balanced back to zero after a query, which turned out to be the wrong assertion for a different reason -- free is a no-op by design, so it cannot balance -- but a test that had merely checked "two boxes collide" would have passed throughout, against an allocator nothing was using. The suite now asserts what is actually provable: that a query allocates from the arena at all, and that the process survives, since glibc aborts when the real free(3) is handed a pointer it never issued. Also here: AKGL_ERR_COLLISION, with the name registered -- tests/error.c asserts the band and the names agree, and it caught the missing one immediately. -fvisibility=hidden and CCD_STATIC_DEFINE keep every ccd* symbol out of libakgl.so's dynamic table, which `nm -D` confirms is empty; AGENTS.md records a shipped defect where an exported `renderer` was preempted by a same-named symbol elsewhere, and a game linking a system libccd would hit exactly that. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-01 23:29:05 -04:00
*/
result = ccdGJKPenetration(&a, &b, &ccd, &ccddepth, &dir, &pos);
if ( result == -2 ) {
FAIL_RETURN(
errctx,
AKGL_ERR_COLLISION,
"Collision arena exhausted at %zu of %d bytes; raise AKGL_CCD_ARENA_BYTES",
akgl_ccd_arena_highwater(),
AKGL_CCD_ARENA_BYTES
);
}
*hit = (result == 0);
*depth = (float32_t)ccddepth;
SUCCEED_RETURN(errctx);
}
Add a pluggable broad phase, and the incremental uniform grid behind it akgl_Partitioner is a record of function pointers and an initializer, the same shape as the render and physics backends. `move` is its own slot rather than remove-then-insert, and that is the whole design: a uniform grid's move is a comparison and a return when the proxy has not left the cells it was in, which is the steady state for a walking actor and the permanent state for a static one. Spelling it as remove-and-insert would turn the incremental grid into the rebuild-every-frame tree that PERFORMANCE.md already measured and rejected. The grid is a dense 128x128 array of cell heads over the world, cells keyed on tile size, with two intrusive chains per entry -- one through the cell so a query can walk it, one through the proxy so removal unlinks a whole span without searching. Everything is an int16_t index rather than a pointer, so the structure is relocatable and clearing it is a memset. A proxy covering more cells than AKGL_COLLISION_GRID_MAX_SPAN spills onto one chain every query walks, which is what makes the cell pool's ceiling provable rather than hopeful. tests/partition.c compares every query against a linear scan over the same proxies and asserts containment in one direction, not equality. That asymmetry is the contract: over-reporting costs a narrowphase call, under-reporting is a wall an actor walks through. The suite is a table over partitioners so the same assertions run against every implementation, which is what the second one landing later will be held to. Three deliberate breaks, and two of them were not caught the first time: - Removing the min-cell rule so a shared-cell pair reports repeatedly: caught. - Dropping the unlink in `move`: **not caught**, initially. Stale entries do not produce wrong query answers, because the bounds re-check filters them out -- they leak the cell pool until the grid stops working, on a timescale no unit test reaches by accident. Counting live pool entries across 200 moves is the only thing that sees it, and now does: 505 entries against an expected 5. - Swapping floorf for a truncating cast: **not caught, and correctly so.** Everything below zero is clamped to cell 0 either way, so today the two are indistinguishable. The comment claiming truncation is a defect was overstating it and now says what is actually true: the clamp is the only thing hiding it, and a negative world origin or a relaxed clamp would make it real with no test standing in front of it. Co-Authored-By: Claude Code <noreply@anthropic.com> Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-Authored-By: Andrew Kesterson <andrew@aklabs.net>
2026-08-02 06:29:02 -04:00
akerr_ErrorContext *akgl_partitioner_factory(akgl_Partitioner *self, char *type)
{
PREPARE_ERROR(errctx);
FAIL_ZERO_RETURN(errctx, self, AKERR_NULLPOINTER, "NULL partitioner reference");
// NULL means "the default", which is the grid. Matched on leading
// characters, the same way akgl_physics_factory matches its backends.
if ( (type == NULL) || (strncmp(type, "grid", 4) == 0) ) {
PASS(errctx, akgl_partitioner_init_grid(self));
SUCCEED_RETURN(errctx);
}
FAIL_RETURN(errctx, AKERR_KEY, "No partitioner named \"%s\"", type);
}
akerr_ErrorContext *akgl_collision_world_init(akgl_CollisionWorld *self, char *type, float32_t cellwidth, float32_t cellheight)
{
PREPARE_ERROR(errctx);
FAIL_ZERO_RETURN(errctx, self, AKERR_NULLPOINTER, "NULL collision world reference");
FAIL_NONZERO_RETURN(errctx, (cellwidth <= 0.0f), AKERR_VALUE,
"Cell width %f is not positive", cellwidth);
FAIL_NONZERO_RETURN(errctx, (cellheight <= 0.0f), AKERR_VALUE,
"Cell height %f is not positive", cellheight);
memset(self, 0x00, sizeof(akgl_CollisionWorld));
self->cellwidth = cellwidth;
self->cellheight = cellheight;
// Every game today is 2D, so the guard is on by default. A caller with a
// real third axis to resolve on clears it.
self->flags = AKGL_COLLISION_TEST_PLANAR;
PASS(errctx, akgl_partitioner_factory(&self->partitioner, type));
PASS(errctx, self->partitioner.reset(&self->partitioner, self));
SUCCEED_RETURN(errctx);
}
Reference in New Issue Copy Permalink