Inverse Purification crashes on non-invertible matrix

[Window-Hero]

Attempting to use Inverse Purification on a non-invertible matrix triggers a crash. This should likely trigger a divide by zero mishap instead.

[object-Object]

Please provide the crash report and latest.log file via https://mclo.gs.

[Window-Hero]

https://mclo.gs/MtaTipb
https://mclo.gs/MjajGFh

[pythonmcpi]

I'm facing a crash with the exact same crashlog (matching error message, stack trace, and line numbers). This replicates on a server running on Pebblehost (Linux amd64), but not in singleplayer (Linux amd64).

However, this crash is triggering on an invertible 3x3 matrix (determinant is 1.0), not only on non-invertible matrices.

Environment: Fabric 1.20.1, Hex Casting 0.11.3, MoreIotas 0.1.1, Hexical 1.5.0, Hexcassettes 1.1.4.

A server log entry immediately before the stack trace reads:

[Server thread/INFO]: [STDERR]: -- org.jblas ERROR Couldn't load copied link file: java.lang.UnsatisfiedLinkError: /tmp/jblas16049075580560245065/libjblas.so: libgfortran.so.5: cannot open shared object file: No such file or directory.

[object-Object]

Apparently this is an upstream issue in jblas: jblas-project/jblas#136
We'll probably have to build jblas ourselves to fix this, since it doesn't seem like it's being maintained anymore.

[object-Object]

Or alternatively we might just need to switch to a different matrix math library. Seems like jblas also might not work on aarch64: jblas-project/jblas#135

https://github.com/lessthanoptimal/ejml seems like a good candidate for a replacement.

[c-Caelum]

Swapping libraries is a breaking change. Is there a nice way to port it over?