Memory corruption inside gradient_clusters?

[Windwoes]

I'm building AprilTag from source (commit bee5efb1cc263ec1459bca46193e3428220d52ba) using a custom CMake file and adding a couple extra files to allow access from the JNI.

The AprilTag source code itself is unchanged. I'm having an issue where often AprilTag will segfault, on both Windows and Linux. The JVM error log shows the fault happened in the gradient_clusters frame.

C  [libc.so.6+0x165ec5]
C  [libapriltag.so+0x39d5f]  gradient_clusters+0x3e1
C  [libapriltag.so+0x3a54e]  apriltag_quad_thresh+0x3e6
C  [libapriltag.so+0x2d6f9]  apriltag_detector_detect+0x3e7

I also sometimes get the message double free or corruption (fasttop)

Interestingly, it works fine on Android. So it seems like there may a memory bug inside gradient_clusters. (Either that or I have a very strange memory issue in my JNI code, but I'm simply pointing AprilTag at a greyscale image buffer, and when it doesn't crash, it runs fine)

[christian-rauch]

Its hard to tell without a backtrace or a reproducible case with the default examples.

There have been some issues reported with ASAN (https://github.com/AprilRobotics/apriltag/pull/213). Maybe fixing those will also solve you issue? Otherwise, can you run your example with ASAN (AddressSanitizer) or Valgrind?

[Windwoes]

I tried to run with ASAN, but it seems that ASAN isn't compatible with the JNI. I will try to find some way to figure out more specifically where the issue is.

[Windwoes]

Update: the crashing on Linux was due to accidentally mixing two versions of the Java side and native side JNI code. I am still having intermittent crashes on Windows which I have not yet been able to track down.

[christian-rauch]

@Windwoes Is your issue resolved? Can we close this?

[Windwoes]

Unfortunately I haven't had any more time to look into the windows issue yet, but I hopefully will soon.