segfault on linux x86-64 for 1.1-snapshot

[dlwh]

qr decomposition produces a guaranteed crasher on linux for me:

https://gist.github.com/dlwh/6626939

It's in dgeqrf

-- David

[dlwh]

(Bug not present in 1.0)

[fommil]

@dlwh On my machine (OS X) I'm getting a crash with this test, (same QR code):

sbt "test-only breeze.linalg.LinearAlgebraTest"

I'm also getting a failure for DenseMatrixTest, but that looks like a machine precision problem:

[info] - Solve *** FAILED ***
[info]   DenseVector(0.18131868131868126, -0.313186813186813, 0.4395604395604394) did not equal DenseVector(0.1813186813186811, -0.3131868131868131, 0.43956043956043944) (DenseMatrixTest.scala:375)

[fommil]

btw, I can confirm that there is no such crash or test failure with F2J backend

SBT_OPTS="-Dcom.github.fommil.netlib.BLAS=com.github.fommil.netlib.F2jBLAS -Dcom.github.fommil.netlib.LAPACK=com.github.fommil.netlib.F2jLAPACK -Dcom.github.fommil.netlib.ARPACK=com.github.fommil.netlib.F2jARPACK" sbt "test-only breeze.linalg.LinearAlgebraTest"

(you might need to edit your SBT launcher script to get this to work)

and it also appears with the NativeRef builds

SBT_OPTS="-Dcom.github.fommil.netlib.BLAS=com.github.fommil.netlib.NativeRefBLAS -Dcom.github.fommil.netlib.LAPACK=com.github.fommil.netlib.NativeRefLAPACK -Dcom.github.fommil.netlib.ARPACK=com.github.fommil.netlib.NativeRefARPACK" sbt "test-only breeze.linalg.LinearAlgebraTest"

However, the problem also arises when I go back to 1.0, so this is not a regression. Oops! Is this a new test?

[fommil]

@dlwh I found it. Technically speaking it's a bug in your code :-P but I think I should support the corner case so I'll fix it on my end.

You're passing null arrays and that means when I try to get a reference to the array in the native code it fails.

[fommil]

@dlwh I've disted new SNAPSHOTs, but I think you need to check your test logic :-P

INFO: successfully loaded /var/folders/lr/331rxc_d187d9pmrwjj99nvr0000gn/T/jniloader2052816085861155377netlib-native_system-osx-x86_64.jnilib
[info] LinearAlgebraTest:
[info] - qr *** FAILED ***
[info]   0.06052275326688006 was not -0.06052275326 plus or minus 1.0E-8 (LinearAlgebraTest.scala:145)

[dlwh]

Hrm, does QR have a nonidentifiability in sign?

Thanks!