Crash at air vent in old comm engineering - dhewm3 1.5.4 on Raspberry Pi 5

[TheMidnightHammer]

Describe the bug
When I try to go back the air vent, the game crashes.

Your System
Ubuntu 24.04 LTS, raspberry pi 5 (arm64 quad core A76, videocore 7, driver V3D 7.1.7.0) and yes i compiled it myself.

To Reproduce
It was on the main campaign with no mods. Normally the floor “exposed” but this time the game crashes. I was just walking normally, i was playing on 1.5.4.

1. Load the attached save game.

2. Walk toward the "exposed" floor section near the air vent.

3. Observe the game freeze and exit to terminal.
   Note : This happens 100% of the time on this hardware configuration.

gameSave.zip

Screencast.from.2026-03-17.09-24-22.webm

The error snippet:

Looks like dhewm3 1.5.4 crashed with signal SIGABRT (6) - sorry!
  281473366566879 (unknown symbol)
  281473356690952 ./nptl/pthread_kill.c:44 __pthread_kill_implementation
  281473356385083 ../sysdeps/posix/raise.c:26 __GI_raise
  281473356299775 ./stdlib/abort.c:79 __GI_abort
  281473356356799 ./assert/assert.c:96 __assert_fail_base
  281473356356911 ./assert/assert.c:105 __assert_fail
  187650997965839 neo/cm/CollisionModel_rotate.cpp:1592 idCollisionModelManagerLocal::Rotation180(trace_s*, idVec3 const&, idVec3 const&, float, float, idVec3 const&, idTraceModel const*, idMat3 const&, int, int, idVec3 const&, idMat3 const&)
  187650997966539 neo/cm/CollisionModel_rotate.cpp:1655 idCollisionModelManagerLocal::Rotation(trace_s*, idVec3 const&, idRotation const&, idTraceModel const*, idMat3 const&, int, int, idVec3 const&, idMat3 const&)
  281472707185179 neo/game/physics/Clip.cpp:1307 idClip::Motion(trace_s&, idVec3 const&, idVec3 const&, idRotation const&, idClipModel const*, idMat3 const&, int, idEntity const*)
  281472707418075 neo/game/physics/Physics_RigidBody.cpp:194 idPhysics_RigidBody::CheckForCollisions(float, rigidBodyPState_s&, trace_s&)
  281472707423363 neo/game/physics/Physics_RigidBody.cpp:904 idPhysics_RigidBody::Evaluate(int, int)
  281472705988487 neo/game/Entity.cpp:2579 idEntity::RunPhysics()
  281472705989235 neo/game/Entity.cpp:838 idEntity::Think()
  281472706107127 neo/game/Game_local.cpp:2344 idGameLocal::RunFrame(usercmd_t const*)
  187650997730435 neo/framework/Session.cpp:2900 idSessionLocal::RunGameTic()
  187650997732867 neo/framework/Session.cpp:2846 idSessionLocal::Frame()
  187650997350027 neo/framework/Common.cpp:2455 idCommonLocal::Frame()
  187650996693127 neo/sys/linux/main.cpp:452 main
  281473356301507 ../sysdeps/nptl/libc_start_call_main.h:58 __libc_start_call_main
  281473356301719 ../csu/libc-start.c:360 __libc_start_main_impl
  187650996694767 _start
  18446744073709551615 (unknown symbol)
Aborted (core dumped)

Expected and actual behavior
I expected to be able to go to the air vent.

log.txt

[DanielGibson]

dhewm3: /home/midnighthammer/Downloads/dhewm3-1.5.4/neo/cm/CollisionModel_rotate.cpp:1592: void idCollisionModelManagerLocal::Rotation180(trace_t*, const idVec3&, const idVec3&, float, float, const idVec3&, const idTraceModel*, const idMat3&, int, cmHandle_t, const idVec3&, const idMat3&): Assertion 'results->fraction <= 1.0f' failed.

Can you run it in a dhewm3 in a debugger and check the actual value of results->fraction when the crash (actually: assertion) happens?
I wonder if it's just some kind of rounding error and it's just above 1.0, or if it's a completely different value that's truly wrong

[TheMidnightHammer]

I fixed it somehow by building in release mod. I added it to the auto builder pr.

I think compiler optimizations (which are present in the CMake Release mod) are doing something.

I did not find the specific value, but i almost certain that it is over 1, it's my third time using GDB so it will take me some time.

[DanielGibson]

That didn't fix it, it just disabled assertions

If gdb won't show you the value (can happen in RelWithDebInfo builds if it has been optimized away), you could also add these lines:

results->fraction = idMath::Fabs( atan( tw.maxTan ) * ( 2.0f * 180.0f / idMath::PI ) / tw.angle );
	}
// before this are lines that were already there, around line 1591, add the next three lines
	if(results->fraction > 1.0f) {
		printf("XXX: result fraction = %f\n", results->fraction);
	}
// after this are lines that were already there, around line 1592
	assert( results->fraction <= 1.0f );
	endRotation.Set( rorg, axis, startAngle + (endAngle-startAngle) * results->fraction );

and then check the terminal output for that "XXX: result ..." line (should be right before the Assertion)

[TheMidnightHammer]

I am still learning where everything is in dhewm3, in what file would this be?

[DanielGibson]

See the backtrace where the assertion happened:
neo/cm/CollisionModel_rotate.cpp:1592 idCollisionModelManagerLocal::Rotation180(trace_s*, idVec3 const&, idVec3 const&, float, float, idVec3 const&, idTraceModel const*, idMat3 const&, int, int, idVec3 const&, idMat3 const&)

[TheMidnightHammer]

Sorry for late respond, the result fraction is XXX: result fraction = 66.647858

Can we just check if its over 1 and if it is just set it to 1? That would change the rotation, but the assertion will not be triggered. Or we investigate the Idmath fabs function.

[TheMidnightHammer]

Since the result must be under 1 we can add a if statement that checks if it over and puts it to 0.9f

[DanielGibson]

no, if it's actually 66.65 that means it's so far off that something is seriously going wrong (not just rounding error) and needs a proper investigation and fix

[TheMidnightHammer]

In 1.5.2 i was not getting this error.

[DanielGibson]

even if you build it yourself, as a Debug or RelWithDebInfo build (i.e. assertions are enabled)?

[TheMidnightHammer]

In 1.5.2 i was not getting this error. I build it using the following code:
cmake ../neo/
make -j4

so i think it is Debug

[DanielGibson]

can you test if the latest git code also has the error?

if it does, can you use git bisect to test the revision between 1.5.2 and 1.5.4 to see what commit introduced the problem?

[TheMidnightHammer]

I tested the latest version 2 times.

The first one the game did not crash, the second the can crash.

[DanielGibson]

So the crash doesn't happen every time you test but only every second time or so?
If I understood that correctly, can you test 1.5.2 a couple of times more?

---

If 1.5.2 really does not show the problem, this is how git bisect works:
In a checked out copy of this git repo run:
git bisect start 1.5.4 1.5.2
Then

1. build dhewm3
2. test it for your problem (if the problem doesn't happen on every run, this means test it several times)
3. • If the problem occurred enter git bisect bad
   • If not, enter git bisect good
4. Either command will select and check out the next revision to test, so start at 1. again, until git tells you it has found the first bad commit.

This should only take around 7 steps

To clean up once everything is done (and you reported the first bad commit ID here), run git bisect reset